Monorepo for Aesthetic.Computer
aesthetic.computer
pop/voice: jeffrey speech synthesis, two lanes (PT + diphone bank)
PT lane (research / vocal-tract music instrument):
- vendor pink trombone from dood.al directly (MIT, sha-pinned), not
a github mirror (zakaton fork is GPL-3 after refactor)
- render-pt.mjs: node vm sandbox loads PT inline, renders sustained
vowels + keyframe trajectories at ~90 ms/render
- fit.py: CMA-ES + log-mel mean+std + librosa pyin F0; PT->PT
self-test recovers params, 2-keyframe joint fit recovers natural
rising F0 contour from jeffrey-pvc word recordings
- measure-jeffrey.py: mediapipe FaceLandmarker on 521 confirmed-jeffrey
IG photos -> 363 measurements -> jeffrey-anthropometry.json with
Fitch-Giedd VTL prior + lip aperture bound
- PHYSIOLOGY.md, MODEL-EXTENSIONS.md frame the personalized-PT paper
Pivot 2026-05-04: PT cannot reach phoneme-level intelligibility from
external trajectory fitting alone. PT reframed as a timbral musical
instrument; new mainline for shippable jeffrey speech is the diphone
bank.
Diphone lane (shippable AC-native TTS):
- diphone-targets.json: 1527 ARPABet pairs with perceptual weights,
532 in tier-1 (vowel-vowel + CV/VC)
- 10 paragraph-form carriers in jeffrey's lowercase voice (~$0.39
ElevenLabs)
- align-paragraphs.py: torchaudio MMS_FA + g2p_en -> 1510 phonemes
aligned across 168.7 s of audio
- extract-diphones.py: 485 unique diphones, 3.2 MB bank, Hann-tapered
- synth-word.py: text -> g2p_en -> overlap-add concat -> jeffrey-shaped
speech, --rate 0.82 default for natural pacing
- end-to-end works on "the music is real" and 10 demo words
Total session ElevenLabs spend: ~$0.65. Output: a 3.2 MB shippable
bank that pronounces arbitrary text in jeffrey's voice (with known
gaps at word boundaries, fixable next pass).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
+12536
-1
+7
-1
pop/SCORE.md
+7
-1
pop/SCORE.md
···
36
36
37
37
See [`big-pictures/README.md`](big-pictures/README.md) for the format spec.
38
38
39
-
### 2. (open)
39
+
### 2. voice (`voice/`)
40
+
41
+
The jeffrey harness for **Pink Trombone** — a tiny, runnable, anatomically grounded jeffrey voice fitted to jeffrey-pvc on a minimal phoneme corpus, with PT parameter bounds derived from the [jeffrey-platter](../papers/jeffrey-platter/) photographs. Research lane, not a track lane: the deliverable is a C/WASM-runnable synth + a paper, not an mp3.
42
+
43
+
See [`voice/README.md`](voice/README.md) for the pipeline. Status: scaffolded 2026-05-04.
44
+
45
+
### 3. (open)
40
46
41
47
More lanes will land here as they prove themselves. Candidates: kidlisp-as-instrument tracks, AC-native ensemble cuts, voice-memo-grade demo lane. None of them have earned a swimlane yet — they need a real track first.
42
48
+25
pop/voice/.gitignore
+25
pop/voice/.gitignore
···
1
+
# vendored third-party (fetch via bin/vendor-pt.sh)
2
+
vendor/pinktrombone/
3
+
4
+
# PT lane — corpus WAVs (jeffrey-pvc derivatives, regenerable)
5
+
corpus/raw/
6
+
corpus/cropped/
7
+
corpus/manifest.json
8
+
9
+
# PT lane — fit outputs (regenerable from corpus + anthropometry)
10
+
fits/
11
+
12
+
# Diphones lane — carrier mp3s + alignment + bank (jeffrey-pvc
13
+
# derivatives + regenerable from raw via bin/{align,extract}-*.py)
14
+
diphones/raw/
15
+
diphones/aligned/
16
+
diphones/bank/
17
+
diphones/manifest.json
18
+
19
+
# Cached model files (mediapipe FaceLandmarker, etc.)
20
+
.cache/
21
+
22
+
# python bytecode + venv crumbs
23
+
__pycache__/
24
+
*.pyc
25
+
.venv/
+87
pop/voice/CORPUS.md
+87
pop/voice/CORPUS.md
···
1
+
# corpus — phoneme target list + recording protocol
2
+
3
+
the audio side of the jeffrey harness. ~40 short jeffrey-pvc clips that the fitter optimizes pink trombone parameters against. each clip is the same phonetic target across multiple carrier contexts so co-articulation effects are observable, not a confound.
4
+
5
+
## scope
6
+
7
+
we are **not** trying to clone arbitrary jeffrey speech. the harness's job is to land coherent vowels and continuants — the things PT models well — and to handle plosives "good enough" for AC's pixelated-voice aesthetic (per `reference_pink_trombone.md`). the corpus reflects that weighting.
8
+
9
+
| category | count | notes |
10
+
| ----------- | ----: | ---------------------------------------------------- |
11
+
| short vowels | 5 | /æ ɛ ɪ ɒ ʊ/ — bat, bet, bit, bot, book |
12
+
| long vowels | 5 | /iː uː ɑː ɔː ɜː/ — see, soup, spa, saw, surf |
13
+
| diphthongs | 5 | /aɪ aʊ eɪ oʊ ɔɪ/ — eye, owl, ate, owe, oil |
14
+
| voiced fric. | 4 | /v ð z ʒ/ — van, this, zoo, vision |
15
+
| unvoiced fric. | 4 | /f θ s ʃ/ — fan, thin, see, ship |
16
+
| voiced plos. | 3 | /b d g/ — buy, do, go |
17
+
| unvoiced plos. | 3 | /p t k/ — pie, two, key |
18
+
| nasals | 3 | /m n ŋ/ — me, no, sing |
19
+
| liquids | 2 | /l ɹ/ — lo, row |
20
+
| glides | 2 | /w j/ — we, you |
21
+
| schwa | 1 | /ə/ — neutral tract reference |
22
+
| ~total | ~37 | |
23
+
24
+
(numbers approximate; final list lives in `corpus/phonemes.json` for code.)
25
+
26
+
## carrier contexts
27
+
28
+
each phoneme is recorded in **three** contexts so the fitter sees it under different articulatory loads:
29
+
30
+
1. **isolation** — sustained for ~1.5s. e.g. for /iː/: just "eeeee". target a steady-state tract pose.
31
+
2. **digit carrier** — embedded inside a counted "one… two… three…" or "a… b… c…" sequence. inherits jeffrey-pvc's prosodic envelope.
32
+
3. **CVC word** — a real english word containing the phoneme in stressed position. provides a co-articulation reference.
33
+
34
+
three contexts × ~37 phonemes = ~111 prompts, ~1.5s each → ~170s of audio.
35
+
36
+
we keep the carrier sentence list **short** (`a b c`, `1 2 3`, "buy", "see", etc.) on purpose. ElevenLabs charges per character, and the per-prompt costs add up if we get expansive. memory: `feedback_jeffrey_pvc_settings.md` — stability ≥ 0.5 to keep voice identity.
37
+
38
+
## recording settings
39
+
40
+
- voice: `provider=jeffrey, voice=neutral:0` (the existing jeffrey-pvc PVC)
41
+
- stability: `0.5` (memory: identity drift below 0.5)
42
+
- similarity: `0.9` (memory)
43
+
- style: `0.0` (we want neutral articulation, not performance)
44
+
- speed: `0.85` for sustained vowels (gives the fitter ~20% more steady-state to chew on); `1.0` for digit/CVC carriers
45
+
- format: server returns mp3; the recorder transcodes to **mono 22050 Hz 16-bit PCM WAV** for librosa/MFCC compatibility (PT's internal sample rate is 48k but the fitter doesn't care, only the spectral envelope matters)
46
+
- with-timestamps: **on** — char-level alignment means the fitter can crop directly to the target phoneme's onset/offset without re-running whisper
47
+
48
+
cache key: sha256 of `{ text, voice, stability, similarity, style, speed }` — same convention as `pop/bin/say.mjs`. reruns of `record-corpus.mjs` are free unless the prompt list changes.
49
+
50
+
## storage layout
51
+
52
+
```
53
+
corpus/
54
+
phonemes.json — master list (committed)
55
+
raw/ — gitignored
56
+
iso/<phoneme-id>.{wav,json} — iso recordings + alignment sidecar
57
+
digit/<phoneme-id>.{wav,json} — digit-carrier recordings
58
+
cvc/<phoneme-id>.{wav,json} — CVC-word recordings
59
+
cropped/ — gitignored
60
+
<phoneme-id>__<context>.wav — cropped to target phoneme span
61
+
using alignment timestamps
62
+
manifest.json — committed; phoneme → file map +
63
+
recording metadata + sha hashes
64
+
```
65
+
66
+
`raw/` and `cropped/` are gitignored because (a) they're regenerable, and
67
+
(b) the WAVs are jeffrey-pvc derivatives — the same closed-clone reasoning
68
+
that keeps `references/` out of the repo, applied conservatively.
69
+
70
+
## what we do **not** record
71
+
72
+
- spontaneous jeffrey speech. the goal is to fit a tract, not to capture
73
+
jeffrey's natural prosody. natural prosody comes back at synth time, not
74
+
at fit time.
75
+
- entire words at fitting resolution. PT can't synthesize an arbitrary
76
+
english word coherently from a static parameter set; that's `render.mjs`'s
77
+
job, not the corpus's.
78
+
- music or singing. if singing voice ever becomes a target, that's a
79
+
separate corpus with `--style 0.6+` and per-pitch sustains.
80
+
81
+
## next
82
+
83
+
- [ ] write `corpus/phonemes.json`
84
+
- [ ] implement `bin/record-corpus.mjs` (mirrors `pop/bin/say.mjs` cache pattern)
85
+
- [ ] verify: one isolation /iː/ recording, listened to, sounds like jeffrey
86
+
- [ ] generate full corpus
87
+
- [ ] crop to per-phoneme spans using alignment sidecars
+104
pop/voice/MODEL-EXTENSIONS.md
+104
pop/voice/MODEL-EXTENSIONS.md
···
1
+
# model extensions — how the platter can improve pink trombone itself
2
+
3
+
PHYSIOLOGY.md uses the platter to *constrain* PT's parameter search. this file goes one step further: the same anthropometric data could *extend* PT's model, not just bound it. that's a research direction, not a deliverable for the harness — but it's worth marking now so we don't accidentally rederive it later, and it's the natural shape of a paper out of this lane.
4
+
5
+
> "the laptop orchestra was a beautiful idea that reached almost no one" — `papers/arxiv-plork/plork.tex` §1
6
+
>
7
+
> pink trombone is the same shape: a beautiful physical model that is one anatomical specimen wide. every voice that comes out of it is the same person.
8
+
9
+
## what PT currently is
10
+
11
+
neil thapen's PT (2017) is a **single-speaker** physical model:
12
+
13
+
- 44 tract segments, each a constant 0.4 cm → tract length pinned at ~17.5 cm
14
+
- one hard-coded neutral-tract area function (close to Maeda 1990 generic adult-male)
15
+
- tongue index/diameter as the user's articulation interface (~Maeda's tongue-body PCA collapsed into 2D)
16
+
- glottal source from Liljencrants-Fant (LF) with a tension knob
17
+
- nasal port branch with constant scaling
18
+
- lip diameter as a single end-of-tube area parameter
19
+
20
+
it sounds remarkable for ~500 lines of JS. but it is *one* mouth.
21
+
22
+
## extensions the platter enables
23
+
24
+
### 1. variable tract length, properly
25
+
26
+
PT exposes `tract.n` (segment count) but the JS UI never varies it. with the Lammert+Narayanan VTL estimate from the platter we can:
27
+
28
+
- set `tract.n` to a per-speaker integer that targets the measured VTL at the existing 0.4 cm/segment quantum, *or*
29
+
- vary the per-segment length (`tract.dx`) and keep `n` fixed.
30
+
31
+
both should be acoustically equivalent at low frequencies; the difference matters at the higher formants. the question worth answering — and it isn't currently in the PT literature — is which formulation is more robust for an *adapted* tract whose length differs by 5–10% from the default.
32
+
33
+
### 2. personalized neutral-tract area function
34
+
35
+
PT ships with a single area function (close to schwa /ə/). every articulation is a deviation from that one neutral pose. the personalization opportunity:
36
+
37
+
> the platter gives us closed-mouth neutral-pose photos. for each, we read the lip aperture, the mandible angle, and (with DECA / EMOCA) the 3D head shape. this anchors **one end** of the area function — the lip end — to jeffrey-specific values. the *interior* of the tract still uses the generic curve, but the end conditions are jeffrey's.
38
+
39
+
acoustically: this should shift F1/F2 of jeffrey's neutral pose by 50–150 Hz versus PT's default. measurable from the schwa recording in CORPUS.md.
40
+
41
+
### 3. mandible kinematics from real jaw rotation
42
+
43
+
PT's "lip diameter" knob secretly controls a small constellation of segments (the front of the tract). with the platter's open-mouth candids vs. closed-mouth neutrals, we can fit jeffrey's actual mandible rotation range and translate that into a *coupled* update of multiple PT segments — front-tract opening, lip protrusion, lower-incisor position — instead of one slider.
44
+
45
+
this is closer to Maeda's original tongue-jaw PCA, but personalized. measurable: open-mouth /ɑː/ vs closed-mouth /uː/ formant shifts should track better against the real jeffrey-pvc recordings than against PT's defaults.
46
+
47
+
### 4. nasal cavity scale from external nostril measurement
48
+
49
+
PT's nasal branch has a constant-scaled area function. nostril-width and inter-alar distance correlate with nasal-cavity volume (Sahin-Yilmaz et al. 2008, Otolaryngology). a one-parameter scale on PT's nasal branch, derived from the platter, should improve nasal-vowel quality (/m n ŋ/ recordings in CORPUS.md become the test set).
50
+
51
+
### 5. F0 envelope from the corpus, not from a knob
52
+
53
+
PT's glottis is parameterized by tension and noise. jeffrey's actual F0 distribution comes out of the alignment-timestamped corpus *for free* — every recorded clip already carries jeffrey-pvc's pitch contour. we can:
54
+
55
+
- estimate jeffrey's neutral F0 (likely 100–130 Hz adult male)
56
+
- estimate his vibrato envelope (probably tiny in the speech corpus, larger if a singing corpus ever lands)
57
+
- use those as the priors on PT's F0 modulation, not a hand-tuned LF
58
+
59
+
### 6. cross-speaker generalization
60
+
61
+
if this works for jeffrey, the same pipeline can be run on any speaker who has both a voice clone and a small photo set. that's a contribution back to PT's user community — a "speaker adaptation" pipeline.
62
+
63
+
## what this would *not* be
64
+
65
+
- **not a paper claiming PT is broken.** it isn't. PT is a celebrated educational artifact and a remarkably accurate generic adult-male tract. the extensions are personalization, not correction.
66
+
- **not a face-to-voice deepfake.** the goal is voice synthesis whose anatomical parameters are *consistent* with the visible face, not whose acoustic fingerprint matches a target. the acoustic fingerprint comes from the jeffrey-pvc corpus, which jeffrey controls.
67
+
- **not a clinical tool.** medical use cases (post-surgery prosody planning, articulator training for speech therapy) need MRI ground truth. this lane is artistic and research-grade.
68
+
69
+
## paper outline (if/when it earns one)
70
+
71
+
```
72
+
title: Personalized Pink Trombone — Anatomically-Grounded Vocal Tract
73
+
Synthesis from Photographs and Speech
74
+
75
+
§1 Introduction — PT as universal-adult model, the personalization gap
76
+
§2 Related work — Maeda 1990, Story 1996 area functions, Lammert 2015
77
+
VTL regression, Fitch-Giedd 1999, voice cloning state
78
+
§3 Anthropometric pipeline — mediapipe → measurements → VTL
79
+
§4 Acoustic fitting — CMA-ES on PT params, MFCC distance, anatomical bounds
80
+
§5 Model extensions — variable-VTL, personalized end-conditions, mandible
81
+
§6 Evaluation — held-out phonemes, A/B vs default-PT, formant accuracy
82
+
§7 AC-native deployment — C/WASM port, runtime budget, integration with
83
+
the recap and big-pictures lanes
84
+
§8 Limitations — external-only measurements, no MRI ground truth, single
85
+
speaker, plosive quality
86
+
§9 Future work — multi-speaker corpus, singing extension, real-time
87
+
inversion (image-to-voice with no acoustic target)
88
+
89
+
cite: Thapen 2017 (PT itself), Maeda 1990, Story et al 1996, Fitch+Giedd
90
+
1999, Lammert+Narayanan 2015, Sahin-Yilmaz et al 2008
91
+
(nasal anatomy), DECA/EMOCA for 3D face fit if used.
92
+
```
93
+
94
+
this fits the **arxiv-ac** lane in `papers/SCORE.md` as a sibling to the
95
+
existing AC-native papers. it doesn't fit the big-pictures rap lane —
96
+
the rap version is just the chorus: *"PT was one mouth; the platter made
97
+
it ours."*
98
+
99
+
## status
100
+
101
+
- 2026-05-04 — outline only. no extensions implemented.
102
+
- gating: ship the harness first (PHYSIOLOGY + fit), then evaluate which
103
+
extensions produce audible/measurable gains, then write the paper if
104
+
any of them do.
+26
pop/voice/NOTICE.md
+26
pop/voice/NOTICE.md
···
1
+
# NOTICE — third-party material
2
+
3
+
## Pink Trombone
4
+
5
+
`vendor/pinktrombone/` (gitignored — fetch via `bin/vendor-pt.sh`) contains the canonical Neil Thapen Pink Trombone source — a single self-contained HTML file with all JS inline.
6
+
7
+
- **upstream:** https://dood.al/pinktrombone/ (the original, not a github mirror)
8
+
- **license: MIT** — *"Copyright 2017 Neil Thapen / Permission is hereby granted, free of charge…"* — verbatim copy in `vendor/pinktrombone/LICENSE`. Read directly from the upstream HTML's comment header (lines 28–46 of v1.1, March 2017).
9
+
- **why dood.al, not a github mirror:** dood.al IS the canonical source. Every github mirror is downstream. Some popular mirrors (e.g. `zakaton/Pink-Trombone`) re-license as **GPL-3.0** after refactoring, which is incompatible with shipping a permissive C/WASM jeffrey voice in fedac/native. Single-file dood.al fetch + sha256 pin is the cleanest provenance contract.
10
+
- **modifications policy:** keep `vendor/pinktrombone/` *unmodified* relative to upstream. all jeffrey-specific extensions live under `pop/voice/bin/` (e.g. `render-pt.mjs`), which read from the vendor tree.
11
+
- **provenance:** `vendor/pinktrombone/UPSTREAM.txt` captures URL, sha256, fetch timestamp, version string, author. Recorded automatically by `bin/vendor-pt.sh`.
12
+
- **MIT compliance:** any redistribution of derived AC artifacts that include the PT runtime must carry the MIT notice. The notice in `LICENSE` is the load-bearing copy; downstream packagers reproduce it verbatim.
13
+
14
+
## ElevenLabs jeffrey-pvc clone
15
+
16
+
The recorded WAVs under `corpus/raw/` and `corpus/cropped/` are derivative outputs of jeffrey's own ElevenLabs Professional Voice Clone (jeffrey-pvc). Jeffrey owns the source recording rights for that clone. These derivative WAVs are:
17
+
18
+
- gitignored (see `.gitignore`)
19
+
- used internally for fitting; not redistributed
20
+
- regenerable from `bin/record-corpus.mjs` (any consumer who needs them can re-pay the ElevenLabs API)
21
+
22
+
If the harness is ever published as a research artifact, the corpus WAVs are *not* published; only the fitted PT parameters are.
23
+
24
+
## jeffrey-platter photographs
25
+
26
+
`measure-jeffrey.py` reads photographs from `papers/jeffrey-platter/` and the AC assets CDN. Those images are subject to the platter's own provenance and licensing (see `papers/jeffrey-platter/README.md`). The harness only stores **derived measurements** (`jeffrey-anthropometry.json`) — never copies of the source photographs.
+148
pop/voice/PHYSIOLOGY.md
+148
pop/voice/PHYSIOLOGY.md
···
1
+
# physiology — anthropometric priors from the platter
2
+
3
+
the visual side of the harness. PT's tract is a 44-segment tube with a default geometry intended for "an adult". if we leave it there, the CMA-ES fitter wastes evals exploring acoustic basins that correspond to anatomically impossible jeffreys. the platter already contains 55 AV-shoot headshots and 38 candid masters — high-res, multi-angle, well-annotated. we measure jeffrey from those, derive tract priors, and clamp the fitter's search space.
4
+
5
+
> "the address is the score" — `papers/arxiv-ac/ac.txt`
6
+
7
+
every photograph already encodes a physiological score. we just need to read it.
8
+
9
+
## what we measure (and why)
10
+
11
+
| measurement | source | reason it matters for PT |
12
+
| --------------------------------- | ----------------------------------------------------------- | --------------------------------------------------------- |
13
+
| **head height** (vertex → menton) | mediapipe face mesh landmarks 10 (forehead) → 152 (chin) | strong correlate of vocal tract length (Fitch+Giedd 1999) |
14
+
| **bizygomatic width** | landmarks 234 ↔ 454 | secondary tract-length predictor; fixes scale |
15
+
| **mandible length** (gonion → menton) | landmarks 172/397 → 152 | mandible ROM constrains jaw rotation in PT |
16
+
| **lip width** (cheilion ↔ cheilion) | landmarks 61 ↔ 291 | upper bound on PT's lip diameter parameter |
17
+
| **lip aperture range** | min/max vertical lip distance across the corpus | lip-protrusion → -aperture mapping for PT |
18
+
| **philtrum length** | landmark 0 (subnasale) → 13 (upper lip cupid's bow center) | minor; informs lip-anterior cavity |
19
+
| **nose-to-chin (subnasale-menton)** | landmark 2 → 152 | proxy for oral-cavity vertical span |
20
+
| **nostril width** | landmarks 219 ↔ 438 | scales nasal cavity volume in PT's nasal port |
21
+
22
+
we do **not** try to estimate velum position, soft-palate angle, or pharyngeal length from external photographs — those need MRI or videofluoroscopy to be honest. they stay free in the fit, bounded only by the published anatomical envelope.
23
+
24
+
## landmarking stack
25
+
26
+
```
27
+
photo (jpg / heic / webp)
28
+
→ opencv read + heif decode (libheif via pillow-heif)
29
+
→ mediapipe face mesh (468 3D landmarks)
30
+
→ insightface buffalo_l (5 landmarks, identity scoring,
31
+
cross-validates that this IS jeffrey;
32
+
reuses the existing portraits/jeffrey
33
+
pipeline's identity threshold)
34
+
→ DECA / EMOCA optional (FLAME params, 3D head fit;
35
+
skipped on first pass — mediapipe is
36
+
enough for ratios)
37
+
```
38
+
39
+
mediapipe is the workhorse. it gives metric 3D landmarks (in a canonical
40
+
head-normalized space), so the ratios survive camera distance. for
41
+
absolute scale we need a reference: jeffrey's actual height (known) or a
42
+
known object in the frame (rare in the platter). the cleanest path is to
43
+
**ratio-normalize** all measurements to head-height-units and then bind
44
+
to absolute centimeters at one point: VTL.
45
+
46
+
## VTL estimation (the load-bearing number)
47
+
48
+
Fitch & Giedd (1999) — JASA 106 — and Lammert & Narayanan (2015) —
49
+
JASA 137 — give regression models from external head measurements to
50
+
internal vocal tract length. simplest defensible form:
51
+
52
+
```
53
+
VTL_male_cm ≈ 17.5 + 0.05 * (height_cm - 175) ± 0.8
54
+
```
55
+
56
+
…with a ±0.8 cm 1σ uncertainty for the population. when we add jeffrey-
57
+
specific head measurements (head-height, bizygomatic), Lammert+Narayanan
58
+
shrink that uncertainty to ~0.4 cm.
59
+
60
+
we use the **lower-uncertainty** estimate as a *narrow prior* on PT's
61
+
tract length parameter, **not** as a hard fix. the fitter can move
62
+
within ±2σ. if the fitter consistently lands at the prior boundary, that
63
+
is a real signal that either (a) the regression is biased for jeffrey or
64
+
(b) PT's idealized tube cannot represent his actual tract. either is a
65
+
research finding, not a bug.
66
+
67
+
## output contract
68
+
69
+
`pop/voice/jeffrey-anthropometry.json` (committed; small, stable):
70
+
71
+
```json
72
+
{
73
+
"version": 1,
74
+
"source_corpus": {
75
+
"shoot": 55,
76
+
"candids": 38,
77
+
"screenshots": 0,
78
+
"rejected_identity": 3,
79
+
"rejected_quality": 7
80
+
},
81
+
"measurements_cm": {
82
+
"head_height": { "median": 23.4, "iqr": [22.9, 23.8], "n": 83 },
83
+
"bizygomatic_width": { "median": 14.1, "iqr": [13.8, 14.4], "n": 83 },
84
+
"mandible_length": { "median": 11.2, "iqr": [10.9, 11.5], "n": 71 },
85
+
"lip_width_neutral": { "median": 5.1, "iqr": [4.9, 5.3], "n": 60 },
86
+
"lip_aperture_max": { "median": 1.8, "iqr": [1.6, 2.0], "n": 22 },
87
+
"philtrum_length": { "median": 1.6, "iqr": [1.5, 1.7], "n": 80 },
88
+
"nose_to_chin": { "median": 6.4, "iqr": [6.2, 6.6], "n": 78 },
89
+
"nostril_width": { "median": 3.2, "iqr": [3.1, 3.3], "n": 65 }
90
+
},
91
+
"derived": {
92
+
"vtl_cm": { "central": 17.6, "ci95": [17.0, 18.2], "model": "lammert-narayanan-2015" },
93
+
"tract_segments_44": { "segment_length_cm": 0.4 },
94
+
"lip_diameter_max_pt": 1.5,
95
+
"mandible_rotation_max_deg": 22
96
+
},
97
+
"notes": [
98
+
"VTL uncertainty narrowed from ±0.8 (Fitch-Giedd, height-only) to ±0.3 (Lammert-Narayanan, multi-feature)",
99
+
"lip_aperture_max samples are scarce — most jeffrey-platter photos are neutral-mouth; resampling needed if singing-voice corpus is added"
100
+
],
101
+
"generated": "2026-05-04",
102
+
"generator": "pop/voice/bin/measure-jeffrey.py v0.1"
103
+
}
104
+
```
105
+
106
+
the JSON is the contract. anything downstream — `fit.py`,
107
+
`render-pt.mjs`, the eventual paper — reads from this file, not from
108
+
re-running the landmarker.
109
+
110
+
## scale-anchor decision
111
+
112
+
absolute scale needs *one* known length. options ranked:
113
+
114
+
1. **jeffrey's stated height** — most reliable, single-number bind, but
115
+
requires jeffrey to fill it in.
116
+
2. **bizygomatic width populational mean** (~13.5–14.5 cm adult male) —
117
+
adds population uncertainty back in.
118
+
3. **interpupillary distance** (median adult ~63 mm) — narrow population
119
+
range, mediapipe gives this directly. defensible default.
120
+
121
+
first pass: option 3 (IPD = 63 mm). if jeffrey provides height, drop in
122
+
option 1 and rerun.
123
+
124
+
## what this is **not**
125
+
126
+
- not a medical-grade measurement. mediapipe landmarks are good to
127
+
~1–2 mm in well-lit frontal portraits, worse at oblique angles.
128
+
- not a clinical tract reconstruction. there is no MRI here. we measure
129
+
what's externally visible, regress to tract length using published
130
+
models, and report the uncertainty honestly.
131
+
- not a face-recognition step. identity validation reuses the existing
132
+
insightface threshold from `portraits/jeffrey/bin/face-match.py` and
133
+
rejects non-jeffrey frames before measurement.
134
+
135
+
## next
136
+
137
+
- [ ] implement `bin/measure-jeffrey.py`
138
+
- [ ] run on all `papers/jeffrey-platter/manifest.json` items, write the JSON
139
+
- [ ] confirm derived VTL is in plausible adult-male range (16.5–18.5 cm)
140
+
- [ ] feed bounds into `fit.py`'s parameter envelope
141
+
142
+
## references
143
+
144
+
- Fitch, W. T., & Giedd, J. (1999). *Morphology and development of the human vocal tract: A study using magnetic resonance imaging.* JASA, 106(3), 1511–1522.
145
+
- Lammert, A. C., & Narayanan, S. S. (2015). *On short-time estimation of vocal tract length from formant frequencies.* JASA, 137(2), 985–995.
146
+
- Story, B. H., Titze, I. R., & Hoffman, E. A. (1996). *Vocal tract area functions from magnetic resonance imaging.* JASA, 100(1), 537–554.
147
+
- Maeda, S. (1990). *Compensatory articulation during speech.* In Speech Production and Speech Modelling.
148
+
- Thapen, N. (2017). *Pink Trombone.* https://dood.al/pinktrombone/
+129
pop/voice/README.md
+129
pop/voice/README.md
···
1
+
# voice — the jeffrey synth lanes
2
+
3
+
a research lane inside `pop/`. **two sub-lanes as of 2026-05-04 pivot:**
4
+
5
+
1. **[`diphones/`](diphones/)** — *the shippable AC-native jeffrey TTS.* jeffrey-pvc ElevenLabs as teacher → forced-aligned diphone bank → C/WASM concat synth. ~3-5 MB shippable assets. Sounds like jeffrey because the diphones ARE jeffrey.
6
+
2. **PT lane (this README + the rest of `pop/voice/`)** — *Pink Trombone as a timbral musical instrument.* Confirmed audibly that PT can't reach phoneme-level intelligibility from external trajectory fitting alone — it's a vocal-tract music tool, not a speech synthesizer. The fit pipeline is still useful: anthropometric priors, single-pose vowel matching, and big-pictures-style sustained-vowel timbres.
7
+
8
+
The PT lane was the first attempt; the pivot below records why it can't ship as jeffrey's voice but stays useful for music. The diphones lane is the new mainline for shippable jeffrey speech.
9
+
10
+
---
11
+
12
+
## PT lane (timbral instrument)
13
+
14
+
original goal — a tiny, runnable, anatomically grounded jeffrey voice — a physical model of jeffrey's vocal tract that we can synthesize from in C/WASM at AC-native budgets — derived from two existing AC corpora:
15
+
16
+
1. the **jeffrey-pvc** ElevenLabs clone (the audio side — already canonical, already wired to `/api/say`)
17
+
2. the **jeffrey-platter** photographic corpus (55 AV-shoot headshots + 38 candid masters + 90 first-person POV captures) — the visual side, already POI-annotated
18
+
19
+
the synthesis layer is **pink trombone** (neil thapen, 2017) — a browser-side physical model of the vocal tract, ~500 lines of JS, MIT-ish, ports cleanly to C and WASM. PT is the structural prior. jeffrey-pvc is the acoustic target. the platter photos give us the *anatomical* prior so the search lives inside a realistic envelope, not a hacky one.
20
+
21
+
> "the gear was the gate, the gate was the price tag" — `pop/big-pictures/plork.txt`
22
+
23
+
the gate here is generic-adult-male defaults. the gate is the price tag.
24
+
25
+
## why this lane exists
26
+
27
+
the prior frankenstein attempt — slicing jeffrey-pvc per phoneme and stitching — produced incoherent monsters because the slices come from different acoustic contexts. PT is a *physical* model: optimizing tract parameters lands on something coherent by construction. and once it's a physical model, the parameters mean something — they're tongue position, lip aperture, glottal tension — not opaque embedding deltas.
28
+
29
+
dropped 2026-05-03: the AC-native formant-resonator path. it sounded like tones, not a voice.
30
+
31
+
picked up 2026-05-03: the pink trombone path, with anatomical priors derived from the platter, with the question of whether the data can also *improve* PT itself.
32
+
33
+
## pipeline
34
+
35
+
```
36
+
platter photos ──▶ measure-jeffrey.py (mediapipe / insightface 3D landmarks)
37
+
│
38
+
▼
39
+
PHYSIOLOGY.md priors
40
+
(VTL, lip aperture range, mandible ROM, nasal scale)
41
+
│
42
+
▼
43
+
jeffrey-pvc /api/say ──▶ record-corpus.mjs ──▶ corpus/raw/*.wav
44
+
│
45
+
▼
46
+
fit.py (CMA-ES + MFCC distance, bounded by priors)
47
+
│
48
+
▼
49
+
fits/<phoneme>.json (per-phoneme PT param trajectories)
50
+
│
51
+
▼
52
+
render.{c,wasm,mjs} (jeffrey voice at synth time)
53
+
```
54
+
55
+
every step caches to disk. reruns cost $0 (the only paid step is the corpus
56
+
recording, and that's content-hashed).
57
+
58
+
## file map
59
+
60
+
```
61
+
README.md — this file
62
+
CORPUS.md — phoneme target list + recording protocol
63
+
PHYSIOLOGY.md — anthropometric pipeline (platter → tract priors)
64
+
MODEL-EXTENSIONS.md — research outline: how the physiology data could
65
+
improve pink trombone itself, not just bound it
66
+
NOTICE.md — third-party license tracking (PT vendor)
67
+
68
+
bin/
69
+
vendor-pt.sh — clone neil thapen's pink trombone into vendor/
70
+
record-corpus.mjs — hits /api/say with jeffrey-pvc per phoneme
71
+
measure-jeffrey.py — extracts anthropometry from platter photos
72
+
fit.py — CMA-ES fit of PT params to recorded WAV
73
+
render-pt.mjs — node-side PT renderer (params → wav, headless)
74
+
75
+
corpus/
76
+
phonemes.json — machine-readable target list
77
+
raw/ — recorded jeffrey-pvc WAVs (gitignored)
78
+
79
+
fits/ — per-phoneme PT param trajectories (gitignored)
80
+
vendor/
81
+
pinktrombone/ — vendored thapen source (gitignored; see NOTICE.md)
82
+
jeffrey-anthropometry.json — derived priors (committed; small JSON)
83
+
```
84
+
85
+
## status
86
+
87
+
- 2026-05-04 — lane scaffolded. all docs landed, all bin/ scripts are stubs.
88
+
- 2026-05-04 — vendored from `dood.al/pinktrombone` directly (single-file HTML, **MIT, Neil Thapen 2017**). A github mirror (`zakaton/Pink-Trombone`) was tried first but turned out GPL-3.0 after refactoring; rejected. License decision resolved — see [NOTICE.md](NOTICE.md).
89
+
- 2026-05-04 — `bin/render-pt.mjs` working. Loads PT inline source via node `vm` sandbox (drift-free; re-extracts from upstream HTML on every call). Renders six sustained vowels (`ah ee oo aa schwa hum`) to WAV in ~90 ms each at 22050 Hz. Smoke set on `~/Desktop` is the audible proof.
90
+
- 2026-05-04 — `bin/fit.py` working. CMA-ES + librosa MFCC distance, calls `render-pt.mjs` as a subprocess. PT→PT self-test recovers every articulator within tolerance. Real fit on the smoke `pt-ee.wav` with `iy` seed: loss 27.8 → 2.5 over 200 evals (~36 s). Recovered params lined up tightly: `tongueDiameter Δ -0.02`, `f0 Δ +0.6 Hz`, `tenseness Δ +0.006`, `lipMul ≈ 1.0`.
91
+
- 2026-05-04 — `bin/measure-jeffrey.py` working (mediapipe Tasks API). Ran on 521 confirmed-jeffrey IG photos → **363 measurements** (74% detection rate). [`jeffrey-anthropometry.json`](jeffrey-anthropometry.json) committed: 8 facial measurements (medians + IQRs) + Fitch-Giedd VTL prior (17.5 cm ± 0.8 σ; population male median, no stated body height yet). `lip_mul_upper_bound = 1.0` consumed by `fit.py`'s bounds clamp.
92
+
- 2026-05-04 — smoke recording: 3 jeffrey-pvc clips (/iy/ /ah/ /uw/, iso context) — server returns raw mp3 without with-timestamps for jeffrey-pvc, recorder falls back gracefully.
93
+
- 2026-05-04 — fit loss v0.1 → v0.2: replaced MFCC time-mean (20-d) with log-mel mean+std (128-d) + librosa.pyin F0 distance in log-cents. v0.1 reported f0=120 Hz on jeffrey's /iy/ but the actual jeffrey-pvc F0 was 144.94 Hz — v0.1 was 25 Hz off and didn't know it. v0.2 recovers F0 to within 1.4 Hz. Throughput 3.7 → 1.8 evals/s (pyin overhead).
94
+
- 2026-05-04 — fit-basin finding: with `--sigma0 0.08 --init iy` the optimizer recovers the *anatomically correct* front-tongue /iy/ pose (`tongueIndex=12.04, tongueDiameter=2.67`). With wider sigma it falls into a back-close basin at slightly lower loss — PT can produce similar vowel-like timbres from multiple tract configurations, so the acoustic loss is multimodal. POSE_INITS + tight sigma is the right disambiguator, not loss-function complexity.
95
+
- 2026-05-04 — trajectory rendering landed. `render-pt.mjs` accepts `--keyframes` arrays of `{t, ...params}` interpolated linearly across the duration; PT's tract.movementSpeed adds another smoothing layer. `--word eye` shortcut + WORD_RECIPES table for /aɪ eɪ oʊ aʊ wi/. 2-keyframe joint fit: take 4 ElevenLabs requests for "eye"/"owe"/"we"/"ay", then run fit.py twice per word with `--window-start/--window-end` to recover start + end poses independently. Composed trajectory recovers anatomically correct front/back tongue positions AND the natural rising F0 contour (e.g. "eye" goes 143 → 198 Hz under stress). PT speaking the fitted-from-jeffrey "eye" is `pt-word-eye-fitted.wav` on Desktop.
96
+
- **2026-05-04 — pivot:** Audible test of PT-saying-words confirmed PT cannot reach phoneme-level intelligibility. New mainline for shippable jeffrey speech is **[`diphones/`](diphones/)** — ElevenLabs-distilled diphone bank, ~3-5 MB shippable, pure C concat at synth time. Diphone target list generated: 1527 ARPABet pairs, 532 in tier-1. PT remains useful for big-pictures sustained-vowel timbres; it's just not the speech engine.
97
+
98
+
```bash
99
+
# render a hand-crafted word
100
+
node bin/render-pt.mjs --word eye --duration 1.0 --out ~/Desktop/pt-eye.wav
101
+
# fit a word from jeffrey's recording (two-pass, start + end)
102
+
.venv/bin/python bin/fit.py --target jeff-eye.mp3 --init ah --window-end 0.45 --out fits/eye-start.json
103
+
.venv/bin/python bin/fit.py --target jeff-eye.mp3 --init iy --window-start 0.55 --out fits/eye-end.json
104
+
# (compose keyframes + render — see one-liner in fits/ pipeline)
105
+
```
106
+
107
+
```bash
108
+
# quick smoke run
109
+
node bin/render-pt.mjs --smoke ~/Desktop
110
+
# single vowel with arbitrary params
111
+
node bin/render-pt.mjs --params '{"pose":"iy","f0":110}' --duration 2 --out ~/Desktop/pt-ee.wav
112
+
# fit any target WAV
113
+
.venv/bin/python bin/fit.py --target ~/Desktop/pt-ee.wav --init iy --max-evals 200
114
+
# end-to-end self-test (no target needed)
115
+
.venv/bin/python bin/fit.py --self-test --max-evals 200
116
+
```
117
+
118
+
## related
119
+
120
+
- memory: `reference_pink_trombone.md` — origin notes
121
+
- memory: `feedback_jeffrey_pvc_settings.md` — stability ≥ 0.5 keeps voice identity
122
+
- memory: `feedback_pop_dual_vocal.md` — AC-native vocal lane history (formant attempt dropped)
123
+
- repo: [`pop/RESEARCH-DIRECTION.md`](../RESEARCH-DIRECTION.md) — vocal lane in the wider pop/ research
124
+
- repo: [`papers/jeffrey-platter/`](../../papers/jeffrey-platter/) — photographic corpus index (the data this lane reads)
125
+
- repo: [`portraits/jeffrey/bin/`](../../portraits/jeffrey/bin/) — existing face-match / face-describe pipeline (insightface, GPT-4o); shares dependency surface with `measure-jeffrey.py`
126
+
127
+
---
128
+
129
+
*maintained by @jeffrey — update STATUS as the lane advances*
+345
pop/voice/bin/fit.py
+345
pop/voice/bin/fit.py
···
1
+
#!/usr/bin/env python3
2
+
"""fit.py — CMA-ES fit of Pink Trombone steady-state params to a target WAV.
3
+
4
+
The minimum-viable version: fit one steady-state vowel pose to a target
5
+
recording by minimizing MFCC distance. Per-frame trajectory fitting and
6
+
per-phoneme corpus iteration come next.
7
+
8
+
How it works:
9
+
target.wav → librosa MFCCs → time-mean vector mfcc_target
10
+
objective(x):
11
+
params = denormalize(x)
12
+
wav = subprocess(render-pt.mjs --params <json> --out - --quiet)
13
+
mfcc_render = librosa MFCCs of wav → time-mean
14
+
return ||mfcc_render - mfcc_target||_2
15
+
CMA-ES minimizes objective over a bounded normalized space.
16
+
17
+
The PT param envelope is bounded by `jeffrey-anthropometry.json` if
18
+
present (clamps tract length, lip aperture, etc); otherwise PT defaults.
19
+
20
+
Usage:
21
+
python bin/fit.py --target ~/Desktop/pt-iy.wav --max-evals 200
22
+
python bin/fit.py --target ~/Desktop/pt-iy.wav --init iy --max-evals 200
23
+
python bin/fit.py --self-test # PT→PT, recovers iy
24
+
"""
25
+
from __future__ import annotations
26
+
27
+
import argparse
28
+
import io
29
+
import json
30
+
import subprocess
31
+
import sys
32
+
import time
33
+
import wave
34
+
from dataclasses import dataclass
35
+
from pathlib import Path
36
+
37
+
import cma # type: ignore
38
+
import librosa # type: ignore
39
+
import numpy as np
40
+
41
+
HERE = Path(__file__).resolve().parent
42
+
ROOT = HERE.parent
43
+
RENDER = HERE / "render-pt.mjs"
44
+
ANTHRO_PATH = ROOT / "jeffrey-anthropometry.json"
45
+
FITS = ROOT / "fits"
46
+
47
+
# ─── Parameter space ────────────────────────────────────────────────────
48
+
# Order is fixed (CMA-ES sees a numpy vector). Each entry: name, lo, hi,
49
+
# initial-pose-defaults map (per-pose) to seed CMA-ES near a sensible spot.
50
+
PARAMS = [
51
+
("tongueIndex", 12.0, 29.0),
52
+
("tongueDiameter", 2.05, 3.5),
53
+
("lipMul", 0.0, 1.0),
54
+
("velumOpen", 0.01, 0.4),
55
+
("f0", 80.0, 200.0),
56
+
("tenseness", 0.0, 1.0),
57
+
]
58
+
59
+
POSE_INITS = {
60
+
"iy": [12.9, 2.43, 1.0, 0.01, 120, 0.6],
61
+
"ah": [22.0, 3.5, 1.0, 0.01, 120, 0.6],
62
+
"uw": [27.0, 2.3, 0.6, 0.01, 120, 0.6],
63
+
"schwa": [17.0, 2.8, 1.0, 0.01, 120, 0.6],
64
+
"m": [17.0, 2.8, 0.0, 0.4, 110, 0.55],
65
+
}
66
+
67
+
68
+
def normalize(values: list[float]) -> np.ndarray:
69
+
out = np.zeros(len(PARAMS))
70
+
for i, ((_, lo, hi), v) in enumerate(zip(PARAMS, values)):
71
+
out[i] = (v - lo) / (hi - lo)
72
+
return out
73
+
74
+
75
+
def denormalize(x: np.ndarray) -> dict:
76
+
return {
77
+
name: float(np.clip(x[i], 0, 1) * (hi - lo) + lo)
78
+
for i, (name, lo, hi) in enumerate(PARAMS)
79
+
}
80
+
81
+
82
+
def apply_anthropometry_bounds(bounds_lo: list[float], bounds_hi: list[float]) -> tuple[list[float], list[float]]:
83
+
"""Tighten lo/hi using anthropometry priors if available."""
84
+
if not ANTHRO_PATH.exists():
85
+
return bounds_lo, bounds_hi
86
+
anthro = json.loads(ANTHRO_PATH.read_text())
87
+
derived = anthro.get("derived", {})
88
+
lip_mul_max = derived.get("lip_mul_upper_bound")
89
+
if lip_mul_max is not None:
90
+
idx = next(i for i, (n, _, _) in enumerate(PARAMS) if n == "lipMul")
91
+
bounds_hi[idx] = min(bounds_hi[idx], float(lip_mul_max))
92
+
return bounds_lo, bounds_hi
93
+
94
+
95
+
# ─── Render bridge ──────────────────────────────────────────────────────
96
+
@dataclass
97
+
class RenderConfig:
98
+
duration_s: float = 1.0
99
+
sample_rate: int = 22050
100
+
seed: int = 1
101
+
102
+
103
+
def render_pt(params: dict, cfg: RenderConfig) -> tuple[np.ndarray, int]:
104
+
"""Spawn render-pt.mjs as a subprocess; return (mono float32, sr)."""
105
+
body = {**params, "durationS": cfg.duration_s, "sampleRate": cfg.sample_rate, "seed": cfg.seed}
106
+
proc = subprocess.run(
107
+
["node", str(RENDER), "--params", json.dumps(body), "--out", "-", "--quiet"],
108
+
capture_output=True, check=True,
109
+
)
110
+
wav_bytes = proc.stdout
111
+
with wave.open(io.BytesIO(wav_bytes), "rb") as w:
112
+
sr = w.getframerate()
113
+
n = w.getnframes()
114
+
raw = w.readframes(n)
115
+
samples = np.frombuffer(raw, dtype="<i2").astype(np.float32) / 32767.0
116
+
return samples, sr
117
+
118
+
119
+
def load_target(path: Path, window_start: float = 0.0,
120
+
window_end: float = 1.0) -> tuple[np.ndarray, int]:
121
+
y, sr = librosa.load(str(path), sr=None, mono=True)
122
+
if (window_start > 0.0) or (window_end < 1.0):
123
+
n = y.shape[0]
124
+
a = int(window_start * n)
125
+
b = int(window_end * n)
126
+
y = y[a:b]
127
+
return y.astype(np.float32), int(sr)
128
+
129
+
130
+
# ─── Spectral distance ──────────────────────────────────────────────────
131
+
# v0.1 used MFCC time-mean — 20 cepstral coefficients averaged. Lost
132
+
# discriminative power between front-close and back-close vowels in the
133
+
# first jeffrey-pvc fit (landed at tongueIndex=28 instead of ~12 for /iy/).
134
+
#
135
+
# v0.2 uses log-mel-spectrogram time-mean *and* time-std — 64 mel bins
136
+
# preserve the spectral envelope (formant pattern) better than 20
137
+
# cepstral coefficients, and the std term captures dynamic variation
138
+
# vs. steady-state. Plus a per-frame MFCC distance via DTW for time-
139
+
# robustness, weighted lighter than the spectral envelope term.
140
+
141
+
def crop_ramps(y: np.ndarray, sr: int) -> np.ndarray:
142
+
head = int(0.1 * sr)
143
+
tail = int(0.15 * sr)
144
+
if y.shape[0] > head + tail + sr * 0.2:
145
+
y = y[head:-tail] if tail else y[head:]
146
+
return y
147
+
148
+
149
+
def spectral_envelope_features(y: np.ndarray, sr: int, n_mels: int = 64) -> np.ndarray:
150
+
"""64-bin log-mel time-mean + time-std → 128-dim feature."""
151
+
y = crop_ramps(y, sr)
152
+
if y.shape[0] < int(sr * 0.2):
153
+
return np.zeros(n_mels * 2, dtype=np.float32)
154
+
S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=n_mels,
155
+
n_fft=2048, hop_length=512, fmax=sr / 2)
156
+
log_S = librosa.power_to_db(S, ref=np.max)
157
+
mean = log_S.mean(axis=1)
158
+
std = log_S.std(axis=1)
159
+
return np.concatenate([mean, std]).astype(np.float32)
160
+
161
+
162
+
def estimate_f0(y: np.ndarray, sr: int) -> float | None:
163
+
"""librosa pyin → median F0 over voiced frames."""
164
+
y = crop_ramps(y, sr)
165
+
if y.shape[0] < int(sr * 0.2):
166
+
return None
167
+
try:
168
+
f0, voiced, _ = librosa.pyin(y, fmin=70, fmax=350, sr=sr,
169
+
frame_length=2048)
170
+
except Exception:
171
+
return None
172
+
f0v = f0[~np.isnan(f0)]
173
+
if f0v.size < 10:
174
+
return None
175
+
return float(np.median(f0v))
176
+
177
+
178
+
def spectral_distance(y_target: np.ndarray, sr_target: int,
179
+
y_render: np.ndarray, sr_render: int,
180
+
f0_target: float | None) -> float:
181
+
"""Composite distance: log-mel envelope (load-bearing) + F0 penalty.
182
+
183
+
The F0 penalty is needed because the log-mel mean is largely insensitive
184
+
to small pitch shifts — but mismatched F0 audibly breaks identity.
185
+
"""
186
+
if sr_target != sr_render:
187
+
y_render = librosa.resample(y_render, orig_sr=sr_render, target_sr=sr_target)
188
+
sr_render = sr_target
189
+
190
+
a = spectral_envelope_features(y_target, sr_target)
191
+
b = spectral_envelope_features(y_render, sr_render)
192
+
envelope_dist = float(np.linalg.norm(a - b)) / np.sqrt(a.size) # per-bin RMS
193
+
194
+
f0_pen = 0.0
195
+
if f0_target is not None:
196
+
f0_render = estimate_f0(y_render, sr_render)
197
+
if f0_render is not None:
198
+
# cents-style log distance, scaled to roughly match envelope dist
199
+
f0_pen = abs(np.log2(f0_render / f0_target)) * 10.0
200
+
201
+
return envelope_dist + f0_pen
202
+
203
+
204
+
# ─── Fit driver ─────────────────────────────────────────────────────────
205
+
def fit(target_path: Path, init_pose: str | None, max_evals: int,
206
+
population: int, sigma0: float, duration_s: float,
207
+
window_start: float = 0.0, window_end: float = 1.0) -> dict:
208
+
y_target, sr_target = load_target(target_path, window_start, window_end)
209
+
cfg = RenderConfig(duration_s=duration_s, sample_rate=sr_target)
210
+
f0_target = estimate_f0(y_target, sr_target)
211
+
212
+
bounds_lo = [lo for _, lo, _ in PARAMS]
213
+
bounds_hi = [hi for _, _, hi in PARAMS]
214
+
bounds_lo, bounds_hi = apply_anthropometry_bounds(bounds_lo, bounds_hi)
215
+
216
+
if init_pose and init_pose in POSE_INITS:
217
+
x0 = normalize(POSE_INITS[init_pose])
218
+
else:
219
+
# midpoint of bounds in normalized space
220
+
x0 = np.full(len(PARAMS), 0.5)
221
+
222
+
es = cma.CMAEvolutionStrategy(
223
+
x0, sigma0,
224
+
{
225
+
"bounds": [[0.0] * len(PARAMS), [1.0] * len(PARAMS)],
226
+
"popsize": population,
227
+
"maxfevals": max_evals,
228
+
"verbose": -9,
229
+
},
230
+
)
231
+
232
+
history = []
233
+
n_evals = 0
234
+
t0 = time.time()
235
+
best = {"loss": float("inf"), "params": None, "eval": -1}
236
+
237
+
while not es.stop() and n_evals < max_evals:
238
+
xs = es.ask()
239
+
losses = []
240
+
for x in xs:
241
+
params = denormalize(np.asarray(x))
242
+
y_r, sr_r = render_pt(params, cfg)
243
+
loss = spectral_distance(y_target, sr_target, y_r, sr_r, f0_target)
244
+
losses.append(loss)
245
+
n_evals += 1
246
+
if loss < best["loss"]:
247
+
best = {"loss": loss, "params": params, "eval": n_evals}
248
+
if n_evals >= max_evals:
249
+
break
250
+
es.tell(xs[: len(losses)], losses)
251
+
history.append({"eval": n_evals, "best_loss": best["loss"], "gen_min": float(min(losses))})
252
+
elapsed = time.time() - t0
253
+
eps = n_evals / elapsed
254
+
sys.stdout.write(
255
+
f"\r eval {n_evals:4d}/{max_evals} best {best['loss']:7.3f} gen-min {min(losses):7.3f} ({eps:4.1f} eval/s)"
256
+
)
257
+
sys.stdout.flush()
258
+
259
+
print()
260
+
return {
261
+
"target": str(target_path),
262
+
"init_pose": init_pose,
263
+
"evals": n_evals,
264
+
"elapsed_s": round(time.time() - t0, 2),
265
+
"best_loss": round(best["loss"], 4),
266
+
"best_params": best["params"],
267
+
"best_eval": best["eval"],
268
+
"history": history,
269
+
"f0_target_hz": round(f0_target, 2) if f0_target else None,
270
+
"loss_version": "log_mel_envelope+f0_log_cents (v0.2)",
271
+
"bounds_used": {n: [lo, hi] for (n, _, _), lo, hi in zip(PARAMS, bounds_lo, bounds_hi)},
272
+
}
273
+
274
+
275
+
# ─── Self-test (PT → PT) ────────────────────────────────────────────────
276
+
def self_test(max_evals: int) -> int:
277
+
"""Render a known-pose vowel via PT, then fit it back. The recovered
278
+
params should be close to the source (or at minimum, the loss should
279
+
drop substantially below a random baseline)."""
280
+
cfg = RenderConfig(duration_s=1.0)
281
+
truth = {"tongueIndex": 12.9, "tongueDiameter": 2.43, "lipMul": 1.0,
282
+
"velumOpen": 0.01, "f0": 120.0, "tenseness": 0.6, "loudness": 0.8}
283
+
print(f"→ self-test: render truth /iy/, then fit it")
284
+
print(f" truth: {truth}")
285
+
y, sr = render_pt(truth, cfg)
286
+
target = Path("/tmp/pt-self-test-target.wav")
287
+
with wave.open(str(target), "wb") as w:
288
+
w.setnchannels(1); w.setsampwidth(2); w.setframerate(sr)
289
+
w.writeframes((np.clip(y, -1, 1) * 32767).astype("<i2").tobytes())
290
+
print(f" target wav: {target}")
291
+
292
+
result = fit(target, init_pose=None, max_evals=max_evals,
293
+
population=8, sigma0=0.4, duration_s=cfg.duration_s)
294
+
print(f" best loss: {result['best_loss']}")
295
+
print(f" best params: {result['best_params']}")
296
+
# Diagnostic — distance per param dimension
297
+
for name, _, _ in PARAMS:
298
+
recovered = result["best_params"][name]
299
+
truth_v = truth[name]
300
+
print(f" {name:18s} truth {truth_v:7.3f} fit {recovered:7.3f} Δ {recovered - truth_v:+.3f}")
301
+
return 0
302
+
303
+
304
+
# ─── CLI ────────────────────────────────────────────────────────────────
305
+
def main() -> int:
306
+
parser = argparse.ArgumentParser()
307
+
parser.add_argument("--target", type=Path, help="path to target WAV")
308
+
parser.add_argument("--init", type=str, default=None,
309
+
choices=list(POSE_INITS.keys()),
310
+
help="seed pose for CMA-ES start point")
311
+
parser.add_argument("--max-evals", type=int, default=200)
312
+
parser.add_argument("--population", type=int, default=8)
313
+
parser.add_argument("--sigma0", type=float, default=0.3)
314
+
parser.add_argument("--duration", type=float, default=1.0,
315
+
help="render duration to compare; should match target")
316
+
parser.add_argument("--self-test", action="store_true")
317
+
parser.add_argument("--out", type=Path, default=None,
318
+
help="write fit JSON here (default: fits/<basename>.json)")
319
+
parser.add_argument("--window-start", type=float, default=0.0,
320
+
help="crop target to [start, end] (fraction of duration); "
321
+
"use to fit one pose to part of a transitional word")
322
+
parser.add_argument("--window-end", type=float, default=1.0)
323
+
args = parser.parse_args()
324
+
325
+
if args.self_test:
326
+
return self_test(args.max_evals)
327
+
328
+
if args.target is None:
329
+
parser.error("--target is required (or use --self-test)")
330
+
331
+
win_label = (f" window=[{args.window_start:.2f}, {args.window_end:.2f}]"
332
+
if (args.window_start > 0 or args.window_end < 1) else "")
333
+
print(f"→ fit {args.target} init={args.init} max-evals={args.max_evals} pop={args.population}{win_label}")
334
+
result = fit(args.target, args.init, args.max_evals, args.population,
335
+
args.sigma0, args.duration,
336
+
args.window_start, args.window_end)
337
+
out_path = args.out or FITS / f"{args.target.stem}.json"
338
+
out_path.parent.mkdir(exist_ok=True)
339
+
out_path.write_text(json.dumps(result, indent=2) + "\n")
340
+
print(f"✓ wrote {out_path}")
341
+
return 0
342
+
343
+
344
+
if __name__ == "__main__":
345
+
raise SystemExit(main())
+370
pop/voice/bin/measure-jeffrey.py
+370
pop/voice/bin/measure-jeffrey.py
···
1
+
#!/usr/bin/env python3
2
+
"""measure-jeffrey.py — anthropometric measurement pass over the platter.
3
+
4
+
Reads photographs (default: `portraits/jeffrey/corpus/shoot/*.jpg`),
5
+
runs MediaPipe Face Mesh with iris-landmarks (468 + 4 iris), extracts
6
+
the head-anthropometry features defined in PHYSIOLOGY.md, anchors
7
+
absolute scale via interpupillary distance (population mean 6.30 cm),
8
+
applies the Lammert+Narayanan VTL regression, and writes
9
+
`pop/voice/jeffrey-anthropometry.json` — the prior file that `fit.py`
10
+
clamps Pink Trombone's parameter envelope against.
11
+
12
+
Identity validation is skipped on the default corpus (the AV shoot is
13
+
all jeffrey by construction). When run against the IG archive — where
14
+
non-jeffrey faces appear — wire `--identity-jsonl` to a face-match.py
15
+
output and we'll filter to confirmed-jeffrey rows.
16
+
17
+
Run:
18
+
pip install mediapipe opencv-python numpy # in pop/.venv
19
+
python bin/measure-jeffrey.py
20
+
python bin/measure-jeffrey.py --glob 'portraits/jeffrey/corpus/shoot/*.jpg'
21
+
python bin/measure-jeffrey.py --debug-out /tmp/measure-debug
22
+
"""
23
+
from __future__ import annotations
24
+
25
+
import argparse
26
+
import glob as globlib
27
+
import json
28
+
import math
29
+
import sys
30
+
from dataclasses import dataclass, field
31
+
from pathlib import Path
32
+
from statistics import median, quantiles
33
+
from typing import Iterator
34
+
35
+
import cv2 # type: ignore
36
+
import mediapipe as mp # type: ignore
37
+
from mediapipe.tasks import python as mp_python # type: ignore
38
+
from mediapipe.tasks.python import vision as mp_vision # type: ignore
39
+
import numpy as np
40
+
41
+
HERE = Path(__file__).resolve().parent
42
+
ROOT = HERE.parent # pop/voice/
43
+
REPO = ROOT.parent.parent # aesthetic-computer/
44
+
PLATTER = REPO / "papers" / "jeffrey-platter"
45
+
OUT_PATH = ROOT / "jeffrey-anthropometry.json"
46
+
DEFAULT_GLOB = str(REPO / "portraits" / "jeffrey" / "corpus" / "shoot" / "*.jpg")
47
+
MODEL_PATH = ROOT / ".cache" / "face_landmarker.task"
48
+
49
+
# ─── MediaPipe Face Mesh landmark indices ──────────────────────────────
50
+
# 468-point canonical face mesh. These indices are stable across mediapipe
51
+
# versions. Cross-reference against
52
+
# https://github.com/google/mediapipe/blob/master/mediapipe/python/solutions/face_mesh.py
53
+
LMK = {
54
+
"forehead": 10, # vertex (top of head proxy — actual vertex is above mesh)
55
+
"chin": 152, # menton
56
+
"left_zygion": 234,
57
+
"right_zygion": 454,
58
+
"left_gonion": 172,
59
+
"right_gonion": 397,
60
+
"left_cheilion": 61,
61
+
"right_cheilion": 291,
62
+
"subnasale": 2,
63
+
"upper_lip_top": 0,
64
+
"upper_lip_cupid": 13,
65
+
"lower_lip_bot": 17,
66
+
"left_alar": 219,
67
+
"right_alar": 438,
68
+
"left_iris_center": 468, # iris landmarks (require refined-landmarks=True)
69
+
"right_iris_center": 473,
70
+
}
71
+
72
+
# Population priors (used when the corresponding measurement is absent
73
+
# from a frame — never as a substitute for measurement).
74
+
INTERPUPILLARY_DISTANCE_CM = 6.30 # adult mean (Dodgson 2004)
75
+
76
+
77
+
# ─── VTL estimation ────────────────────────────────────────────────────
78
+
# We do NOT carry placeholder Lammert+Narayanan coefficients here — the
79
+
# actual paper's regression uses internal MRI-derived measurements that
80
+
# we can't reproduce from external photographs alone. Instead we use
81
+
# Fitch+Giedd 1999's body-height regression as the central estimate
82
+
# (population median when no height is stated), and treat the photo-
83
+
# derived measurements as *diagnostics* + bounds on PT's articulator
84
+
# range — not as inputs to a VTL closed-form.
85
+
#
86
+
# When jeffrey provides --height-cm, we tighten the central estimate
87
+
# accordingly. A real Lammert-style multi-feature fit needs the actual
88
+
# paper coefficients, which are deferred to a future pass.
89
+
90
+
def vtl_fitch_giedd(height_cm: float | None) -> tuple[float, float]:
91
+
"""Body-height VTL regression. Fitch & Giedd 1999.
92
+
93
+
Adult male median: 17.5 cm at 175 cm body height, slope ~0.05 per cm.
94
+
Population residual: ~0.8 cm at 1σ.
95
+
"""
96
+
if height_cm is None:
97
+
return 17.5, 0.8
98
+
return 17.5 + 0.05 * (height_cm - 175.0), 0.8
99
+
100
+
101
+
# ─── Schema ────────────────────────────────────────────────────────────
102
+
@dataclass
103
+
class FrameMeasurements:
104
+
source: str
105
+
# NB. mediapipe FaceMesh has no skull-vertex landmark — landmark 10 is
106
+
# the upper forehead surface. So "head_height" is really
107
+
# forehead-to-chin. Keep that name explicit so we don't accidentally
108
+
# treat it as full cranial height in downstream regressions.
109
+
forehead_to_chin_cm: float | None = None
110
+
bizygomatic_width_cm: float | None = None
111
+
mandible_length_cm: float | None = None
112
+
lip_width_cm: float | None = None
113
+
lip_aperture_cm: float | None = None
114
+
philtrum_cm: float | None = None
115
+
nose_to_chin_cm: float | None = None
116
+
nostril_width_cm: float | None = None
117
+
ipd_px: float | None = None # diagnostic — pixel anchor used for cm scaling
118
+
rejected_reason: str | None = None # e.g. "identity", "no_face", "oblique"
119
+
raw: dict = field(default_factory=dict) # all raw 3D landmarks for debug
120
+
121
+
122
+
# ─── Photo iterator ────────────────────────────────────────────────────
123
+
def iter_glob_photos(pattern: str) -> Iterator[tuple[str, Path]]:
124
+
"""Yield (bucket-tag, path) for files matching the glob. Bucket is
125
+
inferred from the parent directory name."""
126
+
for p in sorted(globlib.glob(pattern)):
127
+
path = Path(p)
128
+
if not path.is_file():
129
+
continue
130
+
yield path.parent.name, path
131
+
132
+
133
+
def iter_paths_file(paths_file: Path) -> Iterator[tuple[str, Path]]:
134
+
"""Yield (bucket-tag, path) for each line in a list-of-paths file."""
135
+
for line in paths_file.read_text().splitlines():
136
+
s = line.strip()
137
+
if not s or s.startswith("#"):
138
+
continue
139
+
path = Path(s)
140
+
if not path.is_file():
141
+
continue
142
+
yield path.parent.name, path
143
+
144
+
145
+
# ─── Landmark → measurement extraction ─────────────────────────────────
146
+
def _dist3(a: np.ndarray, b: np.ndarray) -> float:
147
+
return float(np.linalg.norm(a - b))
148
+
149
+
150
+
def _resize_to_max(image: np.ndarray, max_side: int) -> np.ndarray:
151
+
h, w = image.shape[:2]
152
+
long_side = max(h, w)
153
+
if long_side <= max_side:
154
+
return image
155
+
scale = max_side / long_side
156
+
return cv2.resize(image, (int(w * scale), int(h * scale)), interpolation=cv2.INTER_AREA)
157
+
158
+
159
+
def measure_frame(image: np.ndarray, source: str, landmarker,
160
+
debug_out: Path | None = None) -> FrameMeasurements:
161
+
"""Run MediaPipe FaceLandmarker on the image and extract anthropometry."""
162
+
# mediapipe's detector scales input to ~256-512 internally; on a
163
+
# 6000x4000 master headshot the face occupies <13% of the frame and
164
+
# gets shrunk to <50px which the detector misses. Pre-resize.
165
+
image = _resize_to_max(image, 1280)
166
+
h, w = image.shape[:2]
167
+
rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
168
+
mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb)
169
+
result = landmarker.detect(mp_image)
170
+
if not result.face_landmarks:
171
+
return FrameMeasurements(source=source, rejected_reason="no_face")
172
+
lms = result.face_landmarks[0]
173
+
if len(lms) < 478:
174
+
return FrameMeasurements(source=source, rejected_reason="no_iris")
175
+
176
+
# Pull all required landmarks as (x_px, y_px, z_relative).
177
+
pt = lambda i: np.array([lms[i].x * w, lms[i].y * h, lms[i].z * w], dtype=np.float64)
178
+
P = {name: pt(idx) for name, idx in LMK.items()}
179
+
180
+
# Anchor scale via interpupillary distance.
181
+
ipd_px = _dist3(P["left_iris_center"][:2], P["right_iris_center"][:2])
182
+
if ipd_px < 10:
183
+
return FrameMeasurements(source=source, rejected_reason="oblique_or_small")
184
+
cm_per_px = INTERPUPILLARY_DISTANCE_CM / ipd_px
185
+
186
+
def cm(a: np.ndarray, b: np.ndarray) -> float:
187
+
return _dist3(a[:2], b[:2]) * cm_per_px
188
+
189
+
fm = FrameMeasurements(
190
+
source=source,
191
+
forehead_to_chin_cm = cm(P["forehead"], P["chin"]),
192
+
bizygomatic_width_cm = cm(P["left_zygion"], P["right_zygion"]),
193
+
mandible_length_cm = cm(P["left_gonion"], P["chin"]),
194
+
lip_width_cm = cm(P["left_cheilion"], P["right_cheilion"]),
195
+
lip_aperture_cm = cm(P["upper_lip_top"], P["lower_lip_bot"]),
196
+
philtrum_cm = cm(P["subnasale"], P["upper_lip_cupid"]),
197
+
nose_to_chin_cm = cm(P["subnasale"], P["chin"]),
198
+
nostril_width_cm = cm(P["left_alar"], P["right_alar"]),
199
+
ipd_px = ipd_px,
200
+
)
201
+
202
+
if debug_out is not None:
203
+
debug_out.mkdir(parents=True, exist_ok=True)
204
+
annotated = image.copy()
205
+
for name, p in P.items():
206
+
x, y = int(p[0]), int(p[1])
207
+
cv2.circle(annotated, (x, y), 4, (0, 255, 200), -1)
208
+
cv2.putText(annotated, name, (x + 6, y - 6),
209
+
cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 200), 1)
210
+
cv2.imwrite(str(debug_out / (Path(source).stem + "__landmarks.jpg")), annotated)
211
+
return fm
212
+
213
+
214
+
def load_image(path: Path) -> np.ndarray | None:
215
+
img = cv2.imread(str(path))
216
+
return img
217
+
218
+
219
+
# ─── Aggregation ───────────────────────────────────────────────────────
220
+
def aggregate(frames: list[FrameMeasurements]) -> dict:
221
+
"""Median + IQR for each measurement, ignoring None values."""
222
+
fields = [
223
+
"forehead_to_chin_cm", "bizygomatic_width_cm", "mandible_length_cm",
224
+
"lip_width_cm", "lip_aperture_cm", "philtrum_cm",
225
+
"nose_to_chin_cm", "nostril_width_cm",
226
+
]
227
+
out = {}
228
+
for f in fields:
229
+
values = [getattr(fr, f) for fr in frames if getattr(fr, f) is not None]
230
+
if not values:
231
+
out[f] = {"median": None, "iqr": None, "n": 0}
232
+
continue
233
+
if len(values) >= 4:
234
+
q = quantiles(values, n=4, method="inclusive")
235
+
iqr = [round(q[0], 2), round(q[2], 2)]
236
+
else:
237
+
iqr = [round(min(values), 2), round(max(values), 2)]
238
+
out[f] = {"median": round(median(values), 2), "iqr": iqr, "n": len(values)}
239
+
return out
240
+
241
+
242
+
def derive_pt_priors(measurements: dict, height_cm: float | None) -> dict:
243
+
"""Convert anthropometry into Pink Trombone parameter priors.
244
+
245
+
For now, VTL central is from Fitch-Giedd (body-height-based).
246
+
The photo-derived measurements bound the *articulator* params
247
+
(lip aperture, etc.) but are not yet fed into a closed-form VTL.
248
+
"""
249
+
vtl, sigma = vtl_fitch_giedd(height_cm)
250
+
251
+
# Lip aperture max (cm, from corpus) → PT's lipMul upper bound.
252
+
# PT's restDiameter[lipStart..n-1] = 1.5 * lipMul (where 1.5 cm is the
253
+
# PT-default rest aperture). So a measured 2.02 cm max → lipMul ≤ 2.02/1.5 ≈ 1.35.
254
+
# In normalized [0..1] CMA-ES space we still cap at 1.0 so we don't
255
+
# over-open relative to PT's calibration.
256
+
lip_p75 = (measurements.get("lip_aperture_cm", {}).get("iqr") or [None, None])[1]
257
+
lip_max_cm = lip_p75 if lip_p75 is not None else 1.5
258
+
return {
259
+
"vtl_cm": {
260
+
"central": round(vtl, 2),
261
+
"ci95": [round(vtl - 1.96 * sigma, 2), round(vtl + 1.96 * sigma, 2)],
262
+
"model": "fitch-giedd-1999",
263
+
"input": ("body-height-stated" if height_cm is not None else "population-median-male"),
264
+
},
265
+
# PT default: 44 segments × 0.396 cm each = 17.4 cm. Personalize
266
+
# by varying segment length, keeping segment count constant.
267
+
"tract_segments_44": {"segment_length_cm": round(vtl / 44.0, 4)},
268
+
"lip_aperture_max_cm": round(lip_max_cm, 2),
269
+
"lip_mul_upper_bound": round(min(1.0, lip_max_cm / 1.5), 3),
270
+
"mandible_rotation_max_deg": 22, # placeholder; real value needs jaw-open vs jaw-closed pair
271
+
"notes": [
272
+
"VTL central uses Fitch+Giedd 1999 body-height regression.",
273
+
"Photo-derived measurements (forehead-to-chin, bizygomatic, mandible) are not yet wired into a closed-form VTL — Lammert-Narayanan 2015 needs MRI-derived inputs we don't have.",
274
+
"Lip aperture upper bound comes from corpus 75th percentile (max would be too sensitive to selfie outliers).",
275
+
],
276
+
}
277
+
278
+
279
+
def main() -> int:
280
+
parser = argparse.ArgumentParser()
281
+
parser.add_argument("--glob", default=DEFAULT_GLOB,
282
+
help=f"glob of input images (default: AV shoot cache)")
283
+
parser.add_argument("--paths-file", type=Path, default=None,
284
+
help="read newline-separated image paths from file (overrides --glob)")
285
+
parser.add_argument("--limit", type=int, default=None,
286
+
help="cap on photos processed")
287
+
parser.add_argument("--height-cm", type=float, default=None,
288
+
help="jeffrey's stated body height; tightens VTL prior via Fitch-Giedd")
289
+
parser.add_argument("--debug-out", type=Path, default=None,
290
+
help="write annotated landmark images here")
291
+
parser.add_argument("--out", type=Path, default=OUT_PATH)
292
+
args = parser.parse_args()
293
+
294
+
if not MODEL_PATH.exists():
295
+
print(f"✗ model file missing: {MODEL_PATH}", file=sys.stderr)
296
+
print(" download: curl -sSLfo .cache/face_landmarker.task https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/latest/face_landmarker.task",
297
+
file=sys.stderr)
298
+
return 1
299
+
options = mp_vision.FaceLandmarkerOptions(
300
+
base_options=mp_python.BaseOptions(model_asset_path=str(MODEL_PATH)),
301
+
running_mode=mp_vision.RunningMode.IMAGE,
302
+
num_faces=1,
303
+
output_face_blendshapes=False,
304
+
output_facial_transformation_matrixes=False,
305
+
min_face_detection_confidence=0.5,
306
+
)
307
+
landmarker = mp_vision.FaceLandmarker.create_from_options(options)
308
+
309
+
frames: list[FrameMeasurements] = []
310
+
rejected: dict[str, int] = {}
311
+
counts: dict[str, int] = {}
312
+
313
+
photo_iter = (iter_paths_file(args.paths_file) if args.paths_file
314
+
else iter_glob_photos(args.glob))
315
+
for i, (bucket, path) in enumerate(photo_iter):
316
+
if args.limit is not None and i >= args.limit:
317
+
break
318
+
counts[bucket] = counts.get(bucket, 0) + 1
319
+
img = load_image(path)
320
+
if img is None:
321
+
fm = FrameMeasurements(source=str(path), rejected_reason="read_failed")
322
+
else:
323
+
fm = measure_frame(img, str(path), landmarker, debug_out=args.debug_out)
324
+
frames.append(fm)
325
+
if fm.rejected_reason:
326
+
rejected[fm.rejected_reason] = rejected.get(fm.rejected_reason, 0) + 1
327
+
print(f" - {path.name} rejected: {fm.rejected_reason}")
328
+
else:
329
+
print(f" ✓ {path.name} ipd_px={fm.ipd_px:.1f} "
330
+
f"forehead-chin={fm.forehead_to_chin_cm:.1f}cm bizy={fm.bizygomatic_width_cm:.1f}cm")
331
+
332
+
landmarker.close()
333
+
334
+
if not frames:
335
+
print("✗ no images matched", file=sys.stderr)
336
+
return 1
337
+
338
+
measurements = aggregate(frames)
339
+
derived = derive_pt_priors(measurements, args.height_cm)
340
+
341
+
out = {
342
+
"version": 1,
343
+
"generated": "2026-05-04",
344
+
"generator": "pop/voice/bin/measure-jeffrey.py v0.2",
345
+
"source_glob": args.glob,
346
+
"source_corpus": counts,
347
+
"rejected": rejected,
348
+
"measurements_cm": measurements,
349
+
"derived": derived,
350
+
"scale_anchor": {
351
+
"method": "interpupillary distance",
352
+
"ipd_cm_used": INTERPUPILLARY_DISTANCE_CM,
353
+
"ipd_px_median": round(median([f.ipd_px for f in frames if f.ipd_px]), 1)
354
+
if any(f.ipd_px for f in frames) else None,
355
+
"rationale": "Adult-population mean (Dodgson 2004). Replace with stated height + Lammert 2015 if/when --height-cm is provided.",
356
+
},
357
+
"notes": [
358
+
"Single-pass mediapipe FaceMesh on local AV-shoot cache.",
359
+
"Identity validation skipped (AV shoot is jeffrey by construction). Wire --identity-jsonl when running against IG archive.",
360
+
],
361
+
}
362
+
args.out.write_text(json.dumps(out, indent=2) + "\n")
363
+
n_ok = sum(1 for f in frames if not f.rejected_reason)
364
+
print(f"\n✓ wrote {args.out}")
365
+
print(f" frames: {len(frames)} measured: {n_ok} rejected: {sum(rejected.values())}")
366
+
return 0
367
+
368
+
369
+
if __name__ == "__main__":
370
+
raise SystemExit(main())
+203
pop/voice/bin/record-corpus.mjs
+203
pop/voice/bin/record-corpus.mjs
···
1
+
#!/usr/bin/env node
2
+
// record-corpus.mjs — generate the jeffrey harness phoneme corpus by
3
+
// hitting /api/say with jeffrey-pvc, three carrier contexts per phoneme.
4
+
//
5
+
// Mirrors the cache pattern of pop/bin/say.mjs: every request is hashed
6
+
// against `{ text, voice, settings }` and skipped if the cached WAV +
7
+
// alignment match. Reruns cost $0 unless phonemes.json changes.
8
+
//
9
+
// Usage:
10
+
// node bin/record-corpus.mjs # full corpus
11
+
// node bin/record-corpus.mjs --only iy,ah,schwa # subset
12
+
// node bin/record-corpus.mjs --context iso # just iso recordings
13
+
// node bin/record-corpus.mjs --dry-run # plan only, no API
14
+
// node bin/record-corpus.mjs --force # bypass cache
15
+
16
+
import {
17
+
writeFileSync, readFileSync, mkdirSync, existsSync,
18
+
} from "node:fs";
19
+
import { resolve, dirname } from "node:path";
20
+
import { fileURLToPath } from "node:url";
21
+
import { createHash } from "node:crypto";
22
+
23
+
const HERE = dirname(fileURLToPath(import.meta.url));
24
+
const ROOT = resolve(HERE, "..");
25
+
const CORPUS = resolve(ROOT, "corpus");
26
+
const PHONEMES = JSON.parse(readFileSync(resolve(CORPUS, "phonemes.json"), "utf8"));
27
+
28
+
const argv = process.argv.slice(2);
29
+
const flags = {};
30
+
for (let i = 0; i < argv.length; i++) {
31
+
const a = argv[i];
32
+
if (a.startsWith("--")) {
33
+
const key = a.slice(2);
34
+
const next = argv[i + 1];
35
+
if (next !== undefined && !next.startsWith("--")) { flags[key] = next; i++; }
36
+
else flags[key] = true;
37
+
}
38
+
}
39
+
40
+
const ONLY = flags.only ? new Set(String(flags.only).split(",")) : null;
41
+
const CONTEXTS = flags.context
42
+
? [String(flags.context)]
43
+
: ["iso", "digit", "cvc"];
44
+
const DRY = flags["dry-run"] === true;
45
+
const FORCE = flags.force === true;
46
+
47
+
const SETTINGS = PHONEMES.settings;
48
+
const ENDPOINT = "https://aesthetic.computer/api/say";
49
+
50
+
const manifest = {
51
+
version: 1,
52
+
generated: new Date().toISOString(),
53
+
endpoint: ENDPOINT,
54
+
settings: SETTINGS,
55
+
items: {},
56
+
};
57
+
58
+
function hashOf(body) {
59
+
return createHash("sha256").update(JSON.stringify(body)).digest("hex").slice(0, 16);
60
+
}
61
+
62
+
function buildBody(text, isIso, phoneme) {
63
+
const speed = isIso
64
+
? (phoneme.iso_speed ?? SETTINGS.iso_default_speed)
65
+
: SETTINGS.default_speed;
66
+
return {
67
+
from: text,
68
+
provider: SETTINGS.voice.provider,
69
+
voice: SETTINGS.voice.voice,
70
+
stability: SETTINGS.stability,
71
+
similarity: SETTINGS.similarity,
72
+
style: SETTINGS.style,
73
+
speed,
74
+
withTimestamps: true,
75
+
};
76
+
}
77
+
78
+
async function fetchOne(text, body, outMp3, outAlign) {
79
+
const res = await fetch(ENDPOINT, {
80
+
method: "POST",
81
+
headers: { "Content-Type": "application/json" },
82
+
body: JSON.stringify(body),
83
+
redirect: "follow",
84
+
});
85
+
if (!res.ok) throw new Error(`/api/say ${res.status}: ${await res.text()}`);
86
+
const ct = res.headers.get("content-type") || "";
87
+
if (ct.includes("application/json")) {
88
+
// Server returned JSON-with-timestamps. Save audio + alignment.
89
+
const json = await res.json();
90
+
const buf = Buffer.from(json.audio, "base64");
91
+
writeFileSync(outMp3, buf);
92
+
const a = json.alignment || {};
93
+
const chars = a.characters || [];
94
+
const starts = a.character_start_times_seconds || [];
95
+
const ends = a.character_end_times_seconds || [];
96
+
const words = [];
97
+
let i = 0;
98
+
while (i < chars.length) {
99
+
if (/\s/.test(chars[i])) { i++; continue; }
100
+
const wStart = starts[i];
101
+
let txt = "";
102
+
let lastEnd = ends[i];
103
+
while (i < chars.length && !/\s/.test(chars[i])) {
104
+
txt += chars[i]; lastEnd = ends[i]; i++;
105
+
}
106
+
words.push({ text: txt, fromMs: Math.round(wStart * 1000), toMs: Math.round(lastEnd * 1000) });
107
+
}
108
+
writeFileSync(outAlign, JSON.stringify({
109
+
source: "elevenlabs/with-timestamps",
110
+
text, voice: `${body.provider}/${body.voice}`,
111
+
characters: chars, char_starts_s: starts, char_ends_s: ends, words,
112
+
}, null, 2));
113
+
return { bytes: buf.length, words: words.length, hasAlignment: true };
114
+
}
115
+
// Fallback: server returned raw mp3 (jeffrey-pvc currently doesn't
116
+
// support /with-timestamps). Save the audio; alignment can be re-derived
117
+
// later via WhisperX if needed.
118
+
const buf = Buffer.from(await res.arrayBuffer());
119
+
writeFileSync(outMp3, buf);
120
+
writeFileSync(outAlign, JSON.stringify({
121
+
source: "elevenlabs/no-timestamps",
122
+
text, voice: `${body.provider}/${body.voice}`,
123
+
note: "server returned raw mp3; no character-level alignment available",
124
+
}, null, 2));
125
+
return { bytes: buf.length, words: 0, hasAlignment: false };
126
+
}
127
+
128
+
let plannedRequests = 0;
129
+
let executedRequests = 0;
130
+
let cachedHits = 0;
131
+
const errors = [];
132
+
133
+
for (const phoneme of PHONEMES.phonemes) {
134
+
if (ONLY && !ONLY.has(phoneme.id)) continue;
135
+
manifest.items[phoneme.id] = {
136
+
ipa: phoneme.ipa,
137
+
class: phoneme.class,
138
+
contexts: {},
139
+
};
140
+
141
+
for (const ctx of CONTEXTS) {
142
+
const text = phoneme[ctx];
143
+
if (!text) continue;
144
+
plannedRequests++;
145
+
const body = buildBody(text, ctx === "iso", phoneme);
146
+
const inputHash = hashOf(body);
147
+
148
+
const dir = resolve(CORPUS, "raw", ctx);
149
+
mkdirSync(dir, { recursive: true });
150
+
const stem = resolve(dir, phoneme.id);
151
+
const outMp3 = `${stem}.mp3`;
152
+
const outAlign = `${stem}.json`;
153
+
const outHash = `${stem}.hash`;
154
+
155
+
const cached =
156
+
!FORCE &&
157
+
existsSync(outMp3) &&
158
+
existsSync(outAlign) &&
159
+
existsSync(outHash) &&
160
+
readFileSync(outHash, "utf8").trim() === inputHash;
161
+
162
+
manifest.items[phoneme.id].contexts[ctx] = {
163
+
text,
164
+
hash: inputHash,
165
+
mp3: `corpus/raw/${ctx}/${phoneme.id}.mp3`,
166
+
alignment: `corpus/raw/${ctx}/${phoneme.id}.json`,
167
+
cached,
168
+
};
169
+
170
+
if (cached) {
171
+
cachedHits++;
172
+
continue;
173
+
}
174
+
if (DRY) {
175
+
console.log(` [plan] ${phoneme.id}/${ctx}: ${JSON.stringify(text)}`);
176
+
continue;
177
+
}
178
+
try {
179
+
console.log(`→ ${phoneme.id}/${ctx} (${text.length} chars)`);
180
+
const r = await fetchOne(text, body, outMp3, outAlign);
181
+
writeFileSync(outHash, inputHash + "\n");
182
+
executedRequests++;
183
+
console.log(` ✓ ${(r.bytes / 1024).toFixed(0)} KB · ${r.words} words · hash ${inputHash}`);
184
+
} catch (err) {
185
+
errors.push({ phoneme: phoneme.id, ctx, error: err.message });
186
+
console.error(` ✗ ${err.message}`);
187
+
}
188
+
}
189
+
}
190
+
191
+
if (!DRY) {
192
+
writeFileSync(resolve(CORPUS, "manifest.json"), JSON.stringify(manifest, null, 2));
193
+
}
194
+
195
+
console.log("");
196
+
console.log(`planned: ${plannedRequests}`);
197
+
console.log(`executed: ${executedRequests}`);
198
+
console.log(`cached: ${cachedHits}`);
199
+
if (errors.length) {
200
+
console.log(`errors: ${errors.length}`);
201
+
for (const e of errors) console.log(` - ${e.phoneme}/${e.ctx}: ${e.error}`);
202
+
process.exit(1);
203
+
}
+482
pop/voice/bin/render-pt.mjs
+482
pop/voice/bin/render-pt.mjs
···
1
+
#!/usr/bin/env node
2
+
// render-pt.mjs — headless Pink Trombone renderer.
3
+
//
4
+
// Loads vendor/pinktrombone/index.html, extracts Thapen's Glottis +
5
+
// Tract object literals + simplex noise lib in-place (no transcription,
6
+
// drift-free against upstream), runs them inside a node vm sandbox with
7
+
// shimmed globals, and renders parameter trajectories to mono 16-bit
8
+
// PCM WAV. The biquad-filtered aspiration/fricative noise inputs are
9
+
// approximated as white noise — close enough for vowel quality, will
10
+
// matter more once fricative phonemes become a target.
11
+
//
12
+
// This is what fit.py's render_pt() will eventually call (via subprocess).
13
+
// For now it's also a smoke-test driver: render a few vowels to disk.
14
+
//
15
+
// Usage:
16
+
// node bin/render-pt.mjs --vowel ah --duration 1.5 --out ~/Desktop/pt-ah.wav
17
+
// node bin/render-pt.mjs --smoke ~/Desktop # render the demo set
18
+
// node bin/render-pt.mjs --params '{"tongueIndex":29,"tongueDiameter":2.43,"f0":120}' \
19
+
// --duration 2 --out out.wav
20
+
21
+
import { readFileSync, writeFileSync, mkdirSync } from "node:fs";
22
+
import { resolve, dirname } from "node:path";
23
+
import { fileURLToPath } from "node:url";
24
+
import { homedir } from "node:os";
25
+
import vm from "node:vm";
26
+
27
+
const HERE = dirname(fileURLToPath(import.meta.url));
28
+
const ROOT = resolve(HERE, "..");
29
+
const PT_HTML = resolve(ROOT, "vendor/pinktrombone/index.html");
30
+
31
+
// ── Source extraction ─────────────────────────────────────────────────
32
+
// PT's HTML carries the entire source inline. We slice out the named
33
+
// regions by anchor strings that have been stable since v1.1 (March
34
+
// 2017). If any of these slices fail, the upstream's been re-formatted
35
+
// and we should re-pin.
36
+
function extractRegion(src, startMarker, endMarker, name) {
37
+
const start = src.indexOf(startMarker);
38
+
if (start < 0) throw new Error(`extract '${name}': missing start marker ${JSON.stringify(startMarker)}`);
39
+
const end = src.indexOf(endMarker, start);
40
+
if (end < 0) throw new Error(`extract '${name}': missing end marker ${JSON.stringify(endMarker)}`);
41
+
return src.slice(start, end + endMarker.length);
42
+
}
43
+
44
+
function loadPT() {
45
+
const html = readFileSync(PT_HTML, "utf8");
46
+
47
+
// Math.clamp + Math.moveTowards (two definitions; the second wins,
48
+
// which is the four-arg form we need)
49
+
const mathHelpers = extractRegion(html, "Math.clamp = function", "Math.gaussian", "math-helpers")
50
+
.replace(/Math\.gaussian[\s\S]*$/, "");
51
+
52
+
// Simplex noise IIFE — public domain (Stefan Gustavson)
53
+
const simplexNoise = extractRegion(html, "(function(global){", "})(this);", "simplex-noise");
54
+
55
+
// Glottis + Tract literals — match by their stable terminator pattern.
56
+
const glottisMatch = html.match(/var Glottis =\s*\{[\s\S]*?\n\}\s*\n\s*\n\s*var Tract/);
57
+
if (!glottisMatch) throw new Error("could not extract Glottis block");
58
+
const glottisBlock = glottisMatch[0].replace(/\n\s*var Tract$/, "");
59
+
60
+
const tractMatch = html.match(/var Tract =\s*\{[\s\S]*?\n\};\s*\n/);
61
+
if (!tractMatch) throw new Error("could not extract Tract block");
62
+
const tractBlock = tractMatch[0];
63
+
64
+
return { mathHelpers, simplexNoise, glottisBlock, tractBlock };
65
+
}
66
+
67
+
// ── Sandbox + boot ────────────────────────────────────────────────────
68
+
function bootSandbox(sampleRate) {
69
+
const sandbox = {
70
+
Math: Object.create(Math),
71
+
Float64Array,
72
+
Float32Array,
73
+
Array,
74
+
Date,
75
+
console,
76
+
sampleRate,
77
+
alwaysVoice: false,
78
+
autoWobble: false,
79
+
backCtx: makeNullCtx(),
80
+
UI: { touchesWithMouse: [] },
81
+
AudioSystem: { blockLength: 512, blockTime: 512 / sampleRate },
82
+
palePink: "#FFEEF5",
83
+
time: 0,
84
+
noise: null,
85
+
};
86
+
sandbox.global = sandbox;
87
+
sandbox.window = sandbox;
88
+
sandbox.this = sandbox;
89
+
vm.createContext(sandbox);
90
+
91
+
const { mathHelpers, simplexNoise, glottisBlock, tractBlock } = loadPT();
92
+
93
+
// Run Math helpers (extends sandbox.Math.clamp, Math.moveTowards)
94
+
vm.runInContext(mathHelpers, sandbox);
95
+
// Run simplex noise — installs `noise` on the global (the IIFE writes
96
+
// global.noise = {…}). Our sandbox is the global, so this works.
97
+
vm.runInContext(simplexNoise.replace("})(this);", "})(this);\nthis.noise = noise;"), sandbox);
98
+
// Glottis + Tract literals
99
+
vm.runInContext(glottisBlock + ";\n" + tractBlock, sandbox);
100
+
101
+
// Skip Glottis.init (it draws the keyboard); init waveform manually
102
+
sandbox.Glottis.setupWaveform(0);
103
+
sandbox.Tract.init();
104
+
return sandbox;
105
+
}
106
+
107
+
function makeNullCtx() {
108
+
const noop = () => {};
109
+
return new Proxy({}, { get: () => noop, set: () => true });
110
+
}
111
+
112
+
// ── Parameter shape (the contract fit.py optimizes over) ──────────────
113
+
// All values in PT's native units. The fitter passes a flat dict; it
114
+
// can override any subset, the rest fall back to the per-pose defaults.
115
+
// tongueIndex [12, 29] front-back tongue position
116
+
// tongueDiameter [2.05, 3.5] close-open tongue height
117
+
// lipMul [0, 1.0] lip aperture multiplier (0=closed, 1=open)
118
+
// velumOpen [0.01, 0.4] nasal port (0.01=closed, 0.4=hum)
119
+
// f0 Hz fundamental
120
+
// tenseness [0, 1] glottal tension
121
+
// loudness [0, 1]
122
+
//
123
+
// `pose` (optional) names a vowel preset whose values fill any unset
124
+
// fields. Useful for relative perturbations during fitting.
125
+
126
+
// ── Word recipes — keyframe arrays for short transitions ──────────────
127
+
// `t` is normalized [0, 1] across the render duration. PT's
128
+
// Tract.movementSpeed (15 cm/s) handles fine-grained smoothing inside
129
+
// the gap between any two keyframes for free.
130
+
//
131
+
// These are hand-crafted from the canonical IPA pose values. The fitter
132
+
// can refine them against jeffrey-pvc word recordings later — this is
133
+
// just the "sounds plausible from the phonetics" baseline.
134
+
const WORD_RECIPES = {
135
+
// /aɪ/ "eye" — back-open → front-close glide, mostly voiced
136
+
eye: [
137
+
{ t: 0.0, pose: "ah", f0: 130, tenseness: 0.65 },
138
+
{ t: 0.55, pose: "ah", f0: 130, tenseness: 0.65 }, // hold the start
139
+
{ t: 1.0, pose: "iy", f0: 145, tenseness: 0.60 },
140
+
],
141
+
// /oʊ/ "owe" — back-open-rounded → back-close-rounded
142
+
owe: [
143
+
{ t: 0.0, pose: "aa", f0: 125, tenseness: 0.65, lipMulOverride: 0.65 },
144
+
{ t: 0.5, pose: "aa", f0: 125, tenseness: 0.65, lipMulOverride: 0.55 },
145
+
{ t: 1.0, pose: "uw", f0: 130, tenseness: 0.60 },
146
+
],
147
+
// /eɪ/ "ay" — mid-front → close-front
148
+
ay: [
149
+
{ t: 0.0, tongueIndex: 14.5, tongueDiameter: 2.95, lipMul: 1.0, velumOpen: 0.01, f0: 130, tenseness: 0.6 },
150
+
{ t: 0.5, tongueIndex: 14.5, tongueDiameter: 2.95, lipMul: 1.0, velumOpen: 0.01, f0: 130, tenseness: 0.6 },
151
+
{ t: 1.0, pose: "iy", f0: 140, tenseness: 0.6 },
152
+
],
153
+
// /wi/ "we" — back-close-rounded glide → front-close
154
+
we: [
155
+
{ t: 0.0, pose: "uw", f0: 130, tenseness: 0.55 },
156
+
{ t: 0.25, pose: "uw", f0: 132, tenseness: 0.6 },
157
+
{ t: 1.0, pose: "iy", f0: 145, tenseness: 0.6 },
158
+
],
159
+
// /aʊ/ "ow!" — back-open → back-close-rounded
160
+
ow: [
161
+
{ t: 0.0, pose: "ah", f0: 130, tenseness: 0.7 },
162
+
{ t: 0.5, pose: "ah", f0: 130, tenseness: 0.7 },
163
+
{ t: 1.0, pose: "uw", f0: 130, tenseness: 0.55 },
164
+
],
165
+
};
166
+
167
+
// ── Vowel poses (defaults) ────────────────────────────────────────────
168
+
const VOWEL_POSES = {
169
+
// front-close
170
+
iy: { tongueIndex: 12.9, tongueDiameter: 2.43, velumOpen: 0.01, lipsRound: false },
171
+
// front-open
172
+
ae: { tongueIndex: 14, tongueDiameter: 3.4, velumOpen: 0.01, lipsRound: false },
173
+
// open central
174
+
ah: { tongueIndex: 22.0, tongueDiameter: 3.5, velumOpen: 0.01, lipsRound: false },
175
+
// back-open
176
+
aa: { tongueIndex: 25.0, tongueDiameter: 3.5, velumOpen: 0.01, lipsRound: false },
177
+
// back-close, rounded
178
+
uw: { tongueIndex: 27.0, tongueDiameter: 2.3, velumOpen: 0.01, lipsRound: true },
179
+
// schwa — neutral
180
+
schwa: { tongueIndex: 17.0, tongueDiameter: 2.8, velumOpen: 0.01, lipsRound: false },
181
+
// nasal hum (closed lips, velum open)
182
+
m: { tongueIndex: 17.0, tongueDiameter: 2.8, velumOpen: 0.4, lipsRound: false, lipsClosed: true },
183
+
};
184
+
185
+
// Apply tract pose into Tract.targetDiameter / restDiameter using PT's
186
+
// own setRestDiameter math (transcribed from TractUI line 1541).
187
+
// `params` is a fully-resolved object (no pose lookup here).
188
+
function applyTractParams(s, params) {
189
+
const Tract = s.Tract;
190
+
const tongueIndex = params.tongueIndex;
191
+
const tongueDiameter = params.tongueDiameter;
192
+
const lipMul = params.lipMul;
193
+
const gridOffset = 1.7;
194
+
for (let i = Tract.bladeStart; i < Tract.lipStart; i++) {
195
+
const t = 1.1 * Math.PI * (tongueIndex - i) / (Tract.tipStart - Tract.bladeStart);
196
+
const fixedTongueDiameter = 2 + (tongueDiameter - 2) / 1.5;
197
+
let curve = (1.5 - fixedTongueDiameter + gridOffset) * Math.cos(t);
198
+
if (i == Tract.bladeStart - 2 || i == Tract.lipStart - 1) curve *= 0.8;
199
+
if (i == Tract.bladeStart || i == Tract.lipStart - 2) curve *= 0.94;
200
+
Tract.restDiameter[i] = 1.5 - curve;
201
+
}
202
+
// Lip aperture region (PT segments lipStart..n-1)
203
+
for (let i = Tract.lipStart; i < Tract.n; i++) {
204
+
Tract.restDiameter[i] = 1.5 * lipMul;
205
+
}
206
+
for (let i = 0; i < Tract.n; i++) Tract.targetDiameter[i] = Tract.restDiameter[i];
207
+
Tract.velumTarget = params.velumOpen;
208
+
}
209
+
210
+
function resolveParams(input) {
211
+
// Resolve a partial input against an optional pose preset.
212
+
let base;
213
+
if (input.pose && VOWEL_POSES[input.pose]) {
214
+
const p = VOWEL_POSES[input.pose];
215
+
base = {
216
+
tongueIndex: p.tongueIndex,
217
+
tongueDiameter: p.tongueDiameter,
218
+
lipMul: p.lipsClosed ? 0.0 : (p.lipsRound ? 0.6 : 1.0),
219
+
velumOpen: p.velumOpen ?? 0.01,
220
+
f0: 120,
221
+
tenseness: 0.6,
222
+
loudness: 0.8,
223
+
};
224
+
} else {
225
+
// Schwa-ish defaults if no pose given.
226
+
base = {
227
+
tongueIndex: 17, tongueDiameter: 2.8, lipMul: 1.0, velumOpen: 0.01,
228
+
f0: 120, tenseness: 0.6, loudness: 0.8,
229
+
};
230
+
}
231
+
return { ...base, ...input };
232
+
}
233
+
234
+
// Resolve a single keyframe entry's pose-or-explicit-params shape.
235
+
function resolveKeyframe(kf) {
236
+
const r = resolveParams(kf);
237
+
if (kf.lipMulOverride !== undefined) r.lipMul = kf.lipMulOverride;
238
+
return { t: kf.t, params: r };
239
+
}
240
+
241
+
// Linear interpolation between two resolved param dicts.
242
+
function lerpParams(a, b, alpha) {
243
+
const out = {};
244
+
const keys = ["tongueIndex", "tongueDiameter", "lipMul", "velumOpen",
245
+
"f0", "tenseness", "loudness"];
246
+
for (const k of keys) out[k] = a[k] * (1 - alpha) + b[k] * alpha;
247
+
return out;
248
+
}
249
+
250
+
// Find the params at a given normalized time t ∈ [0, 1] from a sorted
251
+
// keyframe array, by piecewise linear interpolation.
252
+
function paramsAtT(keyframes, t) {
253
+
if (t <= keyframes[0].t) return keyframes[0].params;
254
+
if (t >= keyframes[keyframes.length - 1].t) return keyframes[keyframes.length - 1].params;
255
+
for (let i = 0; i < keyframes.length - 1; i++) {
256
+
const a = keyframes[i], b = keyframes[i + 1];
257
+
if (t >= a.t && t <= b.t) {
258
+
const span = (b.t - a.t) || 1;
259
+
const alpha = (t - a.t) / span;
260
+
return lerpParams(a.params, b.params, alpha);
261
+
}
262
+
}
263
+
return keyframes[keyframes.length - 1].params;
264
+
}
265
+
266
+
// ── Render ─────────────────────────────────────────────────────────────
267
+
function render(input = {}) {
268
+
const sampleRate = input.sampleRate ?? 22050;
269
+
const durationS = input.durationS ?? 1.5;
270
+
const voiced = input.voiced ?? true;
271
+
const rampInS = input.rampInS ?? 0.05;
272
+
const rampOutS = input.rampOutS ?? 0.1;
273
+
const seed = input.seed ?? 1;
274
+
275
+
// Resolve keyframes if provided; otherwise treat input as a single
276
+
// steady-state pose (back-compat with v0.1 callers).
277
+
let keyframes = null;
278
+
if (input.keyframes && Array.isArray(input.keyframes) && input.keyframes.length > 0) {
279
+
keyframes = input.keyframes
280
+
.map(resolveKeyframe)
281
+
.sort((x, y) => x.t - y.t);
282
+
}
283
+
const initialParams = keyframes ? keyframes[0].params : resolveParams(input);
284
+
285
+
const s = bootSandbox(sampleRate);
286
+
applyTractParams(s, initialParams);
287
+
288
+
// Glottis driving values (UI* are the inputs PT reads each block)
289
+
s.Glottis.UIFrequency = initialParams.f0;
290
+
s.Glottis.UITenseness = initialParams.tenseness;
291
+
s.Glottis.smoothFrequency = initialParams.f0;
292
+
s.Glottis.loudness = initialParams.loudness;
293
+
s.alwaysVoice = voiced;
294
+
s.autoWobble = false; // determinism for tests; fitter can flip on
295
+
s.noise.seed(seed);
296
+
297
+
const N = Math.round(durationS * sampleRate);
298
+
const out = new Float32Array(N);
299
+
const blockLen = s.AudioSystem.blockLength;
300
+
const rampInN = Math.round(rampInS * sampleRate);
301
+
const rampOutStart = N - Math.round(rampOutS * sampleRate);
302
+
303
+
// Pre-allocate noise buffer for a block.
304
+
const aspNoise = new Float32Array(blockLen);
305
+
const fricNoise = new Float32Array(blockLen);
306
+
307
+
for (let blockStart = 0; blockStart < N; blockStart += blockLen) {
308
+
// Inject keyframe-interpolated params for this block (if a
309
+
// trajectory was given). PT's tract.movementSpeed adds another
310
+
// layer of smoothing on top of these per-block targets.
311
+
if (keyframes) {
312
+
const tNorm = blockStart / Math.max(1, N - 1);
313
+
const p = paramsAtT(keyframes, tNorm);
314
+
applyTractParams(s, p);
315
+
s.Glottis.UIFrequency = p.f0;
316
+
s.Glottis.UITenseness = p.tenseness;
317
+
s.Glottis.loudness = p.loudness;
318
+
}
319
+
// Fill noise inputs (white noise; biquad filtering is a v0.2 feature).
320
+
for (let j = 0; j < blockLen; j++) {
321
+
aspNoise[j] = Math.random() * 2 - 1;
322
+
fricNoise[j] = Math.random() * 2 - 1;
323
+
}
324
+
const blockN = Math.min(blockLen, N - blockStart);
325
+
for (let j = 0; j < blockN; j++) {
326
+
const lambda1 = j / blockLen;
327
+
const lambda2 = (j + 0.5) / blockLen;
328
+
const glottalOutput = s.Glottis.runStep(lambda1, aspNoise[j]);
329
+
let vocal = 0;
330
+
s.Tract.runStep(glottalOutput, fricNoise[j], lambda1);
331
+
vocal += s.Tract.lipOutput + s.Tract.noseOutput;
332
+
s.Tract.runStep(glottalOutput, fricNoise[j], lambda2);
333
+
vocal += s.Tract.lipOutput + s.Tract.noseOutput;
334
+
let sample = vocal * 0.125;
335
+
// soft attack/release (PT's intensity ramp handles most of this,
336
+
// but a hard tail-clip on stop avoids the residual tract ringing
337
+
// bleeding into a fade)
338
+
const i = blockStart + j;
339
+
if (i < rampInN) sample *= i / rampInN;
340
+
else if (i >= rampOutStart) sample *= Math.max(0, (N - i) / (N - rampOutStart));
341
+
out[i] = sample;
342
+
}
343
+
s.Glottis.finishBlock();
344
+
s.Tract.finishBlock();
345
+
}
346
+
return { audio: out, sampleRate };
347
+
}
348
+
349
+
// ── WAV encoder (16-bit PCM mono) ─────────────────────────────────────
350
+
function encodeWav(float32, sampleRate) {
351
+
const N = float32.length;
352
+
const buf = Buffer.alloc(44 + N * 2);
353
+
// RIFF header
354
+
buf.write("RIFF", 0);
355
+
buf.writeUInt32LE(36 + N * 2, 4);
356
+
buf.write("WAVE", 8);
357
+
buf.write("fmt ", 12);
358
+
buf.writeUInt32LE(16, 16);
359
+
buf.writeUInt16LE(1, 20); // PCM
360
+
buf.writeUInt16LE(1, 22); // mono
361
+
buf.writeUInt32LE(sampleRate, 24);
362
+
buf.writeUInt32LE(sampleRate * 2, 28);
363
+
buf.writeUInt16LE(2, 32);
364
+
buf.writeUInt16LE(16, 34);
365
+
buf.write("data", 36);
366
+
buf.writeUInt32LE(N * 2, 40);
367
+
// Find peak for safe normalization (target -3 dBFS).
368
+
let peak = 1e-9;
369
+
for (let i = 0; i < N; i++) peak = Math.max(peak, Math.abs(float32[i]));
370
+
const gain = 0.7079 / peak; // -3 dBFS
371
+
for (let i = 0; i < N; i++) {
372
+
const v = Math.max(-1, Math.min(1, float32[i] * gain));
373
+
buf.writeInt16LE(Math.round(v * 32767), 44 + i * 2);
374
+
}
375
+
return buf;
376
+
}
377
+
378
+
function expandHome(p) {
379
+
if (!p) return p;
380
+
if (p === "~") return homedir();
381
+
if (p.startsWith("~/")) return resolve(homedir(), p.slice(2));
382
+
return p;
383
+
}
384
+
385
+
// ── CLI ───────────────────────────────────────────────────────────────
386
+
async function main() {
387
+
const argv = process.argv.slice(2);
388
+
const flags = {};
389
+
for (let i = 0; i < argv.length; i++) {
390
+
const a = argv[i];
391
+
if (a.startsWith("--")) {
392
+
const key = a.slice(2);
393
+
const next = argv[i + 1];
394
+
if (next !== undefined && !next.startsWith("--")) { flags[key] = next; i++; }
395
+
else flags[key] = true;
396
+
}
397
+
}
398
+
399
+
if (flags.smoke) {
400
+
const dest = expandHome(typeof flags.smoke === "string" ? flags.smoke : `${homedir()}/Desktop`);
401
+
mkdirSync(dest, { recursive: true });
402
+
const set = [
403
+
{ name: "pt-ah", pose: "ah", f0: 120, tenseness: 0.6, durationS: 1.5 },
404
+
{ name: "pt-ee", pose: "iy", f0: 120, tenseness: 0.6, durationS: 1.5 },
405
+
{ name: "pt-oo", pose: "uw", f0: 120, tenseness: 0.6, durationS: 1.5 },
406
+
{ name: "pt-aa", pose: "aa", f0: 120, tenseness: 0.65, durationS: 1.5 },
407
+
{ name: "pt-schwa", pose: "schwa", f0: 110, tenseness: 0.55, durationS: 1.5 },
408
+
{ name: "pt-hum", pose: "m", f0: 110, tenseness: 0.55, durationS: 1.5 },
409
+
];
410
+
console.log(`→ rendering smoke set into ${dest}`);
411
+
for (const item of set) {
412
+
const t0 = Date.now();
413
+
const { audio, sampleRate } = render(item);
414
+
const wav = encodeWav(audio, sampleRate);
415
+
const out = resolve(dest, `${item.name}.wav`);
416
+
writeFileSync(out, wav);
417
+
console.log(` ✓ ${item.name}.wav (${(wav.length/1024).toFixed(0)} KB · ${item.durationS}s · ${Date.now()-t0}ms)`);
418
+
}
419
+
console.log("\ndone. open ~/Desktop and double-click to listen.");
420
+
return;
421
+
}
422
+
423
+
if (flags.words) {
424
+
// Render every word recipe to ~/Desktop. Optional override of duration.
425
+
const dest = expandHome(typeof flags.words === "string" ? flags.words : `${homedir()}/Desktop`);
426
+
const durationS = Number(flags.duration ?? 1.0);
427
+
mkdirSync(dest, { recursive: true });
428
+
console.log(`→ rendering words into ${dest} (${durationS}s each)`);
429
+
for (const [name, keyframes] of Object.entries(WORD_RECIPES)) {
430
+
const t0 = Date.now();
431
+
const { audio, sampleRate } = render({ keyframes, durationS });
432
+
const wav = encodeWav(audio, sampleRate);
433
+
const out = resolve(dest, `pt-word-${name}.wav`);
434
+
writeFileSync(out, wav);
435
+
console.log(` ✓ pt-word-${name}.wav (${(wav.length/1024).toFixed(0)} KB · ${Date.now()-t0}ms · ${keyframes.length} keyframes)`);
436
+
}
437
+
return;
438
+
}
439
+
440
+
// Single render
441
+
let input;
442
+
if (flags.word) {
443
+
const recipe = WORD_RECIPES[String(flags.word)];
444
+
if (!recipe) {
445
+
console.error(`✗ unknown word: ${flags.word}. options: ${Object.keys(WORD_RECIPES).join(", ")}`);
446
+
process.exit(1);
447
+
}
448
+
input = { keyframes: recipe };
449
+
} else if (flags.params) {
450
+
input = JSON.parse(flags.params);
451
+
} else if (flags.keyframes) {
452
+
input = { keyframes: JSON.parse(flags.keyframes) };
453
+
} else {
454
+
input = { pose: flags.vowel || flags.pose || "schwa" };
455
+
}
456
+
if (flags.duration) input.durationS = Number(flags.duration);
457
+
if (flags.f0) input.f0 = Number(flags.f0);
458
+
if (flags.tenseness) input.tenseness = Number(flags.tenseness);
459
+
if (flags["sample-rate"]) input.sampleRate = Number(flags["sample-rate"]);
460
+
if (flags.seed) input.seed = Number(flags.seed);
461
+
462
+
const out = expandHome(flags.out
463
+
|| `~/Desktop/pt-${flags.word || input.pose || "params"}.wav`);
464
+
const isStdout = out === "-";
465
+
const quiet = flags.quiet === true || isStdout;
466
+
if (!quiet) {
467
+
const summary = JSON.stringify({ ...input, durationS: input.durationS ?? 1.5 });
468
+
console.log(`→ ${summary} → ${out}`);
469
+
}
470
+
const { audio, sampleRate } = render(input);
471
+
const wav = encodeWav(audio, sampleRate);
472
+
if (isStdout) {
473
+
// raw WAV bytes — used by fit.py subprocess
474
+
process.stdout.write(wav);
475
+
} else {
476
+
mkdirSync(dirname(out), { recursive: true });
477
+
writeFileSync(out, wav);
478
+
if (!quiet) console.log(`✓ ${out} (${(wav.length/1024).toFixed(0)} KB)`);
479
+
}
480
+
}
481
+
482
+
main().catch((e) => { console.error(e); process.exit(1); });
+98
pop/voice/bin/vendor-pt.sh
+98
pop/voice/bin/vendor-pt.sh
···
1
+
#!/usr/bin/env bash
2
+
# vendor-pt.sh — fetch Neil Thapen's original Pink Trombone (MIT) into
3
+
# vendor/pinktrombone/.
4
+
#
5
+
# PT lives canonically at https://dood.al/pinktrombone/ as a single
6
+
# self-contained HTML file. The MIT license notice is the comment header
7
+
# at the top of that file (lines 28–46 as of v1.1, March 2017).
8
+
#
9
+
# We fetch the HTML directly with curl rather than cloning a third-party
10
+
# git mirror, because:
11
+
# 1. dood.al IS the canonical source — every github mirror is downstream
12
+
# 2. some mirrors (zakaton/Pink-Trombone, others) re-license under GPL
13
+
# after refactoring; that's incompatible with the AC-native goal of
14
+
# shipping a permissive C/WASM jeffrey voice
15
+
# 3. a single HTML file pinned by sha256 is the simplest possible
16
+
# provenance contract
17
+
#
18
+
# Usage:
19
+
# bin/vendor-pt.sh # fetch from dood.al
20
+
# bin/vendor-pt.sh <url> # override source
21
+
# bin/vendor-pt.sh --dry-run # print what it would do, no fetch
22
+
23
+
set -euo pipefail
24
+
25
+
HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
26
+
ROOT="$(cd "$HERE/.." && pwd)"
27
+
VENDOR_DIR="$ROOT/vendor/pinktrombone"
28
+
29
+
DEFAULT_URL="https://dood.al/pinktrombone/"
30
+
31
+
URL="${1:-$DEFAULT_URL}"
32
+
DRY=0
33
+
[ "${1:-}" = "--dry-run" ] && DRY=1 && URL="$DEFAULT_URL"
34
+
35
+
echo "→ vendoring pink trombone (MIT, Neil Thapen 2017)"
36
+
echo " url: $URL"
37
+
echo " target: $VENDOR_DIR"
38
+
39
+
if [ -f "$VENDOR_DIR/index.html" ] && [ -f "$VENDOR_DIR/UPSTREAM.txt" ]; then
40
+
echo "✗ $VENDOR_DIR already populated. delete it first if you want a fresh fetch."
41
+
exit 1
42
+
fi
43
+
44
+
if [ "$DRY" = 1 ]; then
45
+
echo " (dry run — would curl now)"
46
+
exit 0
47
+
fi
48
+
49
+
mkdir -p "$VENDOR_DIR"
50
+
51
+
TMP="$(mktemp)"
52
+
trap 'rm -f "$TMP"' EXIT
53
+
curl -sSLf "$URL" -o "$TMP"
54
+
55
+
# Sanity-check: confirm the MIT notice we expect is in the file. Refuse
56
+
# to vendor an unverified blob.
57
+
if ! grep -q "Copyright 2017 Neil Thapen" "$TMP"; then
58
+
echo "✗ fetched file missing 'Copyright 2017 Neil Thapen' header — wrong upstream?"
59
+
exit 1
60
+
fi
61
+
if ! grep -q "Permission is hereby granted, free of charge" "$TMP"; then
62
+
echo "✗ fetched file missing MIT preamble — license check failed."
63
+
exit 1
64
+
fi
65
+
66
+
cp "$TMP" "$VENDOR_DIR/index.html"
67
+
68
+
# Extract the MIT block into a stand-alone LICENSE file so anyone reading
69
+
# the vendor tree sees the license without parsing HTML comments.
70
+
awk '
71
+
/Copyright 2017 Neil Thapen/ { writing=1 }
72
+
writing { print }
73
+
/IN THE SOFTWARE\./ { if (writing) { writing=0; exit } }
74
+
' "$TMP" > "$VENDOR_DIR/LICENSE"
75
+
76
+
SHA="$(shasum -a 256 "$VENDOR_DIR/index.html" | awk '{print $1}')"
77
+
SIZE="$(wc -c < "$VENDOR_DIR/index.html" | tr -d ' ')"
78
+
79
+
cat > "$VENDOR_DIR/UPSTREAM.txt" <<EOF
80
+
upstream: $URL
81
+
upstream_url: ${URL}index.html
82
+
fetched_at: $(date -u +"%Y-%m-%dT%H:%M:%SZ")
83
+
sha256: $SHA
84
+
size_bytes: $SIZE
85
+
version: "version 1.1, March 2017" (per index.html line 8)
86
+
author: Neil Thapen <venuspatrol.nfshost.com>
87
+
license: MIT (Copyright 2017 Neil Thapen — see LICENSE in this directory)
88
+
notice: pop/voice/NOTICE.md tracks third-party material for this lane
89
+
EOF
90
+
91
+
echo "✓ vendored pink trombone"
92
+
echo " upstream: $URL"
93
+
echo " sha256: $SHA"
94
+
echo " size: $SIZE bytes"
95
+
echo " files: index.html, LICENSE, UPSTREAM.txt"
96
+
echo
97
+
echo "next: read script/PinkTrombone.js logic from index.html lines 50+,"
98
+
echo " then write bin/render-pt.mjs (node-side renderer)."
+62
pop/voice/corpus/phonemes.json
+62
pop/voice/corpus/phonemes.json
···
1
+
{
2
+
"$schema": "./phonemes.schema.json",
3
+
"version": 1,
4
+
"generated": "2026-05-04",
5
+
"note": "Machine-readable phoneme target list for the jeffrey harness. See pop/voice/CORPUS.md for the design rationale. Each entry has an IPA symbol, an arpabet alias for tools that don't speak IPA, three carrier prompts (iso/digit/cvc), and a 'class' tag for analysis grouping. The 'iso_speed' field is per-entry because diphthongs sustain awkwardly at 0.85 — they need 1.0 to keep the glide audible.",
6
+
"settings": {
7
+
"voice": { "provider": "jeffrey", "voice": "neutral:0" },
8
+
"stability": 0.5,
9
+
"similarity": 0.9,
10
+
"style": 0.0,
11
+
"default_speed": 1.0,
12
+
"iso_default_speed": 0.85
13
+
},
14
+
"phonemes": [
15
+
{ "id": "ae", "ipa": "æ", "arpa": "AE", "class": "short_vowel", "iso": "aaaaa as in cat", "digit": "the cat sat at bat", "cvc": "bat" },
16
+
{ "id": "eh", "ipa": "ɛ", "arpa": "EH", "class": "short_vowel", "iso": "eeeee as in bet", "digit": "let met set bet", "cvc": "bet" },
17
+
{ "id": "ih", "ipa": "ɪ", "arpa": "IH", "class": "short_vowel", "iso": "iiiii as in bit", "digit": "hit fit lit bit", "cvc": "bit" },
18
+
{ "id": "ah", "ipa": "ɒ", "arpa": "AA", "class": "short_vowel", "iso": "ahhhh as in bot", "digit": "got hot lot bot", "cvc": "bot" },
19
+
{ "id": "uh", "ipa": "ʊ", "arpa": "UH", "class": "short_vowel", "iso": "uuuuu as in book", "digit": "look took cook book", "cvc": "book" },
20
+
21
+
{ "id": "iy", "ipa": "iː", "arpa": "IY", "class": "long_vowel", "iso": "eeeee as in see", "digit": "see me three be", "cvc": "see" },
22
+
{ "id": "uw", "ipa": "uː", "arpa": "UW", "class": "long_vowel", "iso": "ooooo as in soup", "digit": "soup loop you do", "cvc": "soup" },
23
+
{ "id": "aa", "ipa": "ɑː", "arpa": "AA2", "class": "long_vowel", "iso": "aaaaa as in spa", "digit": "spa la pa ah", "cvc": "spa" },
24
+
{ "id": "ao", "ipa": "ɔː", "arpa": "AO", "class": "long_vowel", "iso": "aw aw aw as in saw", "digit": "saw paw raw thaw", "cvc": "saw" },
25
+
{ "id": "er", "ipa": "ɜː", "arpa": "ER", "class": "long_vowel", "iso": "errrr as in surf", "digit": "her per stir surf", "cvc": "surf" },
26
+
27
+
{ "id": "ay", "ipa": "aɪ", "arpa": "AY", "class": "diphthong", "iso": "ay ay ay as in eye", "digit": "eye my pie tie", "cvc": "pie", "iso_speed": 1.0 },
28
+
{ "id": "aw", "ipa": "aʊ", "arpa": "AW", "class": "diphthong", "iso": "ow ow ow as in owl", "digit": "owl how now cow", "cvc": "cow", "iso_speed": 1.0 },
29
+
{ "id": "ey", "ipa": "eɪ", "arpa": "EY", "class": "diphthong", "iso": "ay ay ay as in ate", "digit": "ate late date gate", "cvc": "ate", "iso_speed": 1.0 },
30
+
{ "id": "ow", "ipa": "oʊ", "arpa": "OW", "class": "diphthong", "iso": "oh oh oh as in owe", "digit": "owe go so no", "cvc": "owe", "iso_speed": 1.0 },
31
+
{ "id": "oy", "ipa": "ɔɪ", "arpa": "OY", "class": "diphthong", "iso": "oy oy oy as in oil", "digit": "oil boy toy joy", "cvc": "oil", "iso_speed": 1.0 },
32
+
33
+
{ "id": "v", "ipa": "v", "arpa": "V", "class": "voiced_fricative", "iso": "vvvvv as in van", "digit": "van vine vow vote", "cvc": "van" },
34
+
{ "id": "dh", "ipa": "ð", "arpa": "DH", "class": "voiced_fricative", "iso": "this this this", "digit": "this that thus those", "cvc": "this" },
35
+
{ "id": "z", "ipa": "z", "arpa": "Z", "class": "voiced_fricative", "iso": "zzzzz as in zoo", "digit": "zoo buzz fizz haze", "cvc": "zoo" },
36
+
{ "id": "zh", "ipa": "ʒ", "arpa": "ZH", "class": "voiced_fricative", "iso": "zh zh zh as in vision", "digit": "vision azure beige garage", "cvc": "vision" },
37
+
38
+
{ "id": "f", "ipa": "f", "arpa": "F", "class": "unvoiced_fricative", "iso": "fffff as in fan", "digit": "fan fine fee fun", "cvc": "fan" },
39
+
{ "id": "th", "ipa": "θ", "arpa": "TH", "class": "unvoiced_fricative", "iso": "thin thin thin", "digit": "thin think thaw thug", "cvc": "thin" },
40
+
{ "id": "s", "ipa": "s", "arpa": "S", "class": "unvoiced_fricative", "iso": "sssss as in see", "digit": "see say sip sun", "cvc": "see" },
41
+
{ "id": "sh", "ipa": "ʃ", "arpa": "SH", "class": "unvoiced_fricative", "iso": "sh sh sh as in ship", "digit": "ship show sheen shut", "cvc": "ship" },
42
+
43
+
{ "id": "b", "ipa": "b", "arpa": "B", "class": "voiced_plosive", "iso": "buh buh buh", "digit": "buy bee bay bow", "cvc": "buy" },
44
+
{ "id": "d", "ipa": "d", "arpa": "D", "class": "voiced_plosive", "iso": "duh duh duh", "digit": "do day die dough", "cvc": "do" },
45
+
{ "id": "g", "ipa": "g", "arpa": "G", "class": "voiced_plosive", "iso": "guh guh guh", "digit": "go gay guy gum", "cvc": "go" },
46
+
{ "id": "p", "ipa": "p", "arpa": "P", "class": "unvoiced_plosive", "iso": "puh puh puh", "digit": "pie pay pee pow", "cvc": "pie" },
47
+
{ "id": "t", "ipa": "t", "arpa": "T", "class": "unvoiced_plosive", "iso": "tuh tuh tuh", "digit": "two tea tie toe", "cvc": "two" },
48
+
{ "id": "k", "ipa": "k", "arpa": "K", "class": "unvoiced_plosive", "iso": "kuh kuh kuh", "digit": "key cue cow car", "cvc": "key" },
49
+
50
+
{ "id": "m", "ipa": "m", "arpa": "M", "class": "nasal", "iso": "mmmmm as in me", "digit": "me my mow moon", "cvc": "me" },
51
+
{ "id": "n", "ipa": "n", "arpa": "N", "class": "nasal", "iso": "nnnnn as in no", "digit": "no know new neat", "cvc": "no" },
52
+
{ "id": "ng", "ipa": "ŋ", "arpa": "NG", "class": "nasal", "iso": "nggg nggg nggg", "digit": "sing ring king song", "cvc": "sing" },
53
+
54
+
{ "id": "l", "ipa": "l", "arpa": "L", "class": "liquid", "iso": "lllll as in lo", "digit": "lo low lay lip", "cvc": "lo" },
55
+
{ "id": "r", "ipa": "ɹ", "arpa": "R", "class": "liquid", "iso": "rrrrr as in row", "digit": "row ray rye red", "cvc": "row" },
56
+
57
+
{ "id": "w", "ipa": "w", "arpa": "W", "class": "glide", "iso": "wwwww as in we", "digit": "we way why wow", "cvc": "we" },
58
+
{ "id": "y", "ipa": "j", "arpa": "Y", "class": "glide", "iso": "yyyyy as in you", "digit": "you yes yay yam", "cvc": "you" },
59
+
60
+
{ "id": "schwa", "ipa": "ə", "arpa": "AX", "class": "schwa", "iso": "uhhhh", "digit": "the a but um", "cvc": "the" }
61
+
]
62
+
}
+98
pop/voice/diphones/README.md
+98
pop/voice/diphones/README.md
···
1
+
# diphones — the shippable AC-native jeffrey synth
2
+
3
+
The pivot, 2026-05-04: **Pink Trombone is a vocal-tract instrument, not a speech synthesizer.** It can't reach phoneme-level intelligibility from photogrammetric priors + speech samples alone. Confirmed audibly on Desktop's `pt-word-eye-fitted.wav`.
4
+
5
+
This lane reframes around a different question: *what would it take to ship a tiny jeffrey-voiced TTS that runs on AC-native hardware?*
6
+
7
+
## Approach: jeffrey-pvc-distilled diphone bank
8
+
9
+
Use **ElevenLabs jeffrey-pvc as the teacher.** Generate a corpus of carrier sentences chosen to elicit every English diphone (≈1500 transitions). Forced-align with WhisperX (already in the recap pipeline per memory) to extract each ~50 ms slice. Store as a flat lookup: `diphone[a, b] → 50 ms WAV`. Concat with crossfade at synth time.
10
+
11
+
Why this and not Pink Trombone:
12
+
- **It IS jeffrey.** Diphones are jeffrey's actual voice in fragments, not a model's approximation
13
+
- **Tiny shippable size** — 1500 diphones × 50 ms × 22050 Hz × 2 bytes ≈ 3.3 MB raw, less under Opus
14
+
- **No NN at runtime** — pure DSP concat, runs trivially in C/WASM
15
+
- **AC-native by construction** — every diphone is URL-addressable: `assets.aesthetic.computer/jeffrey-diphones/<a>_<b>.opus`
16
+
- **The recap pipeline already has WhisperX** — we reuse the forced-alignment lane
17
+
18
+
Why not (TBD: revisit if diphone bank fails):
19
+
- A tiny RNN/conv vocoder distilled from jeffrey-pvc would be more *natural-sounding* across novel words but adds inference cost + training complexity
20
+
- HTS-style HMM parametric synth is proven shippable (Festvox lineage) but the training pipeline is involved
21
+
22
+
## Cost estimate
23
+
24
+
ARPABet has 39 phonemes + start/end markers ≈ 41². That's 1681 ordered pairs but most are illegal (English doesn't have /ŋ_t/ word-initial, etc.). Realistic English diphone count: **~1500**.
25
+
26
+
Each carrier sentence carries ~10–15 diphones. 1500 / 12 ≈ **~125 carrier sentences**, each ~5 words long. ElevenLabs charges ~30k chars/$5 of credits at jeffrey-pvc rates, so:
27
+
28
+
- 125 carriers × ~25 chars = ~3125 chars
29
+
- That's ~$0.50 of credits
30
+
- Plus regenerations, alignment passes, retries → call it ~**$2-5 in ElevenLabs**
31
+
32
+
This is the cost gate. Negligible compared to a real corpus.
33
+
34
+
## Pipeline
35
+
36
+
```
37
+
diphone-targets.json (the master list of 1500 ARPABet pairs)
38
+
↓
39
+
build-carriers.py (assigns pairs to 125 carrier sentences)
40
+
↓
41
+
carriers.json (sentence text + which diphones it carries)
42
+
↓
43
+
record-carriers.mjs (jeffrey-pvc /api/say with timestamps)
44
+
↓
45
+
carriers/raw/*.mp3 (~125 mp3s)
46
+
↓
47
+
align-carriers.py (WhisperX forced alignment per phoneme)
48
+
↓
49
+
diphone-extraction.py (slice carriers into diphones, ~50 ms each, crossfade-ready)
50
+
↓
51
+
diphones/<a>_<b>.opus (~1500 tiny files)
52
+
↓
53
+
synth.{c,mjs,wasm} (concat + crossfade at synth time)
54
+
```
55
+
56
+
Every step caches by content hash. Reruns cost $0 unless inputs change.
57
+
58
+
## File map (planned)
59
+
60
+
```
61
+
diphones/
62
+
README.md this file
63
+
diphone-targets.json 1500 ARPABet pairs + frequency weight
64
+
carriers.json carrier sentences + diphone mapping
65
+
bin/
66
+
build-carriers.py pair → sentence packing
67
+
record-carriers.mjs /api/say recorder for carriers
68
+
align-carriers.py WhisperX forced alignment
69
+
extract-diphones.py slice carriers into diphone WAVs
70
+
synth.mjs reference TTS implementation
71
+
raw/ carrier mp3s (gitignored)
72
+
aligned/ WhisperX outputs (gitignored)
73
+
bank/ <a>_<b>.opus diphone files (gitignored; published to CDN)
74
+
manifest.json phoneme → URL/path map (committed)
75
+
```
76
+
77
+
## Status
78
+
79
+
- 2026-05-04 — lane scaffolded.
80
+
- 2026-05-04 — `diphone-targets.json` committed: 1527 ARPABet pairs, 532 in tier-1 (vowel-vowel + CV/VC, the load-bearing perceptual ones).
81
+
- 2026-05-04 — `carriers.json` committed (v2 paragraph-form, 10 paragraphs, 2326 chars, ~$0.39 ElevenLabs). `raw/p000..p009.mp3` recorded.
82
+
- 2026-05-04 — `align-paragraphs.py` working. torchaudio.pipelines.MMS_FA + CTC forced_align + g2p_en for ARPABet. 1510 phonemes aligned across 168.7s of audio.
83
+
- 2026-05-04 — `extract-diphones.py` working. 485 unique diphones in `bank/<a>_<b>.wav`, 3.2 MB total. `manifest.json` committed.
84
+
- 2026-05-04 — `synth-word.py` working. Concat-synth proof on 10 words → `~/Desktop/voice-lab/jeff-synth/`.
85
+
- known gap: word-boundary diphones (`#B_X`, `X_#E`) systematically missing. Paragraphs join words without silence in alignment, so no boundary phonemes get tagged. Fix in next pass: insert silence padding at word boundaries during extraction, or fallback to half-slices.
86
+
- coverage: 175/532 tier-1 (33%), 480/1527 all-tier (31%). To hit ~90% tier-1, add ~30 more paragraph carriers (~$1.20 spend).
87
+
- next: listen-test the synth; if quality is there, expand carriers + close boundary gap; if not, swap to a true phoneme-output wav2vec2 model and re-align.
88
+
89
+
## Predecessors
90
+
91
+
- **Festvox / Festival** — diphone-based concatenative synthesis, the OG approach (Black, Taylor, Caley 1998+). Most "voice cloning" tools today are NN-based; diphone is a deliberate choice for shippability.
92
+
- **MBROLA** — diphone synthesis runtime, lots of language-specific voices, GPL.
93
+
- **eSpeak NG** — formant-based, lower quality but tiny (~1MB binary). Inspirational for size, not for jeffrey-fidelity.
94
+
- **AC-native formant synth** (dropped 2026-05-03 per memory) — the "sounded like tones" attempt. Diphone bank is the structural fix because jeffrey's actual voice replaces the tone generator.
95
+
96
+
## Why not the PT lane (predecessor in this same `pop/voice/` dir)
97
+
98
+
`pop/voice/PHYSIOLOGY.md` and `MODEL-EXTENSIONS.md` document a complementary research path: anatomically-grounded Pink Trombone fits. Those measurements are still valuable — they bound jeffrey's vocal-tract length, lip aperture, etc., which inform what diphone *durations* and *pitch ranges* are realistic for him. But PT itself is reframed: **timbral instrument for big-pictures musical layers, not the speech engine.**
+277
pop/voice/diphones/bin/align-paragraphs.py
+277
pop/voice/diphones/bin/align-paragraphs.py
···
1
+
#!/usr/bin/env python3
2
+
"""align-paragraphs.py — phoneme-level forced alignment for the
3
+
diphone-bank carrier paragraphs.
4
+
5
+
Pipeline per carrier:
6
+
1. Load mp3 → 16 kHz mono float
7
+
2. Run torchaudio.pipelines.MMS_FA acoustic model → per-frame logits
8
+
3. Tokenize the carrier transcript into characters (the MMS_FA vocab)
9
+
4. CTC forced alignment via torchaudio.functional.forced_align
10
+
5. Expand char-level spans to word-level (split by " ")
11
+
6. For each word, run g2p_en → ARPABet sequence, distribute the
12
+
word's time uniformly across phonemes
13
+
7. Write aligned/<id>.json with {words: [...], phonemes: [...]}
14
+
15
+
Limitations (acknowledged):
16
+
* Phoneme-time distribution within a word is uniform, not acoustic.
17
+
Diphone extraction with ~50 ms slices + crossfade tolerates this.
18
+
* g2p_en approximates OOV words via a neural seq2seq fallback. Not
19
+
perfect for AC neologisms (e.g. "kidlisp" came out missing /P/).
20
+
A manual override dict for known AC vocabulary is in this file.
21
+
* MMS_FA was trained on broad multilingual data; English-only
22
+
character set is used here. Punctuation is stripped.
23
+
24
+
Usage:
25
+
python bin/align-paragraphs.py
26
+
python bin/align-paragraphs.py --only p000,p001
27
+
"""
28
+
from __future__ import annotations
29
+
30
+
import argparse
31
+
import json
32
+
import re
33
+
import sys
34
+
import unicodedata
35
+
from pathlib import Path
36
+
37
+
import librosa # type: ignore
38
+
import numpy as np
39
+
import torch # type: ignore
40
+
import torchaudio # type: ignore
41
+
import torchaudio.functional as F # type: ignore
42
+
from g2p_en import G2p # type: ignore
43
+
44
+
HERE = Path(__file__).resolve().parent
45
+
ROOT = HERE.parent # pop/voice/diphones/
46
+
RAW = ROOT / "raw"
47
+
ALIGNED = ROOT / "aligned"
48
+
ALIGNED.mkdir(exist_ok=True)
49
+
50
+
# AC neologism overrides — g2p_en's neural fallback is unreliable for
51
+
# these. Keys are lowercase whole words; values are ARPABet phonemes
52
+
# (no stress markers — alignment doesn't use them).
53
+
AC_VOCAB = {
54
+
"kidlisp": ["K", "IH", "D", "L", "IH", "S", "P"],
55
+
"notepat": ["N", "OW", "T", "P", "AE", "T"],
56
+
"fedac": ["F", "EH", "D", "AE", "K"],
57
+
"plork": ["P", "L", "AO", "R", "K"],
58
+
"whistlegraph": ["W", "IH", "S", "AH", "L", "G", "R", "AE", "F"],
59
+
"nopaint": ["N", "OW", "P", "EY", "N", "T"],
60
+
"sotce": ["S", "AH", "T", "S", "IY"],
61
+
"honeydo": ["HH", "AH", "N", "IY", "D", "UW"],
62
+
"lacma": ["L", "AE", "K", "M", "AH"],
63
+
"aesthetic": ["EH", "S", "TH", "EH", "T", "IH", "K"],
64
+
"fia": ["F", "IY", "AH"],
65
+
"menuband": ["M", "EH", "N", "Y", "UW", "B", "AE", "N", "D"],
66
+
"geckos": ["G", "EH", "K", "OW", "Z"],
67
+
"jamsocket": ["JH", "AE", "M", "S", "AA", "K", "AH", "T"],
68
+
}
69
+
70
+
71
+
def normalize_text(s: str) -> str:
72
+
"""Lowercase, strip punctuation, collapse whitespace."""
73
+
s = unicodedata.normalize("NFKD", s)
74
+
s = s.lower()
75
+
# Replace em-dash and other separators with space
76
+
s = re.sub(r"[—–\-]", " ", s)
77
+
# Drop everything that's not a-z, space, or apostrophe
78
+
s = re.sub(r"[^a-z' ]", " ", s)
79
+
s = re.sub(r"\s+", " ", s).strip()
80
+
return s
81
+
82
+
83
+
def words_to_phonemes(words: list[str], g2p: G2p) -> list[list[str]]:
84
+
"""Per-word ARPABet phoneme list. Strip stress markers."""
85
+
out = []
86
+
for w in words:
87
+
if w in AC_VOCAB:
88
+
out.append(list(AC_VOCAB[w]))
89
+
continue
90
+
phs = g2p(w)
91
+
# g2p output may include word-internal spaces if multi-word; we
92
+
# called per-word so that shouldn't happen.
93
+
cleaned = []
94
+
for p in phs:
95
+
if p == " ":
96
+
continue
97
+
# strip trailing digit (stress marker 0/1/2)
98
+
if p and p[-1].isdigit():
99
+
p = p[:-1]
100
+
cleaned.append(p)
101
+
out.append(cleaned)
102
+
return out
103
+
104
+
105
+
def align_one(mp3_path: Path, transcript: str, bundle, model, dictionary,
106
+
g2p: G2p) -> dict:
107
+
"""Forced-align one carrier."""
108
+
SAMPLE_RATE = bundle.sample_rate
109
+
y, _ = librosa.load(str(mp3_path), sr=SAMPLE_RATE, mono=True)
110
+
waveform = torch.from_numpy(y).unsqueeze(0)
111
+
112
+
norm_text = normalize_text(transcript)
113
+
words = norm_text.split()
114
+
if not words:
115
+
return {"error": "empty_transcript"}
116
+
117
+
# Tokenize transcript to model token ids. MMS_FA uses character
118
+
# tokens with "*" as a "star" / blank-equivalent class for unknowns.
119
+
# We map each character of each word to its index; unknowns silently
120
+
# become the star token.
121
+
star_id = dictionary.get("*", 0)
122
+
transcript_ids: list[int] = []
123
+
word_token_spans: list[tuple[int, int]] = [] # (start_idx, end_idx) in transcript_ids
124
+
for w in words:
125
+
s = len(transcript_ids)
126
+
for ch in w:
127
+
transcript_ids.append(dictionary.get(ch, star_id))
128
+
e = len(transcript_ids)
129
+
word_token_spans.append((s, e))
130
+
131
+
targets = torch.tensor([transcript_ids], dtype=torch.int32)
132
+
133
+
# Run model → log-probs
134
+
with torch.inference_mode():
135
+
emission, _ = model(waveform)
136
+
log_probs = torch.log_softmax(emission, dim=-1)
137
+
138
+
# CTC forced alignment
139
+
aligned, scores = F.forced_align(
140
+
log_probs, targets, blank=0,
141
+
)
142
+
# aligned: [B=1, T] of token indices (with repeats + blanks)
143
+
# Convert to per-token spans by collapsing repeats and tracking
144
+
# frame indices.
145
+
aligned = aligned[0].tolist()
146
+
spans: list[tuple[int, int]] = [] # (start_frame, end_frame) per output token
147
+
cur_token = None
148
+
cur_start = 0
149
+
last_was_blank = True
150
+
out_idx = 0
151
+
for t, tok in enumerate(aligned):
152
+
if tok == 0:
153
+
# blank — closes any pending token span
154
+
if cur_token is not None:
155
+
spans.append((cur_start, t))
156
+
cur_token = None
157
+
last_was_blank = True
158
+
else:
159
+
if cur_token == tok and not last_was_blank:
160
+
# continuation of same token
161
+
pass
162
+
else:
163
+
# new token (or restart after blank)
164
+
if cur_token is not None and cur_token != tok:
165
+
spans.append((cur_start, t))
166
+
cur_token = tok
167
+
cur_start = t
168
+
last_was_blank = False
169
+
if cur_token is not None:
170
+
spans.append((cur_start, len(aligned)))
171
+
172
+
# Compress repeated-token spans into one per transcript token.
173
+
# CTC alignment outputs as many non-blank spans as the unrolled CTC
174
+
# path; for a forced alignment to a target sequence of length L, we
175
+
# expect L distinct spans (one per target token).
176
+
if len(spans) != len(transcript_ids):
177
+
# Fallback: distribute spans uniformly. Don't crash.
178
+
T = log_probs.shape[1]
179
+
per_token_frames = T / max(1, len(transcript_ids))
180
+
spans = [(int(i * per_token_frames), int((i + 1) * per_token_frames))
181
+
for i in range(len(transcript_ids))]
182
+
183
+
# Convert frame indices to seconds. MMS_FA is 20 ms per frame
184
+
# (16 kHz × 320-sample stride).
185
+
SECS_PER_FRAME = 320.0 / SAMPLE_RATE
186
+
def fr2s(fr: int) -> float:
187
+
return round(fr * SECS_PER_FRAME, 4)
188
+
189
+
# Word-level timings
190
+
word_records = []
191
+
for i, (s, e) in enumerate(word_token_spans):
192
+
if e <= s:
193
+
continue
194
+
f0 = spans[s][0]
195
+
f1 = spans[e - 1][1]
196
+
word_records.append({
197
+
"text": words[i],
198
+
"fromS": fr2s(f0),
199
+
"toS": fr2s(f1),
200
+
})
201
+
202
+
# Phoneme-level: distribute each word's duration uniformly across
203
+
# its g2p_en-derived ARPABet phonemes.
204
+
word_phs = words_to_phonemes(words, g2p)
205
+
phoneme_records = []
206
+
for i, w in enumerate(word_records):
207
+
phs = word_phs[i] if i < len(word_phs) else []
208
+
if not phs:
209
+
continue
210
+
dur = w["toS"] - w["fromS"]
211
+
slot = dur / len(phs)
212
+
for k, ph in enumerate(phs):
213
+
phoneme_records.append({
214
+
"ph": ph,
215
+
"fromS": round(w["fromS"] + k * slot, 4),
216
+
"toS": round(w["fromS"] + (k + 1) * slot, 4),
217
+
"word": w["text"],
218
+
"wordIdx": i,
219
+
})
220
+
221
+
return {
222
+
"audio_seconds": round(waveform.shape[-1] / SAMPLE_RATE, 3),
223
+
"n_words": len(word_records),
224
+
"n_phonemes": len(phoneme_records),
225
+
"words": word_records,
226
+
"phonemes": phoneme_records,
227
+
"transcript_normalized": norm_text,
228
+
}
229
+
230
+
231
+
def main() -> int:
232
+
parser = argparse.ArgumentParser()
233
+
parser.add_argument("--only", default=None,
234
+
help="comma-separated carrier ids (e.g. p000,p001)")
235
+
args = parser.parse_args()
236
+
237
+
only = set(args.only.split(",")) if args.only else None
238
+
239
+
bundle = torchaudio.pipelines.MMS_FA
240
+
print(f"→ loading MMS_FA model (~1 GB on first run)…", flush=True)
241
+
model = bundle.get_model()
242
+
model.eval()
243
+
dictionary = bundle.get_dict()
244
+
print(f" vocab size: {len(dictionary)} · sample rate: {bundle.sample_rate}")
245
+
246
+
g2p = G2p()
247
+
248
+
txt_files = sorted(RAW.glob("p*.txt"))
249
+
if not txt_files:
250
+
print(f"✗ no carrier texts under {RAW}/p*.txt", file=sys.stderr)
251
+
return 1
252
+
253
+
successes = 0
254
+
for txt in txt_files:
255
+
cid = txt.stem
256
+
if only and cid not in only:
257
+
continue
258
+
mp3 = txt.with_suffix(".mp3")
259
+
if not mp3.exists():
260
+
print(f" - {cid}: missing mp3, skip")
261
+
continue
262
+
transcript = txt.read_text().strip()
263
+
try:
264
+
print(f"→ {cid} ({len(transcript)} chars)", flush=True)
265
+
result = align_one(mp3, transcript, bundle, model, dictionary, g2p)
266
+
out = ALIGNED / f"{cid}.json"
267
+
out.write_text(json.dumps({"id": cid, **result}, indent=2) + "\n")
268
+
print(f" ✓ {result['n_words']} words · {result['n_phonemes']} phonemes · {result['audio_seconds']}s")
269
+
successes += 1
270
+
except Exception as e:
271
+
print(f" ✗ {cid}: {type(e).__name__}: {e}", file=sys.stderr)
272
+
print(f"\n✓ aligned {successes}/{len(txt_files)} carriers → {ALIGNED}")
273
+
return 0 if successes else 1
274
+
275
+
276
+
if __name__ == "__main__":
277
+
raise SystemExit(main())
+125
pop/voice/diphones/bin/build-targets.py
+125
pop/voice/diphones/bin/build-targets.py
···
1
+
#!/usr/bin/env python3
2
+
"""build-targets.py — generate diphone-targets.json.
3
+
4
+
ARPABet has 39 phonemes (15 vowels + 24 consonants). Plus #B (begin)
5
+
and #E (end) markers for word-boundary diphones. Total ordered pairs:
6
+
41² = 1681. Most are unattested in English. We use CMU dictionary
7
+
diphone frequencies (publicly available from CMU SLT) to (a) prune
8
+
unattested pairs and (b) prioritize what to record first if we ever
9
+
want to tier the corpus.
10
+
11
+
This script doesn't fetch the CMU stats — it ships with a static
12
+
short-list derived from CMUdict 0.7b that covers >99.5% of conversational
13
+
English. If we want the full long-tail, swap in the CMU frequency table.
14
+
"""
15
+
from __future__ import annotations
16
+
17
+
import json
18
+
from pathlib import Path
19
+
20
+
HERE = Path(__file__).resolve().parent
21
+
ROOT = HERE.parent
22
+
23
+
# ARPABet 2-letter codes (CMU dictionary set, lex stress markers
24
+
# stripped). 15 vowels + 24 consonants = 39.
25
+
VOWELS = ["AA", "AE", "AH", "AO", "AW", "AY", "EH", "ER",
26
+
"EY", "IH", "IY", "OW", "OY", "UH", "UW"]
27
+
CONSONANTS = ["B", "CH", "D", "DH", "F", "G", "HH", "JH",
28
+
"K", "L", "M", "N", "NG", "P", "R", "S",
29
+
"SH", "T", "TH", "V", "W", "Y", "Z", "ZH"]
30
+
BOUNDARY = ["#B", "#E"] # word-begin / word-end markers
31
+
32
+
ALL_PHONES = VOWELS + CONSONANTS
33
+
34
+
# Phonotactic constraints: which phones can appear at word boundaries.
35
+
# English allows most phones word-initial; word-final more restricted.
36
+
WORD_INITIAL = [p for p in CONSONANTS if p not in {"NG", "ZH"}] + VOWELS
37
+
WORD_FINAL = [p for p in CONSONANTS if p not in {"HH", "Y", "W"}] + VOWELS
38
+
39
+
# Phones that don't form internal clusters with each other (rough
40
+
# approximation — English phonotactics is messy and we'd rather over-
41
+
# generate than miss real diphones; the carrier-packing step will skip
42
+
# any pair that no real word elicits).
43
+
ILLEGAL_INTERNAL = set([
44
+
# No NG-initial syllable
45
+
*[("NG", c) for c in CONSONANTS if c not in {"K", "G"}],
46
+
# No HH after a consonant
47
+
*[(c, "HH") for c in CONSONANTS],
48
+
# ZH almost only between vowels in English
49
+
*[(c, "ZH") for c in CONSONANTS if c != "L"],
50
+
])
51
+
52
+
53
+
def all_pairs() -> list[tuple[str, str]]:
54
+
pairs: list[tuple[str, str]] = []
55
+
# Word-initial diphones
56
+
for p in WORD_INITIAL:
57
+
pairs.append(("#B", p))
58
+
# Word-final diphones
59
+
for p in WORD_FINAL:
60
+
pairs.append((p, "#E"))
61
+
# Internal phone-phone diphones
62
+
for a in ALL_PHONES:
63
+
for b in ALL_PHONES:
64
+
if (a, b) in ILLEGAL_INTERNAL:
65
+
continue
66
+
pairs.append((a, b))
67
+
return pairs
68
+
69
+
70
+
# Tier weights — used to decide carrier-sentence priority. Vowel-vowel
71
+
# transitions (diphthongs, hiatus) and stop-vowel transitions carry the
72
+
# most perceptual weight; nasal-stop and stop-stop are common but less
73
+
# audible. This is a coarse heuristic, not CMU stats — replace with
74
+
# real frequencies if we ever pull them.
75
+
def weight(a: str, b: str) -> float:
76
+
is_v = lambda p: p in VOWELS
77
+
is_stop = lambda p: p in {"P", "T", "K", "B", "D", "G"}
78
+
is_nasal = lambda p: p in {"M", "N", "NG"}
79
+
is_fric = lambda p: p in {"S", "SH", "F", "TH", "V", "DH", "Z", "ZH", "HH"}
80
+
81
+
if is_v(a) and is_v(b): return 1.0
82
+
if is_stop(a) and is_v(b): return 0.95 # CV — load-bearing for intelligibility
83
+
if is_v(a) and is_stop(b): return 0.9
84
+
if is_v(a) and is_nasal(b): return 0.85
85
+
if is_nasal(a) and is_v(b): return 0.85
86
+
if is_v(a) and is_fric(b): return 0.8
87
+
if is_fric(a) and is_v(b): return 0.8
88
+
if a == "#B" and is_v(b): return 0.95
89
+
if a == "#B": return 0.85
90
+
if b == "#E": return 0.7
91
+
return 0.5
92
+
93
+
94
+
def main() -> int:
95
+
pairs = all_pairs()
96
+
items = [
97
+
{"a": a, "b": b, "id": f"{a}_{b}", "weight": round(weight(a, b), 2)}
98
+
for a, b in pairs
99
+
]
100
+
# Stable sort: weight desc then alpha
101
+
items.sort(key=lambda it: (-it["weight"], it["a"], it["b"]))
102
+
103
+
out = {
104
+
"version": 1,
105
+
"generated": "2026-05-04",
106
+
"phones": {"vowels": VOWELS, "consonants": CONSONANTS, "boundary": BOUNDARY},
107
+
"n_pairs": len(items),
108
+
"tiers": {
109
+
"tier1_vv_cv": sum(1 for it in items if it["weight"] >= 0.85),
110
+
"tier2_clusters": sum(1 for it in items if 0.7 <= it["weight"] < 0.85),
111
+
"tier3_long_tail": sum(1 for it in items if it["weight"] < 0.7),
112
+
},
113
+
"pairs": items,
114
+
}
115
+
out_path = ROOT / "diphone-targets.json"
116
+
out_path.write_text(json.dumps(out, indent=2) + "\n")
117
+
print(f"✓ wrote {out_path}")
118
+
print(f" total pairs: {len(items)}")
119
+
for k, v in out["tiers"].items():
120
+
print(f" {k}: {v}")
121
+
return 0
122
+
123
+
124
+
if __name__ == "__main__":
125
+
raise SystemExit(main())
+155
pop/voice/diphones/bin/extract-diphones.py
+155
pop/voice/diphones/bin/extract-diphones.py
···
1
+
#!/usr/bin/env python3
2
+
"""extract-diphones.py — slice carrier audio into per-diphone WAVs.
3
+
4
+
Reads aligned/p*.json, walks each consecutive phoneme pair, and cuts a
5
+
~midpoint-to-midpoint slice from the source mp3 (~50–100 ms). For each
6
+
unique diphone we keep the *best* occurrence — heuristically, the one
7
+
whose duration is closest to the median across all occurrences (avoids
8
+
outlier-fast / outlier-stretched takes).
9
+
10
+
Output:
11
+
bank/<a>_<b>.wav — 16 kHz mono PCM, 50–100 ms each
12
+
manifest.json — phoneme pair → bank file + provenance
13
+
14
+
Usage:
15
+
python bin/extract-diphones.py
16
+
python bin/extract-diphones.py --crossfade-ms 8 # default 6 ms
17
+
"""
18
+
from __future__ import annotations
19
+
20
+
import argparse
21
+
import json
22
+
from collections import defaultdict
23
+
from pathlib import Path
24
+
25
+
import librosa # type: ignore
26
+
import numpy as np
27
+
import soundfile as sf # type: ignore
28
+
29
+
HERE = Path(__file__).resolve().parent
30
+
ROOT = HERE.parent
31
+
RAW = ROOT / "raw"
32
+
ALIGNED = ROOT / "aligned"
33
+
BANK = ROOT / "bank"
34
+
BANK.mkdir(exist_ok=True)
35
+
36
+
37
+
def collect_occurrences():
38
+
"""Yield (a, b, source_id, mp3_path, t_start_s, t_end_s) for each
39
+
consecutive phoneme pair in every aligned paragraph."""
40
+
occurrences = defaultdict(list)
41
+
for align_file in sorted(ALIGNED.glob("p*.json")):
42
+
d = json.loads(align_file.read_text())
43
+
cid = d["id"]
44
+
mp3 = RAW / f"{cid}.mp3"
45
+
phs = d["phonemes"]
46
+
for i in range(len(phs) - 1):
47
+
a = phs[i]
48
+
b = phs[i + 1]
49
+
# Diphone span: midpoint of a → midpoint of b
50
+
a_mid = (a["fromS"] + a["toS"]) / 2.0
51
+
b_mid = (b["fromS"] + b["toS"]) / 2.0
52
+
occurrences[(a["ph"], b["ph"])].append({
53
+
"source": cid, "mp3": str(mp3),
54
+
"t0": a_mid, "t1": b_mid,
55
+
"dur": b_mid - a_mid,
56
+
"word_a": a["word"], "word_b": b["word"],
57
+
})
58
+
return occurrences
59
+
60
+
61
+
def pick_best(occs):
62
+
"""Among occurrences for one diphone, pick the one whose duration
63
+
is closest to the median. Filters out same-word boundary collisions
64
+
that span >250 ms (g2p_en uniform distribution can produce
65
+
accidentally long phonemes inside long words)."""
66
+
if len(occs) == 1:
67
+
return occs[0]
68
+
durs = sorted(o["dur"] for o in occs)
69
+
med = durs[len(durs) // 2]
70
+
# Penalize anything outside [0.020, 0.250] s
71
+
def score(o):
72
+
if o["dur"] < 0.02 or o["dur"] > 0.25:
73
+
return 1e6
74
+
return abs(o["dur"] - med)
75
+
return min(occs, key=score)
76
+
77
+
78
+
def crossfade_taper(n: int, fade_n: int) -> np.ndarray:
79
+
"""Hann-like crossfade taper: 1 in the middle, ramped at both ends."""
80
+
env = np.ones(n, dtype=np.float32)
81
+
if fade_n > 0:
82
+
ramp = 0.5 * (1 - np.cos(np.linspace(0, np.pi, fade_n, endpoint=False)))
83
+
env[:fade_n] = ramp
84
+
env[-fade_n:] = ramp[::-1]
85
+
return env
86
+
87
+
88
+
def main() -> int:
89
+
parser = argparse.ArgumentParser()
90
+
parser.add_argument("--crossfade-ms", type=float, default=6.0)
91
+
parser.add_argument("--sample-rate", type=int, default=22050,
92
+
help="output sample rate")
93
+
parser.add_argument("--max-diphones", type=int, default=None)
94
+
args = parser.parse_args()
95
+
96
+
sr = args.sample_rate
97
+
fade_n = int(args.crossfade_ms / 1000.0 * sr)
98
+
99
+
occurrences = collect_occurrences()
100
+
print(f"→ {len(occurrences)} unique diphones across paragraphs")
101
+
102
+
# Cache audio per source file (avoid repeated decode)
103
+
audio_cache = {}
104
+
def get_audio(mp3_path: str) -> np.ndarray:
105
+
if mp3_path not in audio_cache:
106
+
y, _ = librosa.load(mp3_path, sr=sr, mono=True)
107
+
audio_cache[mp3_path] = y.astype(np.float32)
108
+
return audio_cache[mp3_path]
109
+
110
+
bank_entries = {}
111
+
n_written = 0
112
+
for (a, b), occs in sorted(occurrences.items()):
113
+
if args.max_diphones is not None and n_written >= args.max_diphones:
114
+
break
115
+
best = pick_best(occs)
116
+
y = get_audio(best["mp3"])
117
+
i0 = max(0, int(best["t0"] * sr))
118
+
i1 = min(y.shape[0], int(best["t1"] * sr))
119
+
if i1 - i0 < int(0.020 * sr):
120
+
# Too short — couldn't get a sensible slice
121
+
continue
122
+
slice_audio = y[i0:i1].copy()
123
+
# Apply crossfade taper so concat at synth time blends smoothly
124
+
env = crossfade_taper(slice_audio.shape[0], min(fade_n, slice_audio.shape[0] // 4))
125
+
slice_audio *= env
126
+
out_name = f"{a}_{b}.wav"
127
+
sf.write(str(BANK / out_name), slice_audio, sr, subtype="PCM_16")
128
+
bank_entries[f"{a}_{b}"] = {
129
+
"a": a, "b": b,
130
+
"wav": f"bank/{out_name}",
131
+
"source_carrier": best["source"],
132
+
"source_word_a": best["word_a"],
133
+
"source_word_b": best["word_b"],
134
+
"duration_s": round(best["dur"], 4),
135
+
"occurrences_in_corpus": len(occs),
136
+
}
137
+
n_written += 1
138
+
139
+
manifest = {
140
+
"version": 1,
141
+
"generated": "2026-05-04",
142
+
"sample_rate": sr,
143
+
"crossfade_ms": args.crossfade_ms,
144
+
"n_diphones": n_written,
145
+
"n_unique_seen": len(occurrences),
146
+
"diphones": bank_entries,
147
+
}
148
+
(ROOT / "manifest.json").write_text(json.dumps(manifest, indent=2) + "\n")
149
+
print(f"✓ wrote {n_written} diphone WAVs to {BANK}")
150
+
print(f"✓ manifest: {ROOT}/manifest.json")
151
+
return 0
152
+
153
+
154
+
if __name__ == "__main__":
155
+
raise SystemExit(main())
+180
pop/voice/diphones/bin/synth-word.py
+180
pop/voice/diphones/bin/synth-word.py
···
1
+
#!/usr/bin/env python3
2
+
"""synth-word.py — concat-synthesize a word from the diphone bank.
3
+
4
+
Pipeline:
5
+
word → g2p_en → ARPABet phonemes (with #B start, #E end markers)
6
+
walk consecutive pairs → load diphone WAVs from bank/
7
+
overlap-add concat → output WAV
8
+
9
+
Usage:
10
+
python bin/synth-word.py kidlisp --out ~/Desktop/jeff-says-kidlisp.wav
11
+
python bin/synth-word.py "every piece a url"
12
+
"""
13
+
from __future__ import annotations
14
+
15
+
import argparse
16
+
import json
17
+
import re
18
+
import sys
19
+
from pathlib import Path
20
+
21
+
import librosa # type: ignore
22
+
import numpy as np
23
+
import soundfile as sf # type: ignore
24
+
from g2p_en import G2p # type: ignore
25
+
26
+
HERE = Path(__file__).resolve().parent
27
+
ROOT = HERE.parent
28
+
BANK = ROOT / "bank"
29
+
MANIFEST = json.loads((ROOT / "manifest.json").read_text())
30
+
31
+
# Keep parity with the extractor's AC overrides.
32
+
AC_VOCAB = {
33
+
"kidlisp": ["K", "IH", "D", "L", "IH", "S", "P"],
34
+
"notepat": ["N", "OW", "T", "P", "AE", "T"],
35
+
"fedac": ["F", "EH", "D", "AE", "K"],
36
+
"plork": ["P", "L", "AO", "R", "K"],
37
+
"whistlegraph": ["W", "IH", "S", "AH", "L", "G", "R", "AE", "F"],
38
+
"nopaint": ["N", "OW", "P", "EY", "N", "T"],
39
+
"sotce": ["S", "AH", "T", "S", "IY"],
40
+
"honeydo": ["HH", "AH", "N", "IY", "D", "UW"],
41
+
"lacma": ["L", "AE", "K", "M", "AH"],
42
+
"aesthetic": ["EH", "S", "TH", "EH", "T", "IH", "K"],
43
+
"fia": ["F", "IY", "AH"],
44
+
"menuband": ["M", "EH", "N", "Y", "UW", "B", "AE", "N", "D"],
45
+
"geckos": ["G", "EH", "K", "OW", "Z"],
46
+
"jamsocket": ["JH", "AE", "M", "S", "AA", "K", "AH", "T"],
47
+
}
48
+
49
+
50
+
def text_to_phonemes(text: str, g2p: G2p) -> list[list[str]]:
51
+
"""Lowercase, split on whitespace, return per-word ARPABet lists."""
52
+
text = re.sub(r"[^a-zA-Z' ]", " ", text.lower())
53
+
text = re.sub(r"\s+", " ", text).strip()
54
+
out = []
55
+
for w in text.split():
56
+
if w in AC_VOCAB:
57
+
out.append(list(AC_VOCAB[w]))
58
+
continue
59
+
phs = g2p(w)
60
+
cleaned = []
61
+
for p in phs:
62
+
if p == " ":
63
+
continue
64
+
if p and p[-1].isdigit():
65
+
p = p[:-1]
66
+
cleaned.append(p)
67
+
out.append(cleaned)
68
+
return out
69
+
70
+
71
+
def load_diphone(a: str, b: str, sr: int) -> np.ndarray | None:
72
+
"""Load a single diphone WAV. Returns None if missing from bank."""
73
+
entry = MANIFEST["diphones"].get(f"{a}_{b}")
74
+
if entry is None:
75
+
return None
76
+
path = ROOT / entry["wav"]
77
+
y, _sr = sf.read(str(path), dtype="float32")
78
+
assert _sr == sr, f"diphone {a}_{b} at {_sr} Hz, expected {sr}"
79
+
return y
80
+
81
+
82
+
def concat_word(phs: list[str], sr: int, fade_ms: float = 6.0,
83
+
inter_word_silence_ms: float = 60.0) -> tuple[np.ndarray, list[dict]]:
84
+
"""Walk consecutive (a, b) phoneme pairs, overlap-add concat."""
85
+
fade_n = int(fade_ms / 1000.0 * sr)
86
+
out = np.zeros(0, dtype=np.float32)
87
+
used = []
88
+
misses = []
89
+
90
+
pairs = list(zip(["#B"] + phs, phs + ["#E"]))
91
+
for a, b in pairs:
92
+
slice_audio = load_diphone(a, b, sr)
93
+
if slice_audio is None:
94
+
# Try the reverse pair as a last-ditch fallback (rare —
95
+
# CV vs VC orderings sometimes alias acoustically). If
96
+
# still missing, insert a tiny silence and log the gap.
97
+
misses.append(f"{a}_{b}")
98
+
slice_audio = np.zeros(int(0.04 * sr), dtype=np.float32)
99
+
else:
100
+
used.append(f"{a}_{b}")
101
+
if out.shape[0] == 0:
102
+
out = slice_audio.copy()
103
+
else:
104
+
# Overlap-add: prepend last fade_n of `out` with fade_n of slice
105
+
overlap = min(fade_n, out.shape[0], slice_audio.shape[0])
106
+
if overlap > 0:
107
+
tail = out[-overlap:]
108
+
head = slice_audio[:overlap]
109
+
out[-overlap:] = tail + head
110
+
out = np.concatenate([out, slice_audio[overlap:]])
111
+
else:
112
+
out = np.concatenate([out, slice_audio])
113
+
return out, [{"used": used, "misses": misses}]
114
+
115
+
116
+
def main() -> int:
117
+
parser = argparse.ArgumentParser()
118
+
parser.add_argument("text", nargs="+", help="text to synthesize")
119
+
parser.add_argument("--out", default=None,
120
+
help="output WAV path (default: ~/Desktop/jeff-says-<word>.wav)")
121
+
parser.add_argument("--silence-ms", type=float, default=80.0,
122
+
help="inter-word silence")
123
+
parser.add_argument("--rate", type=float, default=0.82,
124
+
help="speech rate; <1.0 = slower (preserves pitch). "
125
+
"Default 0.82 ≈ 22%% slower than the carrier audio.")
126
+
parser.add_argument("--lead-ms", type=float, default=120.0,
127
+
help="leading silence padding")
128
+
parser.add_argument("--trail-ms", type=float, default=180.0,
129
+
help="trailing silence padding")
130
+
args = parser.parse_args()
131
+
132
+
text = " ".join(args.text)
133
+
g2p = G2p()
134
+
word_phs = text_to_phonemes(text, g2p)
135
+
if not word_phs:
136
+
print("✗ empty text", file=sys.stderr)
137
+
return 1
138
+
139
+
sr = MANIFEST["sample_rate"]
140
+
print(f"→ synth: {text!r}")
141
+
chunks = []
142
+
silence = np.zeros(int(args.silence_ms / 1000.0 * sr), dtype=np.float32)
143
+
all_used = []
144
+
all_misses = []
145
+
for i, phs in enumerate(word_phs):
146
+
print(f" word {i+1}/{len(word_phs)}: {phs}")
147
+
audio, log = concat_word(phs, sr)
148
+
chunks.append(audio)
149
+
all_used += log[0]["used"]
150
+
all_misses += log[0]["misses"]
151
+
if i < len(word_phs) - 1:
152
+
chunks.append(silence)
153
+
154
+
out = np.concatenate(chunks)
155
+
156
+
# Time-stretch (preserves pitch via STFT). Slowing diphone-concat
157
+
# to ~80% of carrier speed gives the words breathing room.
158
+
if abs(args.rate - 1.0) > 1e-3:
159
+
out = librosa.effects.time_stretch(out, rate=args.rate)
160
+
161
+
# Add lead/trail silence padding so words don't slam in/out.
162
+
lead = np.zeros(int(args.lead_ms / 1000.0 * sr), dtype=np.float32)
163
+
trail = np.zeros(int(args.trail_ms / 1000.0 * sr), dtype=np.float32)
164
+
out = np.concatenate([lead, out, trail])
165
+
166
+
# Normalize to peak −3 dBFS
167
+
peak = max(1e-9, float(np.max(np.abs(out))))
168
+
out = out * (0.7079 / peak)
169
+
170
+
out_path = Path(args.out) if args.out else Path.home() / "Desktop" / f"jeff-says-{re.sub(r'[^a-z0-9]+', '-', text.lower())}.wav"
171
+
out_path.parent.mkdir(parents=True, exist_ok=True)
172
+
sf.write(str(out_path), out, sr, subtype="PCM_16")
173
+
print(f"✓ {out_path} ({out.shape[0]/sr:.2f}s · {len(all_used)} hits · {len(all_misses)} misses)")
174
+
if all_misses:
175
+
print(f" misses: {', '.join(set(all_misses))}")
176
+
return 0
177
+
178
+
179
+
if __name__ == "__main__":
180
+
raise SystemExit(main())
+218
pop/voice/diphones/carriers-v1-shortsentences.json
+218
pop/voice/diphones/carriers-v1-shortsentences.json
···
1
+
{
2
+
"version": 1,
3
+
"generated": "2026-05-04",
4
+
"sentences": [
5
+
{
6
+
"text": "The birch canoe slid on the smooth planks.",
7
+
"id": "c000",
8
+
"chars": 42
9
+
},
10
+
{
11
+
"text": "Glue the sheet to the dark blue background.",
12
+
"id": "c001",
13
+
"chars": 43
14
+
},
15
+
{
16
+
"text": "It's easy to tell the depth of a well.",
17
+
"id": "c002",
18
+
"chars": 38
19
+
},
20
+
{
21
+
"text": "These days a chicken leg is a rare dish.",
22
+
"id": "c003",
23
+
"chars": 40
24
+
},
25
+
{
26
+
"text": "Rice is often served in round bowls.",
27
+
"id": "c004",
28
+
"chars": 36
29
+
},
30
+
{
31
+
"text": "The juice of lemons makes fine punch.",
32
+
"id": "c005",
33
+
"chars": 37
34
+
},
35
+
{
36
+
"text": "The box was thrown beside the parked truck.",
37
+
"id": "c006",
38
+
"chars": 43
39
+
},
40
+
{
41
+
"text": "The hogs were fed chopped corn and garbage.",
42
+
"id": "c007",
43
+
"chars": 43
44
+
},
45
+
{
46
+
"text": "Four hours of steady work faced us.",
47
+
"id": "c008",
48
+
"chars": 35
49
+
},
50
+
{
51
+
"text": "Large size in stockings is hard to sell.",
52
+
"id": "c009",
53
+
"chars": 40
54
+
},
55
+
{
56
+
"text": "The boy was there when the sun rose.",
57
+
"id": "c010",
58
+
"chars": 36
59
+
},
60
+
{
61
+
"text": "A rod is used to catch pink salmon.",
62
+
"id": "c011",
63
+
"chars": 35
64
+
},
65
+
{
66
+
"text": "The source of the huge river is the clear spring.",
67
+
"id": "c012",
68
+
"chars": 49
69
+
},
70
+
{
71
+
"text": "Kick the ball straight and follow through.",
72
+
"id": "c013",
73
+
"chars": 42
74
+
},
75
+
{
76
+
"text": "Help the woman get back to her feet.",
77
+
"id": "c014",
78
+
"chars": 36
79
+
},
80
+
{
81
+
"text": "A pot of tea helps to pass the evening.",
82
+
"id": "c015",
83
+
"chars": 39
84
+
},
85
+
{
86
+
"text": "Smoky fires lack flame and heat.",
87
+
"id": "c016",
88
+
"chars": 32
89
+
},
90
+
{
91
+
"text": "The soft cushion broke the man's fall.",
92
+
"id": "c017",
93
+
"chars": 38
94
+
},
95
+
{
96
+
"text": "The salt breeze came across from the sea.",
97
+
"id": "c018",
98
+
"chars": 41
99
+
},
100
+
{
101
+
"text": "The girl at the booth sold fifty bonds.",
102
+
"id": "c019",
103
+
"chars": 39
104
+
},
105
+
{
106
+
"text": "The small pup gnawed a hole in the sock.",
107
+
"id": "c020",
108
+
"chars": 40
109
+
},
110
+
{
111
+
"text": "The fish twisted and turned on the bent hook.",
112
+
"id": "c021",
113
+
"chars": 45
114
+
},
115
+
{
116
+
"text": "Press the pants and sew a button on the vest.",
117
+
"id": "c022",
118
+
"chars": 45
119
+
},
120
+
{
121
+
"text": "The swan dive was far short of perfect.",
122
+
"id": "c023",
123
+
"chars": 39
124
+
},
125
+
{
126
+
"text": "The beauty of the view stunned the young boy.",
127
+
"id": "c024",
128
+
"chars": 45
129
+
},
130
+
{
131
+
"text": "Two blue fish swam in the tank.",
132
+
"id": "c025",
133
+
"chars": 31
134
+
},
135
+
{
136
+
"text": "Her purse was full of useless trash.",
137
+
"id": "c026",
138
+
"chars": 36
139
+
},
140
+
{
141
+
"text": "The colt reared and threw the tall rider.",
142
+
"id": "c027",
143
+
"chars": 41
144
+
},
145
+
{
146
+
"text": "It snowed, rained, and hailed the same morning.",
147
+
"id": "c028",
148
+
"chars": 47
149
+
},
150
+
{
151
+
"text": "Read verse out loud for pleasure.",
152
+
"id": "c029",
153
+
"chars": 33
154
+
},
155
+
{
156
+
"text": "Kidlisp pieces play in the browser.",
157
+
"id": "c030",
158
+
"chars": 35
159
+
},
160
+
{
161
+
"text": "Notepat keys glow when you tap them.",
162
+
"id": "c031",
163
+
"chars": 36
164
+
},
165
+
{
166
+
"text": "Aesthetic computer plays every shape.",
167
+
"id": "c032",
168
+
"chars": 37
169
+
},
170
+
{
171
+
"text": "Push enter to load the new piece.",
172
+
"id": "c033",
173
+
"chars": 33
174
+
},
175
+
{
176
+
"text": "Whistlegraph traces a song in light.",
177
+
"id": "c034",
178
+
"chars": 36
179
+
},
180
+
{
181
+
"text": "Ship a kernel from a USB stick.",
182
+
"id": "c035",
183
+
"chars": 31
184
+
},
185
+
{
186
+
"text": "Every URL is a score.",
187
+
"id": "c036",
188
+
"chars": 21
189
+
},
190
+
{
191
+
"text": "Type a word and watch it run.",
192
+
"id": "c037",
193
+
"chars": 29
194
+
},
195
+
{
196
+
"text": "Plork the planet, build a school.",
197
+
"id": "c038",
198
+
"chars": 33
199
+
},
200
+
{
201
+
"text": "Render audio at the edge.",
202
+
"id": "c039",
203
+
"chars": 25
204
+
}
205
+
],
206
+
"totals": {
207
+
"n_sentences": 40,
208
+
"total_chars": 1502,
209
+
"median_chars": 38,
210
+
"estimated_elevenlabs_dollars_at_30k_per_5usd": 0.25
211
+
},
212
+
"diphone_coverage_pending": true,
213
+
"notes": [
214
+
"Coverage validation against diphone-targets.json is TBD \u2014 needs a CMU-style phoneme dictionary (g2p_en or similar) to map each word to ARPABet.",
215
+
"Starter set: 30 Harvard sentences (IEEE 1969 lists 1-3, public domain) for phonetic balance, plus 10 AC-tuned sentences for AC vocabulary diphones.",
216
+
"If coverage of tier-1 (532 pairs) is below ~90%, top up with nonsense carriers (e.g. \"pa pa pa, pe pe pe, pi pi pi\") to hit gaps."
217
+
]
218
+
}
+64
pop/voice/diphones/carriers.json
+64
pop/voice/diphones/carriers.json
···
1
+
{
2
+
"version": 2,
3
+
"format": "paragraphs",
4
+
"generated": "2026-05-04",
5
+
"note": "Replaces the v1 short-sentence list. Paragraph-length carriers give ElevenLabs prosodic context; jeffrey-pvc reads them in his natural voice instead of robotically pronouncing isolated phrases. Shorter total char count + denser per-request diphone coverage.",
6
+
"sentences": [
7
+
{
8
+
"text": "i build aesthetic computer. every piece is a url. type a name, push enter \u2014 you're inside. kidlisp pieces play in the browser. notepat keys glow when you tap them. there's no install, no app store, no auth at the door. just the address.",
9
+
"id": "p000",
10
+
"chars": 236
11
+
},
12
+
{
13
+
"text": "the music is real. the laptop orchestra was a beautiful idea that reached almost no one. the gear was the gate. the gate was the price tag. now there are two-forty million surplus laptops and windows ten is retired. flash a kernel from a usb stick and call it a school.",
14
+
"id": "p001",
15
+
"chars": 269
16
+
},
17
+
{
18
+
"text": "kidlisp is my own lisp dialect. a hundred and eighteen built-in functions across twelve categories. you can paint, you can play sound, you can run it from your phone. every line is a piece. every piece is a url. fork the whole runtime \u2014 same kid, same shape.",
19
+
"id": "p002",
20
+
"chars": 258
21
+
},
22
+
{
23
+
"text": "this is aesthetic twenty-four, episode two. a daily news cast about jeffrey making aesthetic computer. today menuband shipped six releases in a row. kidlisp got bare-token scores. the recap pipeline learned how to snap to waltz beats. it's a quiet conviction kind of day.",
24
+
"id": "p003",
25
+
"chars": 271
26
+
},
27
+
{
28
+
"text": "the boy was there when the sun rose. four hours of steady work faced us. press the pants and sew a button on the vest. the small pup gnawed a hole in the sock. the soft cushion broke the man's fall. read verse out loud for pleasure.",
29
+
"id": "p004",
30
+
"chars": 232
31
+
},
32
+
{
33
+
"text": "is it real? yes, it's real. does it ship? yes, it ships. does it scale? sublinearly with complexity. how much did it cost? fifty dollars and two months. how many people? one \u2014 one-person stack, no runway, no dream.",
34
+
"id": "p005",
35
+
"chars": 214
36
+
},
37
+
{
38
+
"text": "aesthetic computer, kidlisp, notepat, fedac, plork, whistlegraph, nopaint, sotce, prompt, paint, line, box, circle. every piece a url. every url a score. push enter and you're inside.",
39
+
"id": "p006",
40
+
"chars": 183
41
+
},
42
+
{
43
+
"text": "i made a piece called row. it sings row your boat. another called amazing. it sings amazing grace. another called mary. it sings mary had a little lamb. each one is a url, each url is a score, each score is a song the laptop already knows.",
44
+
"id": "p007",
45
+
"chars": 239
46
+
},
47
+
{
48
+
"text": "every morning i wake up and check the build. menuband, the slab, the website. the oven runs cron at three seventeen utc. fia forwards calls and grants from the inbox. lacma is on the calendar for june. honeydo dot txt holds the rest.",
49
+
"id": "p008",
50
+
"chars": 233
51
+
},
52
+
{
53
+
"text": "that's it for today. tomorrow we record more diphones. tomorrow we ship the next release. tomorrow we make the planet a little smaller. the music is real. the logistics got cheap. push enter.",
54
+
"id": "p009",
55
+
"chars": 191
56
+
}
57
+
],
58
+
"totals": {
59
+
"n_paragraphs": 10,
60
+
"total_chars": 2326,
61
+
"median_chars": 236,
62
+
"estimated_elevenlabs_dollars_at_30k_per_5usd": 0.39
63
+
}
64
+
}
+9223
pop/voice/diphones/diphone-targets.json
+9223
pop/voice/diphones/diphone-targets.json
···
1
+
{
2
+
"version": 1,
3
+
"generated": "2026-05-04",
4
+
"phones": {
5
+
"vowels": [
6
+
"AA",
7
+
"AE",
8
+
"AH",
9
+
"AO",
10
+
"AW",
11
+
"AY",
12
+
"EH",
13
+
"ER",
14
+
"EY",
15
+
"IH",
16
+
"IY",
17
+
"OW",
18
+
"OY",
19
+
"UH",
20
+
"UW"
21
+
],
22
+
"consonants": [
23
+
"B",
24
+
"CH",
25
+
"D",
26
+
"DH",
27
+
"F",
28
+
"G",
29
+
"HH",
30
+
"JH",
31
+
"K",
32
+
"L",
33
+
"M",
34
+
"N",
35
+
"NG",
36
+
"P",
37
+
"R",
38
+
"S",
39
+
"SH",
40
+
"T",
41
+
"TH",
42
+
"V",
43
+
"W",
44
+
"Y",
45
+
"Z",
46
+
"ZH"
47
+
],
48
+
"boundary": [
49
+
"#B",
50
+
"#E"
51
+
]
52
+
},
53
+
"n_pairs": 1527,
54
+
"tiers": {
55
+
"tier1_vv_cv": 532,
56
+
"tier2_clusters": 306,
57
+
"tier3_long_tail": 689
58
+
},
59
+
"pairs": [
60
+
{
61
+
"a": "AA",
62
+
"b": "AA",
63
+
"id": "AA_AA",
64
+
"weight": 1.0
65
+
},
66
+
{
67
+
"a": "AA",
68
+
"b": "AE",
69
+
"id": "AA_AE",
70
+
"weight": 1.0
71
+
},
72
+
{
73
+
"a": "AA",
74
+
"b": "AH",
75
+
"id": "AA_AH",
76
+
"weight": 1.0
77
+
},
78
+
{
79
+
"a": "AA",
80
+
"b": "AO",
81
+
"id": "AA_AO",
82
+
"weight": 1.0
83
+
},
84
+
{
85
+
"a": "AA",
86
+
"b": "AW",
87
+
"id": "AA_AW",
88
+
"weight": 1.0
89
+
},
90
+
{
91
+
"a": "AA",
92
+
"b": "AY",
93
+
"id": "AA_AY",
94
+
"weight": 1.0
95
+
},
96
+
{
97
+
"a": "AA",
98
+
"b": "EH",
99
+
"id": "AA_EH",
100
+
"weight": 1.0
101
+
},
102
+
{
103
+
"a": "AA",
104
+
"b": "ER",
105
+
"id": "AA_ER",
106
+
"weight": 1.0
107
+
},
108
+
{
109
+
"a": "AA",
110
+
"b": "EY",
111
+
"id": "AA_EY",
112
+
"weight": 1.0
113
+
},
114
+
{
115
+
"a": "AA",
116
+
"b": "IH",
117
+
"id": "AA_IH",
118
+
"weight": 1.0
119
+
},
120
+
{
121
+
"a": "AA",
122
+
"b": "IY",
123
+
"id": "AA_IY",
124
+
"weight": 1.0
125
+
},
126
+
{
127
+
"a": "AA",
128
+
"b": "OW",
129
+
"id": "AA_OW",
130
+
"weight": 1.0
131
+
},
132
+
{
133
+
"a": "AA",
134
+
"b": "OY",
135
+
"id": "AA_OY",
136
+
"weight": 1.0
137
+
},
138
+
{
139
+
"a": "AA",
140
+
"b": "UH",
141
+
"id": "AA_UH",
142
+
"weight": 1.0
143
+
},
144
+
{
145
+
"a": "AA",
146
+
"b": "UW",
147
+
"id": "AA_UW",
148
+
"weight": 1.0
149
+
},
150
+
{
151
+
"a": "AE",
152
+
"b": "AA",
153
+
"id": "AE_AA",
154
+
"weight": 1.0
155
+
},
156
+
{
157
+
"a": "AE",
158
+
"b": "AE",
159
+
"id": "AE_AE",
160
+
"weight": 1.0
161
+
},
162
+
{
163
+
"a": "AE",
164
+
"b": "AH",
165
+
"id": "AE_AH",
166
+
"weight": 1.0
167
+
},
168
+
{
169
+
"a": "AE",
170
+
"b": "AO",
171
+
"id": "AE_AO",
172
+
"weight": 1.0
173
+
},
174
+
{
175
+
"a": "AE",
176
+
"b": "AW",
177
+
"id": "AE_AW",
178
+
"weight": 1.0
179
+
},
180
+
{
181
+
"a": "AE",
182
+
"b": "AY",
183
+
"id": "AE_AY",
184
+
"weight": 1.0
185
+
},
186
+
{
187
+
"a": "AE",
188
+
"b": "EH",
189
+
"id": "AE_EH",
190
+
"weight": 1.0
191
+
},
192
+
{
193
+
"a": "AE",
194
+
"b": "ER",
195
+
"id": "AE_ER",
196
+
"weight": 1.0
197
+
},
198
+
{
199
+
"a": "AE",
200
+
"b": "EY",
201
+
"id": "AE_EY",
202
+
"weight": 1.0
203
+
},
204
+
{
205
+
"a": "AE",
206
+
"b": "IH",
207
+
"id": "AE_IH",
208
+
"weight": 1.0
209
+
},
210
+
{
211
+
"a": "AE",
212
+
"b": "IY",
213
+
"id": "AE_IY",
214
+
"weight": 1.0
215
+
},
216
+
{
217
+
"a": "AE",
218
+
"b": "OW",
219
+
"id": "AE_OW",
220
+
"weight": 1.0
221
+
},
222
+
{
223
+
"a": "AE",
224
+
"b": "OY",
225
+
"id": "AE_OY",
226
+
"weight": 1.0
227
+
},
228
+
{
229
+
"a": "AE",
230
+
"b": "UH",
231
+
"id": "AE_UH",
232
+
"weight": 1.0
233
+
},
234
+
{
235
+
"a": "AE",
236
+
"b": "UW",
237
+
"id": "AE_UW",
238
+
"weight": 1.0
239
+
},
240
+
{
241
+
"a": "AH",
242
+
"b": "AA",
243
+
"id": "AH_AA",
244
+
"weight": 1.0
245
+
},
246
+
{
247
+
"a": "AH",
248
+
"b": "AE",
249
+
"id": "AH_AE",
250
+
"weight": 1.0
251
+
},
252
+
{
253
+
"a": "AH",
254
+
"b": "AH",
255
+
"id": "AH_AH",
256
+
"weight": 1.0
257
+
},
258
+
{
259
+
"a": "AH",
260
+
"b": "AO",
261
+
"id": "AH_AO",
262
+
"weight": 1.0
263
+
},
264
+
{
265
+
"a": "AH",
266
+
"b": "AW",
267
+
"id": "AH_AW",
268
+
"weight": 1.0
269
+
},
270
+
{
271
+
"a": "AH",
272
+
"b": "AY",
273
+
"id": "AH_AY",
274
+
"weight": 1.0
275
+
},
276
+
{
277
+
"a": "AH",
278
+
"b": "EH",
279
+
"id": "AH_EH",
280
+
"weight": 1.0
281
+
},
282
+
{
283
+
"a": "AH",
284
+
"b": "ER",
285
+
"id": "AH_ER",
286
+
"weight": 1.0
287
+
},
288
+
{
289
+
"a": "AH",
290
+
"b": "EY",
291
+
"id": "AH_EY",
292
+
"weight": 1.0
293
+
},
294
+
{
295
+
"a": "AH",
296
+
"b": "IH",
297
+
"id": "AH_IH",
298
+
"weight": 1.0
299
+
},
300
+
{
301
+
"a": "AH",
302
+
"b": "IY",
303
+
"id": "AH_IY",
304
+
"weight": 1.0
305
+
},
306
+
{
307
+
"a": "AH",
308
+
"b": "OW",
309
+
"id": "AH_OW",
310
+
"weight": 1.0
311
+
},
312
+
{
313
+
"a": "AH",
314
+
"b": "OY",
315
+
"id": "AH_OY",
316
+
"weight": 1.0
317
+
},
318
+
{
319
+
"a": "AH",
320
+
"b": "UH",
321
+
"id": "AH_UH",
322
+
"weight": 1.0
323
+
},
324
+
{
325
+
"a": "AH",
326
+
"b": "UW",
327
+
"id": "AH_UW",
328
+
"weight": 1.0
329
+
},
330
+
{
331
+
"a": "AO",
332
+
"b": "AA",
333
+
"id": "AO_AA",
334
+
"weight": 1.0
335
+
},
336
+
{
337
+
"a": "AO",
338
+
"b": "AE",
339
+
"id": "AO_AE",
340
+
"weight": 1.0
341
+
},
342
+
{
343
+
"a": "AO",
344
+
"b": "AH",
345
+
"id": "AO_AH",
346
+
"weight": 1.0
347
+
},
348
+
{
349
+
"a": "AO",
350
+
"b": "AO",
351
+
"id": "AO_AO",
352
+
"weight": 1.0
353
+
},
354
+
{
355
+
"a": "AO",
356
+
"b": "AW",
357
+
"id": "AO_AW",
358
+
"weight": 1.0
359
+
},
360
+
{
361
+
"a": "AO",
362
+
"b": "AY",
363
+
"id": "AO_AY",
364
+
"weight": 1.0
365
+
},
366
+
{
367
+
"a": "AO",
368
+
"b": "EH",
369
+
"id": "AO_EH",
370
+
"weight": 1.0
371
+
},
372
+
{
373
+
"a": "AO",
374
+
"b": "ER",
375
+
"id": "AO_ER",
376
+
"weight": 1.0
377
+
},
378
+
{
379
+
"a": "AO",
380
+
"b": "EY",
381
+
"id": "AO_EY",
382
+
"weight": 1.0
383
+
},
384
+
{
385
+
"a": "AO",
386
+
"b": "IH",
387
+
"id": "AO_IH",
388
+
"weight": 1.0
389
+
},
390
+
{
391
+
"a": "AO",
392
+
"b": "IY",
393
+
"id": "AO_IY",
394
+
"weight": 1.0
395
+
},
396
+
{
397
+
"a": "AO",
398
+
"b": "OW",
399
+
"id": "AO_OW",
400
+
"weight": 1.0
401
+
},
402
+
{
403
+
"a": "AO",
404
+
"b": "OY",
405
+
"id": "AO_OY",
406
+
"weight": 1.0
407
+
},
408
+
{
409
+
"a": "AO",
410
+
"b": "UH",
411
+
"id": "AO_UH",
412
+
"weight": 1.0
413
+
},
414
+
{
415
+
"a": "AO",
416
+
"b": "UW",
417
+
"id": "AO_UW",
418
+
"weight": 1.0
419
+
},
420
+
{
421
+
"a": "AW",
422
+
"b": "AA",
423
+
"id": "AW_AA",
424
+
"weight": 1.0
425
+
},
426
+
{
427
+
"a": "AW",
428
+
"b": "AE",
429
+
"id": "AW_AE",
430
+
"weight": 1.0
431
+
},
432
+
{
433
+
"a": "AW",
434
+
"b": "AH",
435
+
"id": "AW_AH",
436
+
"weight": 1.0
437
+
},
438
+
{
439
+
"a": "AW",
440
+
"b": "AO",
441
+
"id": "AW_AO",
442
+
"weight": 1.0
443
+
},
444
+
{
445
+
"a": "AW",
446
+
"b": "AW",
447
+
"id": "AW_AW",
448
+
"weight": 1.0
449
+
},
450
+
{
451
+
"a": "AW",
452
+
"b": "AY",
453
+
"id": "AW_AY",
454
+
"weight": 1.0
455
+
},
456
+
{
457
+
"a": "AW",
458
+
"b": "EH",
459
+
"id": "AW_EH",
460
+
"weight": 1.0
461
+
},
462
+
{
463
+
"a": "AW",
464
+
"b": "ER",
465
+
"id": "AW_ER",
466
+
"weight": 1.0
467
+
},
468
+
{
469
+
"a": "AW",
470
+
"b": "EY",
471
+
"id": "AW_EY",
472
+
"weight": 1.0
473
+
},
474
+
{
475
+
"a": "AW",
476
+
"b": "IH",
477
+
"id": "AW_IH",
478
+
"weight": 1.0
479
+
},
480
+
{
481
+
"a": "AW",
482
+
"b": "IY",
483
+
"id": "AW_IY",
484
+
"weight": 1.0
485
+
},
486
+
{
487
+
"a": "AW",
488
+
"b": "OW",
489
+
"id": "AW_OW",
490
+
"weight": 1.0
491
+
},
492
+
{
493
+
"a": "AW",
494
+
"b": "OY",
495
+
"id": "AW_OY",
496
+
"weight": 1.0
497
+
},
498
+
{
499
+
"a": "AW",
500
+
"b": "UH",
501
+
"id": "AW_UH",
502
+
"weight": 1.0
503
+
},
504
+
{
505
+
"a": "AW",
506
+
"b": "UW",
507
+
"id": "AW_UW",
508
+
"weight": 1.0
509
+
},
510
+
{
511
+
"a": "AY",
512
+
"b": "AA",
513
+
"id": "AY_AA",
514
+
"weight": 1.0
515
+
},
516
+
{
517
+
"a": "AY",
518
+
"b": "AE",
519
+
"id": "AY_AE",
520
+
"weight": 1.0
521
+
},
522
+
{
523
+
"a": "AY",
524
+
"b": "AH",
525
+
"id": "AY_AH",
526
+
"weight": 1.0
527
+
},
528
+
{
529
+
"a": "AY",
530
+
"b": "AO",
531
+
"id": "AY_AO",
532
+
"weight": 1.0
533
+
},
534
+
{
535
+
"a": "AY",
536
+
"b": "AW",
537
+
"id": "AY_AW",
538
+
"weight": 1.0
539
+
},
540
+
{
541
+
"a": "AY",
542
+
"b": "AY",
543
+
"id": "AY_AY",
544
+
"weight": 1.0
545
+
},
546
+
{
547
+
"a": "AY",
548
+
"b": "EH",
549
+
"id": "AY_EH",
550
+
"weight": 1.0
551
+
},
552
+
{
553
+
"a": "AY",
554
+
"b": "ER",
555
+
"id": "AY_ER",
556
+
"weight": 1.0
557
+
},
558
+
{
559
+
"a": "AY",
560
+
"b": "EY",
561
+
"id": "AY_EY",
562
+
"weight": 1.0
563
+
},
564
+
{
565
+
"a": "AY",
566
+
"b": "IH",
567
+
"id": "AY_IH",
568
+
"weight": 1.0
569
+
},
570
+
{
571
+
"a": "AY",
572
+
"b": "IY",
573
+
"id": "AY_IY",
574
+
"weight": 1.0
575
+
},
576
+
{
577
+
"a": "AY",
578
+
"b": "OW",
579
+
"id": "AY_OW",
580
+
"weight": 1.0
581
+
},
582
+
{
583
+
"a": "AY",
584
+
"b": "OY",
585
+
"id": "AY_OY",
586
+
"weight": 1.0
587
+
},
588
+
{
589
+
"a": "AY",
590
+
"b": "UH",
591
+
"id": "AY_UH",
592
+
"weight": 1.0
593
+
},
594
+
{
595
+
"a": "AY",
596
+
"b": "UW",
597
+
"id": "AY_UW",
598
+
"weight": 1.0
599
+
},
600
+
{
601
+
"a": "EH",
602
+
"b": "AA",
603
+
"id": "EH_AA",
604
+
"weight": 1.0
605
+
},
606
+
{
607
+
"a": "EH",
608
+
"b": "AE",
609
+
"id": "EH_AE",
610
+
"weight": 1.0
611
+
},
612
+
{
613
+
"a": "EH",
614
+
"b": "AH",
615
+
"id": "EH_AH",
616
+
"weight": 1.0
617
+
},
618
+
{
619
+
"a": "EH",
620
+
"b": "AO",
621
+
"id": "EH_AO",
622
+
"weight": 1.0
623
+
},
624
+
{
625
+
"a": "EH",
626
+
"b": "AW",
627
+
"id": "EH_AW",
628
+
"weight": 1.0
629
+
},
630
+
{
631
+
"a": "EH",
632
+
"b": "AY",
633
+
"id": "EH_AY",
634
+
"weight": 1.0
635
+
},
636
+
{
637
+
"a": "EH",
638
+
"b": "EH",
639
+
"id": "EH_EH",
640
+
"weight": 1.0
641
+
},
642
+
{
643
+
"a": "EH",
644
+
"b": "ER",
645
+
"id": "EH_ER",
646
+
"weight": 1.0
647
+
},
648
+
{
649
+
"a": "EH",
650
+
"b": "EY",
651
+
"id": "EH_EY",
652
+
"weight": 1.0
653
+
},
654
+
{
655
+
"a": "EH",
656
+
"b": "IH",
657
+
"id": "EH_IH",
658
+
"weight": 1.0
659
+
},
660
+
{
661
+
"a": "EH",
662
+
"b": "IY",
663
+
"id": "EH_IY",
664
+
"weight": 1.0
665
+
},
666
+
{
667
+
"a": "EH",
668
+
"b": "OW",
669
+
"id": "EH_OW",
670
+
"weight": 1.0
671
+
},
672
+
{
673
+
"a": "EH",
674
+
"b": "OY",
675
+
"id": "EH_OY",
676
+
"weight": 1.0
677
+
},
678
+
{
679
+
"a": "EH",
680
+
"b": "UH",
681
+
"id": "EH_UH",
682
+
"weight": 1.0
683
+
},
684
+
{
685
+
"a": "EH",
686
+
"b": "UW",
687
+
"id": "EH_UW",
688
+
"weight": 1.0
689
+
},
690
+
{
691
+
"a": "ER",
692
+
"b": "AA",
693
+
"id": "ER_AA",
694
+
"weight": 1.0
695
+
},
696
+
{
697
+
"a": "ER",
698
+
"b": "AE",
699
+
"id": "ER_AE",
700
+
"weight": 1.0
701
+
},
702
+
{
703
+
"a": "ER",
704
+
"b": "AH",
705
+
"id": "ER_AH",
706
+
"weight": 1.0
707
+
},
708
+
{
709
+
"a": "ER",
710
+
"b": "AO",
711
+
"id": "ER_AO",
712
+
"weight": 1.0
713
+
},
714
+
{
715
+
"a": "ER",
716
+
"b": "AW",
717
+
"id": "ER_AW",
718
+
"weight": 1.0
719
+
},
720
+
{
721
+
"a": "ER",
722
+
"b": "AY",
723
+
"id": "ER_AY",
724
+
"weight": 1.0
725
+
},
726
+
{
727
+
"a": "ER",
728
+
"b": "EH",
729
+
"id": "ER_EH",
730
+
"weight": 1.0
731
+
},
732
+
{
733
+
"a": "ER",
734
+
"b": "ER",
735
+
"id": "ER_ER",
736
+
"weight": 1.0
737
+
},
738
+
{
739
+
"a": "ER",
740
+
"b": "EY",
741
+
"id": "ER_EY",
742
+
"weight": 1.0
743
+
},
744
+
{
745
+
"a": "ER",
746
+
"b": "IH",
747
+
"id": "ER_IH",
748
+
"weight": 1.0
749
+
},
750
+
{
751
+
"a": "ER",
752
+
"b": "IY",
753
+
"id": "ER_IY",
754
+
"weight": 1.0
755
+
},
756
+
{
757
+
"a": "ER",
758
+
"b": "OW",
759
+
"id": "ER_OW",
760
+
"weight": 1.0
761
+
},
762
+
{
763
+
"a": "ER",
764
+
"b": "OY",
765
+
"id": "ER_OY",
766
+
"weight": 1.0
767
+
},
768
+
{
769
+
"a": "ER",
770
+
"b": "UH",
771
+
"id": "ER_UH",
772
+
"weight": 1.0
773
+
},
774
+
{
775
+
"a": "ER",
776
+
"b": "UW",
777
+
"id": "ER_UW",
778
+
"weight": 1.0
779
+
},
780
+
{
781
+
"a": "EY",
782
+
"b": "AA",
783
+
"id": "EY_AA",
784
+
"weight": 1.0
785
+
},
786
+
{
787
+
"a": "EY",
788
+
"b": "AE",
789
+
"id": "EY_AE",
790
+
"weight": 1.0
791
+
},
792
+
{
793
+
"a": "EY",
794
+
"b": "AH",
795
+
"id": "EY_AH",
796
+
"weight": 1.0
797
+
},
798
+
{
799
+
"a": "EY",
800
+
"b": "AO",
801
+
"id": "EY_AO",
802
+
"weight": 1.0
803
+
},
804
+
{
805
+
"a": "EY",
806
+
"b": "AW",
807
+
"id": "EY_AW",
808
+
"weight": 1.0
809
+
},
810
+
{
811
+
"a": "EY",
812
+
"b": "AY",
813
+
"id": "EY_AY",
814
+
"weight": 1.0
815
+
},
816
+
{
817
+
"a": "EY",
818
+
"b": "EH",
819
+
"id": "EY_EH",
820
+
"weight": 1.0
821
+
},
822
+
{
823
+
"a": "EY",
824
+
"b": "ER",
825
+
"id": "EY_ER",
826
+
"weight": 1.0
827
+
},
828
+
{
829
+
"a": "EY",
830
+
"b": "EY",
831
+
"id": "EY_EY",
832
+
"weight": 1.0
833
+
},
834
+
{
835
+
"a": "EY",
836
+
"b": "IH",
837
+
"id": "EY_IH",
838
+
"weight": 1.0
839
+
},
840
+
{
841
+
"a": "EY",
842
+
"b": "IY",
843
+
"id": "EY_IY",
844
+
"weight": 1.0
845
+
},
846
+
{
847
+
"a": "EY",
848
+
"b": "OW",
849
+
"id": "EY_OW",
850
+
"weight": 1.0
851
+
},
852
+
{
853
+
"a": "EY",
854
+
"b": "OY",
855
+
"id": "EY_OY",
856
+
"weight": 1.0
857
+
},
858
+
{
859
+
"a": "EY",
860
+
"b": "UH",
861
+
"id": "EY_UH",
862
+
"weight": 1.0
863
+
},
864
+
{
865
+
"a": "EY",
866
+
"b": "UW",
867
+
"id": "EY_UW",
868
+
"weight": 1.0
869
+
},
870
+
{
871
+
"a": "IH",
872
+
"b": "AA",
873
+
"id": "IH_AA",
874
+
"weight": 1.0
875
+
},
876
+
{
877
+
"a": "IH",
878
+
"b": "AE",
879
+
"id": "IH_AE",
880
+
"weight": 1.0
881
+
},
882
+
{
883
+
"a": "IH",
884
+
"b": "AH",
885
+
"id": "IH_AH",
886
+
"weight": 1.0
887
+
},
888
+
{
889
+
"a": "IH",
890
+
"b": "AO",
891
+
"id": "IH_AO",
892
+
"weight": 1.0
893
+
},
894
+
{
895
+
"a": "IH",
896
+
"b": "AW",
897
+
"id": "IH_AW",
898
+
"weight": 1.0
899
+
},
900
+
{
901
+
"a": "IH",
902
+
"b": "AY",
903
+
"id": "IH_AY",
904
+
"weight": 1.0
905
+
},
906
+
{
907
+
"a": "IH",
908
+
"b": "EH",
909
+
"id": "IH_EH",
910
+
"weight": 1.0
911
+
},
912
+
{
913
+
"a": "IH",
914
+
"b": "ER",
915
+
"id": "IH_ER",
916
+
"weight": 1.0
917
+
},
918
+
{
919
+
"a": "IH",
920
+
"b": "EY",
921
+
"id": "IH_EY",
922
+
"weight": 1.0
923
+
},
924
+
{
925
+
"a": "IH",
926
+
"b": "IH",
927
+
"id": "IH_IH",
928
+
"weight": 1.0
929
+
},
930
+
{
931
+
"a": "IH",
932
+
"b": "IY",
933
+
"id": "IH_IY",
934
+
"weight": 1.0
935
+
},
936
+
{
937
+
"a": "IH",
938
+
"b": "OW",
939
+
"id": "IH_OW",
940
+
"weight": 1.0
941
+
},
942
+
{
943
+
"a": "IH",
944
+
"b": "OY",
945
+
"id": "IH_OY",
946
+
"weight": 1.0
947
+
},
948
+
{
949
+
"a": "IH",
950
+
"b": "UH",
951
+
"id": "IH_UH",
952
+
"weight": 1.0
953
+
},
954
+
{
955
+
"a": "IH",
956
+
"b": "UW",
957
+
"id": "IH_UW",
958
+
"weight": 1.0
959
+
},
960
+
{
961
+
"a": "IY",
962
+
"b": "AA",
963
+
"id": "IY_AA",
964
+
"weight": 1.0
965
+
},
966
+
{
967
+
"a": "IY",
968
+
"b": "AE",
969
+
"id": "IY_AE",
970
+
"weight": 1.0
971
+
},
972
+
{
973
+
"a": "IY",
974
+
"b": "AH",
975
+
"id": "IY_AH",
976
+
"weight": 1.0
977
+
},
978
+
{
979
+
"a": "IY",
980
+
"b": "AO",
981
+
"id": "IY_AO",
982
+
"weight": 1.0
983
+
},
984
+
{
985
+
"a": "IY",
986
+
"b": "AW",
987
+
"id": "IY_AW",
988
+
"weight": 1.0
989
+
},
990
+
{
991
+
"a": "IY",
992
+
"b": "AY",
993
+
"id": "IY_AY",
994
+
"weight": 1.0
995
+
},
996
+
{
997
+
"a": "IY",
998
+
"b": "EH",
999
+
"id": "IY_EH",
1000
+
"weight": 1.0
1001
+
},
1002
+
{
1003
+
"a": "IY",
1004
+
"b": "ER",
1005
+
"id": "IY_ER",
1006
+
"weight": 1.0
1007
+
},
1008
+
{
1009
+
"a": "IY",
1010
+
"b": "EY",
1011
+
"id": "IY_EY",
1012
+
"weight": 1.0
1013
+
},
1014
+
{
1015
+
"a": "IY",
1016
+
"b": "IH",
1017
+
"id": "IY_IH",
1018
+
"weight": 1.0
1019
+
},
1020
+
{
1021
+
"a": "IY",
1022
+
"b": "IY",
1023
+
"id": "IY_IY",
1024
+
"weight": 1.0
1025
+
},
1026
+
{
1027
+
"a": "IY",
1028
+
"b": "OW",
1029
+
"id": "IY_OW",
1030
+
"weight": 1.0
1031
+
},
1032
+
{
1033
+
"a": "IY",
1034
+
"b": "OY",
1035
+
"id": "IY_OY",
1036
+
"weight": 1.0
1037
+
},
1038
+
{
1039
+
"a": "IY",
1040
+
"b": "UH",
1041
+
"id": "IY_UH",
1042
+
"weight": 1.0
1043
+
},
1044
+
{
1045
+
"a": "IY",
1046
+
"b": "UW",
1047
+
"id": "IY_UW",
1048
+
"weight": 1.0
1049
+
},
1050
+
{
1051
+
"a": "OW",
1052
+
"b": "AA",
1053
+
"id": "OW_AA",
1054
+
"weight": 1.0
1055
+
},
1056
+
{
1057
+
"a": "OW",
1058
+
"b": "AE",
1059
+
"id": "OW_AE",
1060
+
"weight": 1.0
1061
+
},
1062
+
{
1063
+
"a": "OW",
1064
+
"b": "AH",
1065
+
"id": "OW_AH",
1066
+
"weight": 1.0
1067
+
},
1068
+
{
1069
+
"a": "OW",
1070
+
"b": "AO",
1071
+
"id": "OW_AO",
1072
+
"weight": 1.0
1073
+
},
1074
+
{
1075
+
"a": "OW",
1076
+
"b": "AW",
1077
+
"id": "OW_AW",
1078
+
"weight": 1.0
1079
+
},
1080
+
{
1081
+
"a": "OW",
1082
+
"b": "AY",
1083
+
"id": "OW_AY",
1084
+
"weight": 1.0
1085
+
},
1086
+
{
1087
+
"a": "OW",
1088
+
"b": "EH",
1089
+
"id": "OW_EH",
1090
+
"weight": 1.0
1091
+
},
1092
+
{
1093
+
"a": "OW",
1094
+
"b": "ER",
1095
+
"id": "OW_ER",
1096
+
"weight": 1.0
1097
+
},
1098
+
{
1099
+
"a": "OW",
1100
+
"b": "EY",
1101
+
"id": "OW_EY",
1102
+
"weight": 1.0
1103
+
},
1104
+
{
1105
+
"a": "OW",
1106
+
"b": "IH",
1107
+
"id": "OW_IH",
1108
+
"weight": 1.0
1109
+
},
1110
+
{
1111
+
"a": "OW",
1112
+
"b": "IY",
1113
+
"id": "OW_IY",
1114
+
"weight": 1.0
1115
+
},
1116
+
{
1117
+
"a": "OW",
1118
+
"b": "OW",
1119
+
"id": "OW_OW",
1120
+
"weight": 1.0
1121
+
},
1122
+
{
1123
+
"a": "OW",
1124
+
"b": "OY",
1125
+
"id": "OW_OY",
1126
+
"weight": 1.0
1127
+
},
1128
+
{
1129
+
"a": "OW",
1130
+
"b": "UH",
1131
+
"id": "OW_UH",
1132
+
"weight": 1.0
1133
+
},
1134
+
{
1135
+
"a": "OW",
1136
+
"b": "UW",
1137
+
"id": "OW_UW",
1138
+
"weight": 1.0
1139
+
},
1140
+
{
1141
+
"a": "OY",
1142
+
"b": "AA",
1143
+
"id": "OY_AA",
1144
+
"weight": 1.0
1145
+
},
1146
+
{
1147
+
"a": "OY",
1148
+
"b": "AE",
1149
+
"id": "OY_AE",
1150
+
"weight": 1.0
1151
+
},
1152
+
{
1153
+
"a": "OY",
1154
+
"b": "AH",
1155
+
"id": "OY_AH",
1156
+
"weight": 1.0
1157
+
},
1158
+
{
1159
+
"a": "OY",
1160
+
"b": "AO",
1161
+
"id": "OY_AO",
1162
+
"weight": 1.0
1163
+
},
1164
+
{
1165
+
"a": "OY",
1166
+
"b": "AW",
1167
+
"id": "OY_AW",
1168
+
"weight": 1.0
1169
+
},
1170
+
{
1171
+
"a": "OY",
1172
+
"b": "AY",
1173
+
"id": "OY_AY",
1174
+
"weight": 1.0
1175
+
},
1176
+
{
1177
+
"a": "OY",
1178
+
"b": "EH",
1179
+
"id": "OY_EH",
1180
+
"weight": 1.0
1181
+
},
1182
+
{
1183
+
"a": "OY",
1184
+
"b": "ER",
1185
+
"id": "OY_ER",
1186
+
"weight": 1.0
1187
+
},
1188
+
{
1189
+
"a": "OY",
1190
+
"b": "EY",
1191
+
"id": "OY_EY",
1192
+
"weight": 1.0
1193
+
},
1194
+
{
1195
+
"a": "OY",
1196
+
"b": "IH",
1197
+
"id": "OY_IH",
1198
+
"weight": 1.0
1199
+
},
1200
+
{
1201
+
"a": "OY",
1202
+
"b": "IY",
1203
+
"id": "OY_IY",
1204
+
"weight": 1.0
1205
+
},
1206
+
{
1207
+
"a": "OY",
1208
+
"b": "OW",
1209
+
"id": "OY_OW",
1210
+
"weight": 1.0
1211
+
},
1212
+
{
1213
+
"a": "OY",
1214
+
"b": "OY",
1215
+
"id": "OY_OY",
1216
+
"weight": 1.0
1217
+
},
1218
+
{
1219
+
"a": "OY",
1220
+
"b": "UH",
1221
+
"id": "OY_UH",
1222
+
"weight": 1.0
1223
+
},
1224
+
{
1225
+
"a": "OY",
1226
+
"b": "UW",
1227
+
"id": "OY_UW",
1228
+
"weight": 1.0
1229
+
},
1230
+
{
1231
+
"a": "UH",
1232
+
"b": "AA",
1233
+
"id": "UH_AA",
1234
+
"weight": 1.0
1235
+
},
1236
+
{
1237
+
"a": "UH",
1238
+
"b": "AE",
1239
+
"id": "UH_AE",
1240
+
"weight": 1.0
1241
+
},
1242
+
{
1243
+
"a": "UH",
1244
+
"b": "AH",
1245
+
"id": "UH_AH",
1246
+
"weight": 1.0
1247
+
},
1248
+
{
1249
+
"a": "UH",
1250
+
"b": "AO",
1251
+
"id": "UH_AO",
1252
+
"weight": 1.0
1253
+
},
1254
+
{
1255
+
"a": "UH",
1256
+
"b": "AW",
1257
+
"id": "UH_AW",
1258
+
"weight": 1.0
1259
+
},
1260
+
{
1261
+
"a": "UH",
1262
+
"b": "AY",
1263
+
"id": "UH_AY",
1264
+
"weight": 1.0
1265
+
},
1266
+
{
1267
+
"a": "UH",
1268
+
"b": "EH",
1269
+
"id": "UH_EH",
1270
+
"weight": 1.0
1271
+
},
1272
+
{
1273
+
"a": "UH",
1274
+
"b": "ER",
1275
+
"id": "UH_ER",
1276
+
"weight": 1.0
1277
+
},
1278
+
{
1279
+
"a": "UH",
1280
+
"b": "EY",
1281
+
"id": "UH_EY",
1282
+
"weight": 1.0
1283
+
},
1284
+
{
1285
+
"a": "UH",
1286
+
"b": "IH",
1287
+
"id": "UH_IH",
1288
+
"weight": 1.0
1289
+
},
1290
+
{
1291
+
"a": "UH",
1292
+
"b": "IY",
1293
+
"id": "UH_IY",
1294
+
"weight": 1.0
1295
+
},
1296
+
{
1297
+
"a": "UH",
1298
+
"b": "OW",
1299
+
"id": "UH_OW",
1300
+
"weight": 1.0
1301
+
},
1302
+
{
1303
+
"a": "UH",
1304
+
"b": "OY",
1305
+
"id": "UH_OY",
1306
+
"weight": 1.0
1307
+
},
1308
+
{
1309
+
"a": "UH",
1310
+
"b": "UH",
1311
+
"id": "UH_UH",
1312
+
"weight": 1.0
1313
+
},
1314
+
{
1315
+
"a": "UH",
1316
+
"b": "UW",
1317
+
"id": "UH_UW",
1318
+
"weight": 1.0
1319
+
},
1320
+
{
1321
+
"a": "UW",
1322
+
"b": "AA",
1323
+
"id": "UW_AA",
1324
+
"weight": 1.0
1325
+
},
1326
+
{
1327
+
"a": "UW",
1328
+
"b": "AE",
1329
+
"id": "UW_AE",
1330
+
"weight": 1.0
1331
+
},
1332
+
{
1333
+
"a": "UW",
1334
+
"b": "AH",
1335
+
"id": "UW_AH",
1336
+
"weight": 1.0
1337
+
},
1338
+
{
1339
+
"a": "UW",
1340
+
"b": "AO",
1341
+
"id": "UW_AO",
1342
+
"weight": 1.0
1343
+
},
1344
+
{
1345
+
"a": "UW",
1346
+
"b": "AW",
1347
+
"id": "UW_AW",
1348
+
"weight": 1.0
1349
+
},
1350
+
{
1351
+
"a": "UW",
1352
+
"b": "AY",
1353
+
"id": "UW_AY",
1354
+
"weight": 1.0
1355
+
},
1356
+
{
1357
+
"a": "UW",
1358
+
"b": "EH",
1359
+
"id": "UW_EH",
1360
+
"weight": 1.0
1361
+
},
1362
+
{
1363
+
"a": "UW",
1364
+
"b": "ER",
1365
+
"id": "UW_ER",
1366
+
"weight": 1.0
1367
+
},
1368
+
{
1369
+
"a": "UW",
1370
+
"b": "EY",
1371
+
"id": "UW_EY",
1372
+
"weight": 1.0
1373
+
},
1374
+
{
1375
+
"a": "UW",
1376
+
"b": "IH",
1377
+
"id": "UW_IH",
1378
+
"weight": 1.0
1379
+
},
1380
+
{
1381
+
"a": "UW",
1382
+
"b": "IY",
1383
+
"id": "UW_IY",
1384
+
"weight": 1.0
1385
+
},
1386
+
{
1387
+
"a": "UW",
1388
+
"b": "OW",
1389
+
"id": "UW_OW",
1390
+
"weight": 1.0
1391
+
},
1392
+
{
1393
+
"a": "UW",
1394
+
"b": "OY",
1395
+
"id": "UW_OY",
1396
+
"weight": 1.0
1397
+
},
1398
+
{
1399
+
"a": "UW",
1400
+
"b": "UH",
1401
+
"id": "UW_UH",
1402
+
"weight": 1.0
1403
+
},
1404
+
{
1405
+
"a": "UW",
1406
+
"b": "UW",
1407
+
"id": "UW_UW",
1408
+
"weight": 1.0
1409
+
},
1410
+
{
1411
+
"a": "#B",
1412
+
"b": "AA",
1413
+
"id": "#B_AA",
1414
+
"weight": 0.95
1415
+
},
1416
+
{
1417
+
"a": "#B",
1418
+
"b": "AE",
1419
+
"id": "#B_AE",
1420
+
"weight": 0.95
1421
+
},
1422
+
{
1423
+
"a": "#B",
1424
+
"b": "AH",
1425
+
"id": "#B_AH",
1426
+
"weight": 0.95
1427
+
},
1428
+
{
1429
+
"a": "#B",
1430
+
"b": "AO",
1431
+
"id": "#B_AO",
1432
+
"weight": 0.95
1433
+
},
1434
+
{
1435
+
"a": "#B",
1436
+
"b": "AW",
1437
+
"id": "#B_AW",
1438
+
"weight": 0.95
1439
+
},
1440
+
{
1441
+
"a": "#B",
1442
+
"b": "AY",
1443
+
"id": "#B_AY",
1444
+
"weight": 0.95
1445
+
},
1446
+
{
1447
+
"a": "#B",
1448
+
"b": "EH",
1449
+
"id": "#B_EH",
1450
+
"weight": 0.95
1451
+
},
1452
+
{
1453
+
"a": "#B",
1454
+
"b": "ER",
1455
+
"id": "#B_ER",
1456
+
"weight": 0.95
1457
+
},
1458
+
{
1459
+
"a": "#B",
1460
+
"b": "EY",
1461
+
"id": "#B_EY",
1462
+
"weight": 0.95
1463
+
},
1464
+
{
1465
+
"a": "#B",
1466
+
"b": "IH",
1467
+
"id": "#B_IH",
1468
+
"weight": 0.95
1469
+
},
1470
+
{
1471
+
"a": "#B",
1472
+
"b": "IY",
1473
+
"id": "#B_IY",
1474
+
"weight": 0.95
1475
+
},
1476
+
{
1477
+
"a": "#B",
1478
+
"b": "OW",
1479
+
"id": "#B_OW",
1480
+
"weight": 0.95
1481
+
},
1482
+
{
1483
+
"a": "#B",
1484
+
"b": "OY",
1485
+
"id": "#B_OY",
1486
+
"weight": 0.95
1487
+
},
1488
+
{
1489
+
"a": "#B",
1490
+
"b": "UH",
1491
+
"id": "#B_UH",
1492
+
"weight": 0.95
1493
+
},
1494
+
{
1495
+
"a": "#B",
1496
+
"b": "UW",
1497
+
"id": "#B_UW",
1498
+
"weight": 0.95
1499
+
},
1500
+
{
1501
+
"a": "B",
1502
+
"b": "AA",
1503
+
"id": "B_AA",
1504
+
"weight": 0.95
1505
+
},
1506
+
{
1507
+
"a": "B",
1508
+
"b": "AE",
1509
+
"id": "B_AE",
1510
+
"weight": 0.95
1511
+
},
1512
+
{
1513
+
"a": "B",
1514
+
"b": "AH",
1515
+
"id": "B_AH",
1516
+
"weight": 0.95
1517
+
},
1518
+
{
1519
+
"a": "B",
1520
+
"b": "AO",
1521
+
"id": "B_AO",
1522
+
"weight": 0.95
1523
+
},
1524
+
{
1525
+
"a": "B",
1526
+
"b": "AW",
1527
+
"id": "B_AW",
1528
+
"weight": 0.95
1529
+
},
1530
+
{
1531
+
"a": "B",
1532
+
"b": "AY",
1533
+
"id": "B_AY",
1534
+
"weight": 0.95
1535
+
},
1536
+
{
1537
+
"a": "B",
1538
+
"b": "EH",
1539
+
"id": "B_EH",
1540
+
"weight": 0.95
1541
+
},
1542
+
{
1543
+
"a": "B",
1544
+
"b": "ER",
1545
+
"id": "B_ER",
1546
+
"weight": 0.95
1547
+
},
1548
+
{
1549
+
"a": "B",
1550
+
"b": "EY",
1551
+
"id": "B_EY",
1552
+
"weight": 0.95
1553
+
},
1554
+
{
1555
+
"a": "B",
1556
+
"b": "IH",
1557
+
"id": "B_IH",
1558
+
"weight": 0.95
1559
+
},
1560
+
{
1561
+
"a": "B",
1562
+
"b": "IY",
1563
+
"id": "B_IY",
1564
+
"weight": 0.95
1565
+
},
1566
+
{
1567
+
"a": "B",
1568
+
"b": "OW",
1569
+
"id": "B_OW",
1570
+
"weight": 0.95
1571
+
},
1572
+
{
1573
+
"a": "B",
1574
+
"b": "OY",
1575
+
"id": "B_OY",
1576
+
"weight": 0.95
1577
+
},
1578
+
{
1579
+
"a": "B",
1580
+
"b": "UH",
1581
+
"id": "B_UH",
1582
+
"weight": 0.95
1583
+
},
1584
+
{
1585
+
"a": "B",
1586
+
"b": "UW",
1587
+
"id": "B_UW",
1588
+
"weight": 0.95
1589
+
},
1590
+
{
1591
+
"a": "D",
1592
+
"b": "AA",
1593
+
"id": "D_AA",
1594
+
"weight": 0.95
1595
+
},
1596
+
{
1597
+
"a": "D",
1598
+
"b": "AE",
1599
+
"id": "D_AE",
1600
+
"weight": 0.95
1601
+
},
1602
+
{
1603
+
"a": "D",
1604
+
"b": "AH",
1605
+
"id": "D_AH",
1606
+
"weight": 0.95
1607
+
},
1608
+
{
1609
+
"a": "D",
1610
+
"b": "AO",
1611
+
"id": "D_AO",
1612
+
"weight": 0.95
1613
+
},
1614
+
{
1615
+
"a": "D",
1616
+
"b": "AW",
1617
+
"id": "D_AW",
1618
+
"weight": 0.95
1619
+
},
1620
+
{
1621
+
"a": "D",
1622
+
"b": "AY",
1623
+
"id": "D_AY",
1624
+
"weight": 0.95
1625
+
},
1626
+
{
1627
+
"a": "D",
1628
+
"b": "EH",
1629
+
"id": "D_EH",
1630
+
"weight": 0.95
1631
+
},
1632
+
{
1633
+
"a": "D",
1634
+
"b": "ER",
1635
+
"id": "D_ER",
1636
+
"weight": 0.95
1637
+
},
1638
+
{
1639
+
"a": "D",
1640
+
"b": "EY",
1641
+
"id": "D_EY",
1642
+
"weight": 0.95
1643
+
},
1644
+
{
1645
+
"a": "D",
1646
+
"b": "IH",
1647
+
"id": "D_IH",
1648
+
"weight": 0.95
1649
+
},
1650
+
{
1651
+
"a": "D",
1652
+
"b": "IY",
1653
+
"id": "D_IY",
1654
+
"weight": 0.95
1655
+
},
1656
+
{
1657
+
"a": "D",
1658
+
"b": "OW",
1659
+
"id": "D_OW",
1660
+
"weight": 0.95
1661
+
},
1662
+
{
1663
+
"a": "D",
1664
+
"b": "OY",
1665
+
"id": "D_OY",
1666
+
"weight": 0.95
1667
+
},
1668
+
{
1669
+
"a": "D",
1670
+
"b": "UH",
1671
+
"id": "D_UH",
1672
+
"weight": 0.95
1673
+
},
1674
+
{
1675
+
"a": "D",
1676
+
"b": "UW",
1677
+
"id": "D_UW",
1678
+
"weight": 0.95
1679
+
},
1680
+
{
1681
+
"a": "G",
1682
+
"b": "AA",
1683
+
"id": "G_AA",
1684
+
"weight": 0.95
1685
+
},
1686
+
{
1687
+
"a": "G",
1688
+
"b": "AE",
1689
+
"id": "G_AE",
1690
+
"weight": 0.95
1691
+
},
1692
+
{
1693
+
"a": "G",
1694
+
"b": "AH",
1695
+
"id": "G_AH",
1696
+
"weight": 0.95
1697
+
},
1698
+
{
1699
+
"a": "G",
1700
+
"b": "AO",
1701
+
"id": "G_AO",
1702
+
"weight": 0.95
1703
+
},
1704
+
{
1705
+
"a": "G",
1706
+
"b": "AW",
1707
+
"id": "G_AW",
1708
+
"weight": 0.95
1709
+
},
1710
+
{
1711
+
"a": "G",
1712
+
"b": "AY",
1713
+
"id": "G_AY",
1714
+
"weight": 0.95
1715
+
},
1716
+
{
1717
+
"a": "G",
1718
+
"b": "EH",
1719
+
"id": "G_EH",
1720
+
"weight": 0.95
1721
+
},
1722
+
{
1723
+
"a": "G",
1724
+
"b": "ER",
1725
+
"id": "G_ER",
1726
+
"weight": 0.95
1727
+
},
1728
+
{
1729
+
"a": "G",
1730
+
"b": "EY",
1731
+
"id": "G_EY",
1732
+
"weight": 0.95
1733
+
},
1734
+
{
1735
+
"a": "G",
1736
+
"b": "IH",
1737
+
"id": "G_IH",
1738
+
"weight": 0.95
1739
+
},
1740
+
{
1741
+
"a": "G",
1742
+
"b": "IY",
1743
+
"id": "G_IY",
1744
+
"weight": 0.95
1745
+
},
1746
+
{
1747
+
"a": "G",
1748
+
"b": "OW",
1749
+
"id": "G_OW",
1750
+
"weight": 0.95
1751
+
},
1752
+
{
1753
+
"a": "G",
1754
+
"b": "OY",
1755
+
"id": "G_OY",
1756
+
"weight": 0.95
1757
+
},
1758
+
{
1759
+
"a": "G",
1760
+
"b": "UH",
1761
+
"id": "G_UH",
1762
+
"weight": 0.95
1763
+
},
1764
+
{
1765
+
"a": "G",
1766
+
"b": "UW",
1767
+
"id": "G_UW",
1768
+
"weight": 0.95
1769
+
},
1770
+
{
1771
+
"a": "K",
1772
+
"b": "AA",
1773
+
"id": "K_AA",
1774
+
"weight": 0.95
1775
+
},
1776
+
{
1777
+
"a": "K",
1778
+
"b": "AE",
1779
+
"id": "K_AE",
1780
+
"weight": 0.95
1781
+
},
1782
+
{
1783
+
"a": "K",
1784
+
"b": "AH",
1785
+
"id": "K_AH",
1786
+
"weight": 0.95
1787
+
},
1788
+
{
1789
+
"a": "K",
1790
+
"b": "AO",
1791
+
"id": "K_AO",
1792
+
"weight": 0.95
1793
+
},
1794
+
{
1795
+
"a": "K",
1796
+
"b": "AW",
1797
+
"id": "K_AW",
1798
+
"weight": 0.95
1799
+
},
1800
+
{
1801
+
"a": "K",
1802
+
"b": "AY",
1803
+
"id": "K_AY",
1804
+
"weight": 0.95
1805
+
},
1806
+
{
1807
+
"a": "K",
1808
+
"b": "EH",
1809
+
"id": "K_EH",
1810
+
"weight": 0.95
1811
+
},
1812
+
{
1813
+
"a": "K",
1814
+
"b": "ER",
1815
+
"id": "K_ER",
1816
+
"weight": 0.95
1817
+
},
1818
+
{
1819
+
"a": "K",
1820
+
"b": "EY",
1821
+
"id": "K_EY",
1822
+
"weight": 0.95
1823
+
},
1824
+
{
1825
+
"a": "K",
1826
+
"b": "IH",
1827
+
"id": "K_IH",
1828
+
"weight": 0.95
1829
+
},
1830
+
{
1831
+
"a": "K",
1832
+
"b": "IY",
1833
+
"id": "K_IY",
1834
+
"weight": 0.95
1835
+
},
1836
+
{
1837
+
"a": "K",
1838
+
"b": "OW",
1839
+
"id": "K_OW",
1840
+
"weight": 0.95
1841
+
},
1842
+
{
1843
+
"a": "K",
1844
+
"b": "OY",
1845
+
"id": "K_OY",
1846
+
"weight": 0.95
1847
+
},
1848
+
{
1849
+
"a": "K",
1850
+
"b": "UH",
1851
+
"id": "K_UH",
1852
+
"weight": 0.95
1853
+
},
1854
+
{
1855
+
"a": "K",
1856
+
"b": "UW",
1857
+
"id": "K_UW",
1858
+
"weight": 0.95
1859
+
},
1860
+
{
1861
+
"a": "P",
1862
+
"b": "AA",
1863
+
"id": "P_AA",
1864
+
"weight": 0.95
1865
+
},
1866
+
{
1867
+
"a": "P",
1868
+
"b": "AE",
1869
+
"id": "P_AE",
1870
+
"weight": 0.95
1871
+
},
1872
+
{
1873
+
"a": "P",
1874
+
"b": "AH",
1875
+
"id": "P_AH",
1876
+
"weight": 0.95
1877
+
},
1878
+
{
1879
+
"a": "P",
1880
+
"b": "AO",
1881
+
"id": "P_AO",
1882
+
"weight": 0.95
1883
+
},
1884
+
{
1885
+
"a": "P",
1886
+
"b": "AW",
1887
+
"id": "P_AW",
1888
+
"weight": 0.95
1889
+
},
1890
+
{
1891
+
"a": "P",
1892
+
"b": "AY",
1893
+
"id": "P_AY",
1894
+
"weight": 0.95
1895
+
},
1896
+
{
1897
+
"a": "P",
1898
+
"b": "EH",
1899
+
"id": "P_EH",
1900
+
"weight": 0.95
1901
+
},
1902
+
{
1903
+
"a": "P",
1904
+
"b": "ER",
1905
+
"id": "P_ER",
1906
+
"weight": 0.95
1907
+
},
1908
+
{
1909
+
"a": "P",
1910
+
"b": "EY",
1911
+
"id": "P_EY",
1912
+
"weight": 0.95
1913
+
},
1914
+
{
1915
+
"a": "P",
1916
+
"b": "IH",
1917
+
"id": "P_IH",
1918
+
"weight": 0.95
1919
+
},
1920
+
{
1921
+
"a": "P",
1922
+
"b": "IY",
1923
+
"id": "P_IY",
1924
+
"weight": 0.95
1925
+
},
1926
+
{
1927
+
"a": "P",
1928
+
"b": "OW",
1929
+
"id": "P_OW",
1930
+
"weight": 0.95
1931
+
},
1932
+
{
1933
+
"a": "P",
1934
+
"b": "OY",
1935
+
"id": "P_OY",
1936
+
"weight": 0.95
1937
+
},
1938
+
{
1939
+
"a": "P",
1940
+
"b": "UH",
1941
+
"id": "P_UH",
1942
+
"weight": 0.95
1943
+
},
1944
+
{
1945
+
"a": "P",
1946
+
"b": "UW",
1947
+
"id": "P_UW",
1948
+
"weight": 0.95
1949
+
},
1950
+
{
1951
+
"a": "T",
1952
+
"b": "AA",
1953
+
"id": "T_AA",
1954
+
"weight": 0.95
1955
+
},
1956
+
{
1957
+
"a": "T",
1958
+
"b": "AE",
1959
+
"id": "T_AE",
1960
+
"weight": 0.95
1961
+
},
1962
+
{
1963
+
"a": "T",
1964
+
"b": "AH",
1965
+
"id": "T_AH",
1966
+
"weight": 0.95
1967
+
},
1968
+
{
1969
+
"a": "T",
1970
+
"b": "AO",
1971
+
"id": "T_AO",
1972
+
"weight": 0.95
1973
+
},
1974
+
{
1975
+
"a": "T",
1976
+
"b": "AW",
1977
+
"id": "T_AW",
1978
+
"weight": 0.95
1979
+
},
1980
+
{
1981
+
"a": "T",
1982
+
"b": "AY",
1983
+
"id": "T_AY",
1984
+
"weight": 0.95
1985
+
},
1986
+
{
1987
+
"a": "T",
1988
+
"b": "EH",
1989
+
"id": "T_EH",
1990
+
"weight": 0.95
1991
+
},
1992
+
{
1993
+
"a": "T",
1994
+
"b": "ER",
1995
+
"id": "T_ER",
1996
+
"weight": 0.95
1997
+
},
1998
+
{
1999
+
"a": "T",
2000
+
"b": "EY",
2001
+
"id": "T_EY",
2002
+
"weight": 0.95
2003
+
},
2004
+
{
2005
+
"a": "T",
2006
+
"b": "IH",
2007
+
"id": "T_IH",
2008
+
"weight": 0.95
2009
+
},
2010
+
{
2011
+
"a": "T",
2012
+
"b": "IY",
2013
+
"id": "T_IY",
2014
+
"weight": 0.95
2015
+
},
2016
+
{
2017
+
"a": "T",
2018
+
"b": "OW",
2019
+
"id": "T_OW",
2020
+
"weight": 0.95
2021
+
},
2022
+
{
2023
+
"a": "T",
2024
+
"b": "OY",
2025
+
"id": "T_OY",
2026
+
"weight": 0.95
2027
+
},
2028
+
{
2029
+
"a": "T",
2030
+
"b": "UH",
2031
+
"id": "T_UH",
2032
+
"weight": 0.95
2033
+
},
2034
+
{
2035
+
"a": "T",
2036
+
"b": "UW",
2037
+
"id": "T_UW",
2038
+
"weight": 0.95
2039
+
},
2040
+
{
2041
+
"a": "AA",
2042
+
"b": "B",
2043
+
"id": "AA_B",
2044
+
"weight": 0.9
2045
+
},
2046
+
{
2047
+
"a": "AA",
2048
+
"b": "D",
2049
+
"id": "AA_D",
2050
+
"weight": 0.9
2051
+
},
2052
+
{
2053
+
"a": "AA",
2054
+
"b": "G",
2055
+
"id": "AA_G",
2056
+
"weight": 0.9
2057
+
},
2058
+
{
2059
+
"a": "AA",
2060
+
"b": "K",
2061
+
"id": "AA_K",
2062
+
"weight": 0.9
2063
+
},
2064
+
{
2065
+
"a": "AA",
2066
+
"b": "P",
2067
+
"id": "AA_P",
2068
+
"weight": 0.9
2069
+
},
2070
+
{
2071
+
"a": "AA",
2072
+
"b": "T",
2073
+
"id": "AA_T",
2074
+
"weight": 0.9
2075
+
},
2076
+
{
2077
+
"a": "AE",
2078
+
"b": "B",
2079
+
"id": "AE_B",
2080
+
"weight": 0.9
2081
+
},
2082
+
{
2083
+
"a": "AE",
2084
+
"b": "D",
2085
+
"id": "AE_D",
2086
+
"weight": 0.9
2087
+
},
2088
+
{
2089
+
"a": "AE",
2090
+
"b": "G",
2091
+
"id": "AE_G",
2092
+
"weight": 0.9
2093
+
},
2094
+
{
2095
+
"a": "AE",
2096
+
"b": "K",
2097
+
"id": "AE_K",
2098
+
"weight": 0.9
2099
+
},
2100
+
{
2101
+
"a": "AE",
2102
+
"b": "P",
2103
+
"id": "AE_P",
2104
+
"weight": 0.9
2105
+
},
2106
+
{
2107
+
"a": "AE",
2108
+
"b": "T",
2109
+
"id": "AE_T",
2110
+
"weight": 0.9
2111
+
},
2112
+
{
2113
+
"a": "AH",
2114
+
"b": "B",
2115
+
"id": "AH_B",
2116
+
"weight": 0.9
2117
+
},
2118
+
{
2119
+
"a": "AH",
2120
+
"b": "D",
2121
+
"id": "AH_D",
2122
+
"weight": 0.9
2123
+
},
2124
+
{
2125
+
"a": "AH",
2126
+
"b": "G",
2127
+
"id": "AH_G",
2128
+
"weight": 0.9
2129
+
},
2130
+
{
2131
+
"a": "AH",
2132
+
"b": "K",
2133
+
"id": "AH_K",
2134
+
"weight": 0.9
2135
+
},
2136
+
{
2137
+
"a": "AH",
2138
+
"b": "P",
2139
+
"id": "AH_P",
2140
+
"weight": 0.9
2141
+
},
2142
+
{
2143
+
"a": "AH",
2144
+
"b": "T",
2145
+
"id": "AH_T",
2146
+
"weight": 0.9
2147
+
},
2148
+
{
2149
+
"a": "AO",
2150
+
"b": "B",
2151
+
"id": "AO_B",
2152
+
"weight": 0.9
2153
+
},
2154
+
{
2155
+
"a": "AO",
2156
+
"b": "D",
2157
+
"id": "AO_D",
2158
+
"weight": 0.9
2159
+
},
2160
+
{
2161
+
"a": "AO",
2162
+
"b": "G",
2163
+
"id": "AO_G",
2164
+
"weight": 0.9
2165
+
},
2166
+
{
2167
+
"a": "AO",
2168
+
"b": "K",
2169
+
"id": "AO_K",
2170
+
"weight": 0.9
2171
+
},
2172
+
{
2173
+
"a": "AO",
2174
+
"b": "P",
2175
+
"id": "AO_P",
2176
+
"weight": 0.9
2177
+
},
2178
+
{
2179
+
"a": "AO",
2180
+
"b": "T",
2181
+
"id": "AO_T",
2182
+
"weight": 0.9
2183
+
},
2184
+
{
2185
+
"a": "AW",
2186
+
"b": "B",
2187
+
"id": "AW_B",
2188
+
"weight": 0.9
2189
+
},
2190
+
{
2191
+
"a": "AW",
2192
+
"b": "D",
2193
+
"id": "AW_D",
2194
+
"weight": 0.9
2195
+
},
2196
+
{
2197
+
"a": "AW",
2198
+
"b": "G",
2199
+
"id": "AW_G",
2200
+
"weight": 0.9
2201
+
},
2202
+
{
2203
+
"a": "AW",
2204
+
"b": "K",
2205
+
"id": "AW_K",
2206
+
"weight": 0.9
2207
+
},
2208
+
{
2209
+
"a": "AW",
2210
+
"b": "P",
2211
+
"id": "AW_P",
2212
+
"weight": 0.9
2213
+
},
2214
+
{
2215
+
"a": "AW",
2216
+
"b": "T",
2217
+
"id": "AW_T",
2218
+
"weight": 0.9
2219
+
},
2220
+
{
2221
+
"a": "AY",
2222
+
"b": "B",
2223
+
"id": "AY_B",
2224
+
"weight": 0.9
2225
+
},
2226
+
{
2227
+
"a": "AY",
2228
+
"b": "D",
2229
+
"id": "AY_D",
2230
+
"weight": 0.9
2231
+
},
2232
+
{
2233
+
"a": "AY",
2234
+
"b": "G",
2235
+
"id": "AY_G",
2236
+
"weight": 0.9
2237
+
},
2238
+
{
2239
+
"a": "AY",
2240
+
"b": "K",
2241
+
"id": "AY_K",
2242
+
"weight": 0.9
2243
+
},
2244
+
{
2245
+
"a": "AY",
2246
+
"b": "P",
2247
+
"id": "AY_P",
2248
+
"weight": 0.9
2249
+
},
2250
+
{
2251
+
"a": "AY",
2252
+
"b": "T",
2253
+
"id": "AY_T",
2254
+
"weight": 0.9
2255
+
},
2256
+
{
2257
+
"a": "EH",
2258
+
"b": "B",
2259
+
"id": "EH_B",
2260
+
"weight": 0.9
2261
+
},
2262
+
{
2263
+
"a": "EH",
2264
+
"b": "D",
2265
+
"id": "EH_D",
2266
+
"weight": 0.9
2267
+
},
2268
+
{
2269
+
"a": "EH",
2270
+
"b": "G",
2271
+
"id": "EH_G",
2272
+
"weight": 0.9
2273
+
},
2274
+
{
2275
+
"a": "EH",
2276
+
"b": "K",
2277
+
"id": "EH_K",
2278
+
"weight": 0.9
2279
+
},
2280
+
{
2281
+
"a": "EH",
2282
+
"b": "P",
2283
+
"id": "EH_P",
2284
+
"weight": 0.9
2285
+
},
2286
+
{
2287
+
"a": "EH",
2288
+
"b": "T",
2289
+
"id": "EH_T",
2290
+
"weight": 0.9
2291
+
},
2292
+
{
2293
+
"a": "ER",
2294
+
"b": "B",
2295
+
"id": "ER_B",
2296
+
"weight": 0.9
2297
+
},
2298
+
{
2299
+
"a": "ER",
2300
+
"b": "D",
2301
+
"id": "ER_D",
2302
+
"weight": 0.9
2303
+
},
2304
+
{
2305
+
"a": "ER",
2306
+
"b": "G",
2307
+
"id": "ER_G",
2308
+
"weight": 0.9
2309
+
},
2310
+
{
2311
+
"a": "ER",
2312
+
"b": "K",
2313
+
"id": "ER_K",
2314
+
"weight": 0.9
2315
+
},
2316
+
{
2317
+
"a": "ER",
2318
+
"b": "P",
2319
+
"id": "ER_P",
2320
+
"weight": 0.9
2321
+
},
2322
+
{
2323
+
"a": "ER",
2324
+
"b": "T",
2325
+
"id": "ER_T",
2326
+
"weight": 0.9
2327
+
},
2328
+
{
2329
+
"a": "EY",
2330
+
"b": "B",
2331
+
"id": "EY_B",
2332
+
"weight": 0.9
2333
+
},
2334
+
{
2335
+
"a": "EY",
2336
+
"b": "D",
2337
+
"id": "EY_D",
2338
+
"weight": 0.9
2339
+
},
2340
+
{
2341
+
"a": "EY",
2342
+
"b": "G",
2343
+
"id": "EY_G",
2344
+
"weight": 0.9
2345
+
},
2346
+
{
2347
+
"a": "EY",
2348
+
"b": "K",
2349
+
"id": "EY_K",
2350
+
"weight": 0.9
2351
+
},
2352
+
{
2353
+
"a": "EY",
2354
+
"b": "P",
2355
+
"id": "EY_P",
2356
+
"weight": 0.9
2357
+
},
2358
+
{
2359
+
"a": "EY",
2360
+
"b": "T",
2361
+
"id": "EY_T",
2362
+
"weight": 0.9
2363
+
},
2364
+
{
2365
+
"a": "IH",
2366
+
"b": "B",
2367
+
"id": "IH_B",
2368
+
"weight": 0.9
2369
+
},
2370
+
{
2371
+
"a": "IH",
2372
+
"b": "D",
2373
+
"id": "IH_D",
2374
+
"weight": 0.9
2375
+
},
2376
+
{
2377
+
"a": "IH",
2378
+
"b": "G",
2379
+
"id": "IH_G",
2380
+
"weight": 0.9
2381
+
},
2382
+
{
2383
+
"a": "IH",
2384
+
"b": "K",
2385
+
"id": "IH_K",
2386
+
"weight": 0.9
2387
+
},
2388
+
{
2389
+
"a": "IH",
2390
+
"b": "P",
2391
+
"id": "IH_P",
2392
+
"weight": 0.9
2393
+
},
2394
+
{
2395
+
"a": "IH",
2396
+
"b": "T",
2397
+
"id": "IH_T",
2398
+
"weight": 0.9
2399
+
},
2400
+
{
2401
+
"a": "IY",
2402
+
"b": "B",
2403
+
"id": "IY_B",
2404
+
"weight": 0.9
2405
+
},
2406
+
{
2407
+
"a": "IY",
2408
+
"b": "D",
2409
+
"id": "IY_D",
2410
+
"weight": 0.9
2411
+
},
2412
+
{
2413
+
"a": "IY",
2414
+
"b": "G",
2415
+
"id": "IY_G",
2416
+
"weight": 0.9
2417
+
},
2418
+
{
2419
+
"a": "IY",
2420
+
"b": "K",
2421
+
"id": "IY_K",
2422
+
"weight": 0.9
2423
+
},
2424
+
{
2425
+
"a": "IY",
2426
+
"b": "P",
2427
+
"id": "IY_P",
2428
+
"weight": 0.9
2429
+
},
2430
+
{
2431
+
"a": "IY",
2432
+
"b": "T",
2433
+
"id": "IY_T",
2434
+
"weight": 0.9
2435
+
},
2436
+
{
2437
+
"a": "OW",
2438
+
"b": "B",
2439
+
"id": "OW_B",
2440
+
"weight": 0.9
2441
+
},
2442
+
{
2443
+
"a": "OW",
2444
+
"b": "D",
2445
+
"id": "OW_D",
2446
+
"weight": 0.9
2447
+
},
2448
+
{
2449
+
"a": "OW",
2450
+
"b": "G",
2451
+
"id": "OW_G",
2452
+
"weight": 0.9
2453
+
},
2454
+
{
2455
+
"a": "OW",
2456
+
"b": "K",
2457
+
"id": "OW_K",
2458
+
"weight": 0.9
2459
+
},
2460
+
{
2461
+
"a": "OW",
2462
+
"b": "P",
2463
+
"id": "OW_P",
2464
+
"weight": 0.9
2465
+
},
2466
+
{
2467
+
"a": "OW",
2468
+
"b": "T",
2469
+
"id": "OW_T",
2470
+
"weight": 0.9
2471
+
},
2472
+
{
2473
+
"a": "OY",
2474
+
"b": "B",
2475
+
"id": "OY_B",
2476
+
"weight": 0.9
2477
+
},
2478
+
{
2479
+
"a": "OY",
2480
+
"b": "D",
2481
+
"id": "OY_D",
2482
+
"weight": 0.9
2483
+
},
2484
+
{
2485
+
"a": "OY",
2486
+
"b": "G",
2487
+
"id": "OY_G",
2488
+
"weight": 0.9
2489
+
},
2490
+
{
2491
+
"a": "OY",
2492
+
"b": "K",
2493
+
"id": "OY_K",
2494
+
"weight": 0.9
2495
+
},
2496
+
{
2497
+
"a": "OY",
2498
+
"b": "P",
2499
+
"id": "OY_P",
2500
+
"weight": 0.9
2501
+
},
2502
+
{
2503
+
"a": "OY",
2504
+
"b": "T",
2505
+
"id": "OY_T",
2506
+
"weight": 0.9
2507
+
},
2508
+
{
2509
+
"a": "UH",
2510
+
"b": "B",
2511
+
"id": "UH_B",
2512
+
"weight": 0.9
2513
+
},
2514
+
{
2515
+
"a": "UH",
2516
+
"b": "D",
2517
+
"id": "UH_D",
2518
+
"weight": 0.9
2519
+
},
2520
+
{
2521
+
"a": "UH",
2522
+
"b": "G",
2523
+
"id": "UH_G",
2524
+
"weight": 0.9
2525
+
},
2526
+
{
2527
+
"a": "UH",
2528
+
"b": "K",
2529
+
"id": "UH_K",
2530
+
"weight": 0.9
2531
+
},
2532
+
{
2533
+
"a": "UH",
2534
+
"b": "P",
2535
+
"id": "UH_P",
2536
+
"weight": 0.9
2537
+
},
2538
+
{
2539
+
"a": "UH",
2540
+
"b": "T",
2541
+
"id": "UH_T",
2542
+
"weight": 0.9
2543
+
},
2544
+
{
2545
+
"a": "UW",
2546
+
"b": "B",
2547
+
"id": "UW_B",
2548
+
"weight": 0.9
2549
+
},
2550
+
{
2551
+
"a": "UW",
2552
+
"b": "D",
2553
+
"id": "UW_D",
2554
+
"weight": 0.9
2555
+
},
2556
+
{
2557
+
"a": "UW",
2558
+
"b": "G",
2559
+
"id": "UW_G",
2560
+
"weight": 0.9
2561
+
},
2562
+
{
2563
+
"a": "UW",
2564
+
"b": "K",
2565
+
"id": "UW_K",
2566
+
"weight": 0.9
2567
+
},
2568
+
{
2569
+
"a": "UW",
2570
+
"b": "P",
2571
+
"id": "UW_P",
2572
+
"weight": 0.9
2573
+
},
2574
+
{
2575
+
"a": "UW",
2576
+
"b": "T",
2577
+
"id": "UW_T",
2578
+
"weight": 0.9
2579
+
},
2580
+
{
2581
+
"a": "#B",
2582
+
"b": "B",
2583
+
"id": "#B_B",
2584
+
"weight": 0.85
2585
+
},
2586
+
{
2587
+
"a": "#B",
2588
+
"b": "CH",
2589
+
"id": "#B_CH",
2590
+
"weight": 0.85
2591
+
},
2592
+
{
2593
+
"a": "#B",
2594
+
"b": "D",
2595
+
"id": "#B_D",
2596
+
"weight": 0.85
2597
+
},
2598
+
{
2599
+
"a": "#B",
2600
+
"b": "DH",
2601
+
"id": "#B_DH",
2602
+
"weight": 0.85
2603
+
},
2604
+
{
2605
+
"a": "#B",
2606
+
"b": "F",
2607
+
"id": "#B_F",
2608
+
"weight": 0.85
2609
+
},
2610
+
{
2611
+
"a": "#B",
2612
+
"b": "G",
2613
+
"id": "#B_G",
2614
+
"weight": 0.85
2615
+
},
2616
+
{
2617
+
"a": "#B",
2618
+
"b": "HH",
2619
+
"id": "#B_HH",
2620
+
"weight": 0.85
2621
+
},
2622
+
{
2623
+
"a": "#B",
2624
+
"b": "JH",
2625
+
"id": "#B_JH",
2626
+
"weight": 0.85
2627
+
},
2628
+
{
2629
+
"a": "#B",
2630
+
"b": "K",
2631
+
"id": "#B_K",
2632
+
"weight": 0.85
2633
+
},
2634
+
{
2635
+
"a": "#B",
2636
+
"b": "L",
2637
+
"id": "#B_L",
2638
+
"weight": 0.85
2639
+
},
2640
+
{
2641
+
"a": "#B",
2642
+
"b": "M",
2643
+
"id": "#B_M",
2644
+
"weight": 0.85
2645
+
},
2646
+
{
2647
+
"a": "#B",
2648
+
"b": "N",
2649
+
"id": "#B_N",
2650
+
"weight": 0.85
2651
+
},
2652
+
{
2653
+
"a": "#B",
2654
+
"b": "P",
2655
+
"id": "#B_P",
2656
+
"weight": 0.85
2657
+
},
2658
+
{
2659
+
"a": "#B",
2660
+
"b": "R",
2661
+
"id": "#B_R",
2662
+
"weight": 0.85
2663
+
},
2664
+
{
2665
+
"a": "#B",
2666
+
"b": "S",
2667
+
"id": "#B_S",
2668
+
"weight": 0.85
2669
+
},
2670
+
{
2671
+
"a": "#B",
2672
+
"b": "SH",
2673
+
"id": "#B_SH",
2674
+
"weight": 0.85
2675
+
},
2676
+
{
2677
+
"a": "#B",
2678
+
"b": "T",
2679
+
"id": "#B_T",
2680
+
"weight": 0.85
2681
+
},
2682
+
{
2683
+
"a": "#B",
2684
+
"b": "TH",
2685
+
"id": "#B_TH",
2686
+
"weight": 0.85
2687
+
},
2688
+
{
2689
+
"a": "#B",
2690
+
"b": "V",
2691
+
"id": "#B_V",
2692
+
"weight": 0.85
2693
+
},
2694
+
{
2695
+
"a": "#B",
2696
+
"b": "W",
2697
+
"id": "#B_W",
2698
+
"weight": 0.85
2699
+
},
2700
+
{
2701
+
"a": "#B",
2702
+
"b": "Y",
2703
+
"id": "#B_Y",
2704
+
"weight": 0.85
2705
+
},
2706
+
{
2707
+
"a": "#B",
2708
+
"b": "Z",
2709
+
"id": "#B_Z",
2710
+
"weight": 0.85
2711
+
},
2712
+
{
2713
+
"a": "AA",
2714
+
"b": "M",
2715
+
"id": "AA_M",
2716
+
"weight": 0.85
2717
+
},
2718
+
{
2719
+
"a": "AA",
2720
+
"b": "N",
2721
+
"id": "AA_N",
2722
+
"weight": 0.85
2723
+
},
2724
+
{
2725
+
"a": "AA",
2726
+
"b": "NG",
2727
+
"id": "AA_NG",
2728
+
"weight": 0.85
2729
+
},
2730
+
{
2731
+
"a": "AE",
2732
+
"b": "M",
2733
+
"id": "AE_M",
2734
+
"weight": 0.85
2735
+
},
2736
+
{
2737
+
"a": "AE",
2738
+
"b": "N",
2739
+
"id": "AE_N",
2740
+
"weight": 0.85
2741
+
},
2742
+
{
2743
+
"a": "AE",
2744
+
"b": "NG",
2745
+
"id": "AE_NG",
2746
+
"weight": 0.85
2747
+
},
2748
+
{
2749
+
"a": "AH",
2750
+
"b": "M",
2751
+
"id": "AH_M",
2752
+
"weight": 0.85
2753
+
},
2754
+
{
2755
+
"a": "AH",
2756
+
"b": "N",
2757
+
"id": "AH_N",
2758
+
"weight": 0.85
2759
+
},
2760
+
{
2761
+
"a": "AH",
2762
+
"b": "NG",
2763
+
"id": "AH_NG",
2764
+
"weight": 0.85
2765
+
},
2766
+
{
2767
+
"a": "AO",
2768
+
"b": "M",
2769
+
"id": "AO_M",
2770
+
"weight": 0.85
2771
+
},
2772
+
{
2773
+
"a": "AO",
2774
+
"b": "N",
2775
+
"id": "AO_N",
2776
+
"weight": 0.85
2777
+
},
2778
+
{
2779
+
"a": "AO",
2780
+
"b": "NG",
2781
+
"id": "AO_NG",
2782
+
"weight": 0.85
2783
+
},
2784
+
{
2785
+
"a": "AW",
2786
+
"b": "M",
2787
+
"id": "AW_M",
2788
+
"weight": 0.85
2789
+
},
2790
+
{
2791
+
"a": "AW",
2792
+
"b": "N",
2793
+
"id": "AW_N",
2794
+
"weight": 0.85
2795
+
},
2796
+
{
2797
+
"a": "AW",
2798
+
"b": "NG",
2799
+
"id": "AW_NG",
2800
+
"weight": 0.85
2801
+
},
2802
+
{
2803
+
"a": "AY",
2804
+
"b": "M",
2805
+
"id": "AY_M",
2806
+
"weight": 0.85
2807
+
},
2808
+
{
2809
+
"a": "AY",
2810
+
"b": "N",
2811
+
"id": "AY_N",
2812
+
"weight": 0.85
2813
+
},
2814
+
{
2815
+
"a": "AY",
2816
+
"b": "NG",
2817
+
"id": "AY_NG",
2818
+
"weight": 0.85
2819
+
},
2820
+
{
2821
+
"a": "EH",
2822
+
"b": "M",
2823
+
"id": "EH_M",
2824
+
"weight": 0.85
2825
+
},
2826
+
{
2827
+
"a": "EH",
2828
+
"b": "N",
2829
+
"id": "EH_N",
2830
+
"weight": 0.85
2831
+
},
2832
+
{
2833
+
"a": "EH",
2834
+
"b": "NG",
2835
+
"id": "EH_NG",
2836
+
"weight": 0.85
2837
+
},
2838
+
{
2839
+
"a": "ER",
2840
+
"b": "M",
2841
+
"id": "ER_M",
2842
+
"weight": 0.85
2843
+
},
2844
+
{
2845
+
"a": "ER",
2846
+
"b": "N",
2847
+
"id": "ER_N",
2848
+
"weight": 0.85
2849
+
},
2850
+
{
2851
+
"a": "ER",
2852
+
"b": "NG",
2853
+
"id": "ER_NG",
2854
+
"weight": 0.85
2855
+
},
2856
+
{
2857
+
"a": "EY",
2858
+
"b": "M",
2859
+
"id": "EY_M",
2860
+
"weight": 0.85
2861
+
},
2862
+
{
2863
+
"a": "EY",
2864
+
"b": "N",
2865
+
"id": "EY_N",
2866
+
"weight": 0.85
2867
+
},
2868
+
{
2869
+
"a": "EY",
2870
+
"b": "NG",
2871
+
"id": "EY_NG",
2872
+
"weight": 0.85
2873
+
},
2874
+
{
2875
+
"a": "IH",
2876
+
"b": "M",
2877
+
"id": "IH_M",
2878
+
"weight": 0.85
2879
+
},
2880
+
{
2881
+
"a": "IH",
2882
+
"b": "N",
2883
+
"id": "IH_N",
2884
+
"weight": 0.85
2885
+
},
2886
+
{
2887
+
"a": "IH",
2888
+
"b": "NG",
2889
+
"id": "IH_NG",
2890
+
"weight": 0.85
2891
+
},
2892
+
{
2893
+
"a": "IY",
2894
+
"b": "M",
2895
+
"id": "IY_M",
2896
+
"weight": 0.85
2897
+
},
2898
+
{
2899
+
"a": "IY",
2900
+
"b": "N",
2901
+
"id": "IY_N",
2902
+
"weight": 0.85
2903
+
},
2904
+
{
2905
+
"a": "IY",
2906
+
"b": "NG",
2907
+
"id": "IY_NG",
2908
+
"weight": 0.85
2909
+
},
2910
+
{
2911
+
"a": "M",
2912
+
"b": "AA",
2913
+
"id": "M_AA",
2914
+
"weight": 0.85
2915
+
},
2916
+
{
2917
+
"a": "M",
2918
+
"b": "AE",
2919
+
"id": "M_AE",
2920
+
"weight": 0.85
2921
+
},
2922
+
{
2923
+
"a": "M",
2924
+
"b": "AH",
2925
+
"id": "M_AH",
2926
+
"weight": 0.85
2927
+
},
2928
+
{
2929
+
"a": "M",
2930
+
"b": "AO",
2931
+
"id": "M_AO",
2932
+
"weight": 0.85
2933
+
},
2934
+
{
2935
+
"a": "M",
2936
+
"b": "AW",
2937
+
"id": "M_AW",
2938
+
"weight": 0.85
2939
+
},
2940
+
{
2941
+
"a": "M",
2942
+
"b": "AY",
2943
+
"id": "M_AY",
2944
+
"weight": 0.85
2945
+
},
2946
+
{
2947
+
"a": "M",
2948
+
"b": "EH",
2949
+
"id": "M_EH",
2950
+
"weight": 0.85
2951
+
},
2952
+
{
2953
+
"a": "M",
2954
+
"b": "ER",
2955
+
"id": "M_ER",
2956
+
"weight": 0.85
2957
+
},
2958
+
{
2959
+
"a": "M",
2960
+
"b": "EY",
2961
+
"id": "M_EY",
2962
+
"weight": 0.85
2963
+
},
2964
+
{
2965
+
"a": "M",
2966
+
"b": "IH",
2967
+
"id": "M_IH",
2968
+
"weight": 0.85
2969
+
},
2970
+
{
2971
+
"a": "M",
2972
+
"b": "IY",
2973
+
"id": "M_IY",
2974
+
"weight": 0.85
2975
+
},
2976
+
{
2977
+
"a": "M",
2978
+
"b": "OW",
2979
+
"id": "M_OW",
2980
+
"weight": 0.85
2981
+
},
2982
+
{
2983
+
"a": "M",
2984
+
"b": "OY",
2985
+
"id": "M_OY",
2986
+
"weight": 0.85
2987
+
},
2988
+
{
2989
+
"a": "M",
2990
+
"b": "UH",
2991
+
"id": "M_UH",
2992
+
"weight": 0.85
2993
+
},
2994
+
{
2995
+
"a": "M",
2996
+
"b": "UW",
2997
+
"id": "M_UW",
2998
+
"weight": 0.85
2999
+
},
3000
+
{
3001
+
"a": "N",
3002
+
"b": "AA",
3003
+
"id": "N_AA",
3004
+
"weight": 0.85
3005
+
},
3006
+
{
3007
+
"a": "N",
3008
+
"b": "AE",
3009
+
"id": "N_AE",
3010
+
"weight": 0.85
3011
+
},
3012
+
{
3013
+
"a": "N",
3014
+
"b": "AH",
3015
+
"id": "N_AH",
3016
+
"weight": 0.85
3017
+
},
3018
+
{
3019
+
"a": "N",
3020
+
"b": "AO",
3021
+
"id": "N_AO",
3022
+
"weight": 0.85
3023
+
},
3024
+
{
3025
+
"a": "N",
3026
+
"b": "AW",
3027
+
"id": "N_AW",
3028
+
"weight": 0.85
3029
+
},
3030
+
{
3031
+
"a": "N",
3032
+
"b": "AY",
3033
+
"id": "N_AY",
3034
+
"weight": 0.85
3035
+
},
3036
+
{
3037
+
"a": "N",
3038
+
"b": "EH",
3039
+
"id": "N_EH",
3040
+
"weight": 0.85
3041
+
},
3042
+
{
3043
+
"a": "N",
3044
+
"b": "ER",
3045
+
"id": "N_ER",
3046
+
"weight": 0.85
3047
+
},
3048
+
{
3049
+
"a": "N",
3050
+
"b": "EY",
3051
+
"id": "N_EY",
3052
+
"weight": 0.85
3053
+
},
3054
+
{
3055
+
"a": "N",
3056
+
"b": "IH",
3057
+
"id": "N_IH",
3058
+
"weight": 0.85
3059
+
},
3060
+
{
3061
+
"a": "N",
3062
+
"b": "IY",
3063
+
"id": "N_IY",
3064
+
"weight": 0.85
3065
+
},
3066
+
{
3067
+
"a": "N",
3068
+
"b": "OW",
3069
+
"id": "N_OW",
3070
+
"weight": 0.85
3071
+
},
3072
+
{
3073
+
"a": "N",
3074
+
"b": "OY",
3075
+
"id": "N_OY",
3076
+
"weight": 0.85
3077
+
},
3078
+
{
3079
+
"a": "N",
3080
+
"b": "UH",
3081
+
"id": "N_UH",
3082
+
"weight": 0.85
3083
+
},
3084
+
{
3085
+
"a": "N",
3086
+
"b": "UW",
3087
+
"id": "N_UW",
3088
+
"weight": 0.85
3089
+
},
3090
+
{
3091
+
"a": "NG",
3092
+
"b": "AA",
3093
+
"id": "NG_AA",
3094
+
"weight": 0.85
3095
+
},
3096
+
{
3097
+
"a": "NG",
3098
+
"b": "AE",
3099
+
"id": "NG_AE",
3100
+
"weight": 0.85
3101
+
},
3102
+
{
3103
+
"a": "NG",
3104
+
"b": "AH",
3105
+
"id": "NG_AH",
3106
+
"weight": 0.85
3107
+
},
3108
+
{
3109
+
"a": "NG",
3110
+
"b": "AO",
3111
+
"id": "NG_AO",
3112
+
"weight": 0.85
3113
+
},
3114
+
{
3115
+
"a": "NG",
3116
+
"b": "AW",
3117
+
"id": "NG_AW",
3118
+
"weight": 0.85
3119
+
},
3120
+
{
3121
+
"a": "NG",
3122
+
"b": "AY",
3123
+
"id": "NG_AY",
3124
+
"weight": 0.85
3125
+
},
3126
+
{
3127
+
"a": "NG",
3128
+
"b": "EH",
3129
+
"id": "NG_EH",
3130
+
"weight": 0.85
3131
+
},
3132
+
{
3133
+
"a": "NG",
3134
+
"b": "ER",
3135
+
"id": "NG_ER",
3136
+
"weight": 0.85
3137
+
},
3138
+
{
3139
+
"a": "NG",
3140
+
"b": "EY",
3141
+
"id": "NG_EY",
3142
+
"weight": 0.85
3143
+
},
3144
+
{
3145
+
"a": "NG",
3146
+
"b": "IH",
3147
+
"id": "NG_IH",
3148
+
"weight": 0.85
3149
+
},
3150
+
{
3151
+
"a": "NG",
3152
+
"b": "IY",
3153
+
"id": "NG_IY",
3154
+
"weight": 0.85
3155
+
},
3156
+
{
3157
+
"a": "NG",
3158
+
"b": "OW",
3159
+
"id": "NG_OW",
3160
+
"weight": 0.85
3161
+
},
3162
+
{
3163
+
"a": "NG",
3164
+
"b": "OY",
3165
+
"id": "NG_OY",
3166
+
"weight": 0.85
3167
+
},
3168
+
{
3169
+
"a": "NG",
3170
+
"b": "UH",
3171
+
"id": "NG_UH",
3172
+
"weight": 0.85
3173
+
},
3174
+
{
3175
+
"a": "NG",
3176
+
"b": "UW",
3177
+
"id": "NG_UW",
3178
+
"weight": 0.85
3179
+
},
3180
+
{
3181
+
"a": "OW",
3182
+
"b": "M",
3183
+
"id": "OW_M",
3184
+
"weight": 0.85
3185
+
},
3186
+
{
3187
+
"a": "OW",
3188
+
"b": "N",
3189
+
"id": "OW_N",
3190
+
"weight": 0.85
3191
+
},
3192
+
{
3193
+
"a": "OW",
3194
+
"b": "NG",
3195
+
"id": "OW_NG",
3196
+
"weight": 0.85
3197
+
},
3198
+
{
3199
+
"a": "OY",
3200
+
"b": "M",
3201
+
"id": "OY_M",
3202
+
"weight": 0.85
3203
+
},
3204
+
{
3205
+
"a": "OY",
3206
+
"b": "N",
3207
+
"id": "OY_N",
3208
+
"weight": 0.85
3209
+
},
3210
+
{
3211
+
"a": "OY",
3212
+
"b": "NG",
3213
+
"id": "OY_NG",
3214
+
"weight": 0.85
3215
+
},
3216
+
{
3217
+
"a": "UH",
3218
+
"b": "M",
3219
+
"id": "UH_M",
3220
+
"weight": 0.85
3221
+
},
3222
+
{
3223
+
"a": "UH",
3224
+
"b": "N",
3225
+
"id": "UH_N",
3226
+
"weight": 0.85
3227
+
},
3228
+
{
3229
+
"a": "UH",
3230
+
"b": "NG",
3231
+
"id": "UH_NG",
3232
+
"weight": 0.85
3233
+
},
3234
+
{
3235
+
"a": "UW",
3236
+
"b": "M",
3237
+
"id": "UW_M",
3238
+
"weight": 0.85
3239
+
},
3240
+
{
3241
+
"a": "UW",
3242
+
"b": "N",
3243
+
"id": "UW_N",
3244
+
"weight": 0.85
3245
+
},
3246
+
{
3247
+
"a": "UW",
3248
+
"b": "NG",
3249
+
"id": "UW_NG",
3250
+
"weight": 0.85
3251
+
},
3252
+
{
3253
+
"a": "AA",
3254
+
"b": "DH",
3255
+
"id": "AA_DH",
3256
+
"weight": 0.8
3257
+
},
3258
+
{
3259
+
"a": "AA",
3260
+
"b": "F",
3261
+
"id": "AA_F",
3262
+
"weight": 0.8
3263
+
},
3264
+
{
3265
+
"a": "AA",
3266
+
"b": "HH",
3267
+
"id": "AA_HH",
3268
+
"weight": 0.8
3269
+
},
3270
+
{
3271
+
"a": "AA",
3272
+
"b": "S",
3273
+
"id": "AA_S",
3274
+
"weight": 0.8
3275
+
},
3276
+
{
3277
+
"a": "AA",
3278
+
"b": "SH",
3279
+
"id": "AA_SH",
3280
+
"weight": 0.8
3281
+
},
3282
+
{
3283
+
"a": "AA",
3284
+
"b": "TH",
3285
+
"id": "AA_TH",
3286
+
"weight": 0.8
3287
+
},
3288
+
{
3289
+
"a": "AA",
3290
+
"b": "V",
3291
+
"id": "AA_V",
3292
+
"weight": 0.8
3293
+
},
3294
+
{
3295
+
"a": "AA",
3296
+
"b": "Z",
3297
+
"id": "AA_Z",
3298
+
"weight": 0.8
3299
+
},
3300
+
{
3301
+
"a": "AA",
3302
+
"b": "ZH",
3303
+
"id": "AA_ZH",
3304
+
"weight": 0.8
3305
+
},
3306
+
{
3307
+
"a": "AE",
3308
+
"b": "DH",
3309
+
"id": "AE_DH",
3310
+
"weight": 0.8
3311
+
},
3312
+
{
3313
+
"a": "AE",
3314
+
"b": "F",
3315
+
"id": "AE_F",
3316
+
"weight": 0.8
3317
+
},
3318
+
{
3319
+
"a": "AE",
3320
+
"b": "HH",
3321
+
"id": "AE_HH",
3322
+
"weight": 0.8
3323
+
},
3324
+
{
3325
+
"a": "AE",
3326
+
"b": "S",
3327
+
"id": "AE_S",
3328
+
"weight": 0.8
3329
+
},
3330
+
{
3331
+
"a": "AE",
3332
+
"b": "SH",
3333
+
"id": "AE_SH",
3334
+
"weight": 0.8
3335
+
},
3336
+
{
3337
+
"a": "AE",
3338
+
"b": "TH",
3339
+
"id": "AE_TH",
3340
+
"weight": 0.8
3341
+
},
3342
+
{
3343
+
"a": "AE",
3344
+
"b": "V",
3345
+
"id": "AE_V",
3346
+
"weight": 0.8
3347
+
},
3348
+
{
3349
+
"a": "AE",
3350
+
"b": "Z",
3351
+
"id": "AE_Z",
3352
+
"weight": 0.8
3353
+
},
3354
+
{
3355
+
"a": "AE",
3356
+
"b": "ZH",
3357
+
"id": "AE_ZH",
3358
+
"weight": 0.8
3359
+
},
3360
+
{
3361
+
"a": "AH",
3362
+
"b": "DH",
3363
+
"id": "AH_DH",
3364
+
"weight": 0.8
3365
+
},
3366
+
{
3367
+
"a": "AH",
3368
+
"b": "F",
3369
+
"id": "AH_F",
3370
+
"weight": 0.8
3371
+
},
3372
+
{
3373
+
"a": "AH",
3374
+
"b": "HH",
3375
+
"id": "AH_HH",
3376
+
"weight": 0.8
3377
+
},
3378
+
{
3379
+
"a": "AH",
3380
+
"b": "S",
3381
+
"id": "AH_S",
3382
+
"weight": 0.8
3383
+
},
3384
+
{
3385
+
"a": "AH",
3386
+
"b": "SH",
3387
+
"id": "AH_SH",
3388
+
"weight": 0.8
3389
+
},
3390
+
{
3391
+
"a": "AH",
3392
+
"b": "TH",
3393
+
"id": "AH_TH",
3394
+
"weight": 0.8
3395
+
},
3396
+
{
3397
+
"a": "AH",
3398
+
"b": "V",
3399
+
"id": "AH_V",
3400
+
"weight": 0.8
3401
+
},
3402
+
{
3403
+
"a": "AH",
3404
+
"b": "Z",
3405
+
"id": "AH_Z",
3406
+
"weight": 0.8
3407
+
},
3408
+
{
3409
+
"a": "AH",
3410
+
"b": "ZH",
3411
+
"id": "AH_ZH",
3412
+
"weight": 0.8
3413
+
},
3414
+
{
3415
+
"a": "AO",
3416
+
"b": "DH",
3417
+
"id": "AO_DH",
3418
+
"weight": 0.8
3419
+
},
3420
+
{
3421
+
"a": "AO",
3422
+
"b": "F",
3423
+
"id": "AO_F",
3424
+
"weight": 0.8
3425
+
},
3426
+
{
3427
+
"a": "AO",
3428
+
"b": "HH",
3429
+
"id": "AO_HH",
3430
+
"weight": 0.8
3431
+
},
3432
+
{
3433
+
"a": "AO",
3434
+
"b": "S",
3435
+
"id": "AO_S",
3436
+
"weight": 0.8
3437
+
},
3438
+
{
3439
+
"a": "AO",
3440
+
"b": "SH",
3441
+
"id": "AO_SH",
3442
+
"weight": 0.8
3443
+
},
3444
+
{
3445
+
"a": "AO",
3446
+
"b": "TH",
3447
+
"id": "AO_TH",
3448
+
"weight": 0.8
3449
+
},
3450
+
{
3451
+
"a": "AO",
3452
+
"b": "V",
3453
+
"id": "AO_V",
3454
+
"weight": 0.8
3455
+
},
3456
+
{
3457
+
"a": "AO",
3458
+
"b": "Z",
3459
+
"id": "AO_Z",
3460
+
"weight": 0.8
3461
+
},
3462
+
{
3463
+
"a": "AO",
3464
+
"b": "ZH",
3465
+
"id": "AO_ZH",
3466
+
"weight": 0.8
3467
+
},
3468
+
{
3469
+
"a": "AW",
3470
+
"b": "DH",
3471
+
"id": "AW_DH",
3472
+
"weight": 0.8
3473
+
},
3474
+
{
3475
+
"a": "AW",
3476
+
"b": "F",
3477
+
"id": "AW_F",
3478
+
"weight": 0.8
3479
+
},
3480
+
{
3481
+
"a": "AW",
3482
+
"b": "HH",
3483
+
"id": "AW_HH",
3484
+
"weight": 0.8
3485
+
},
3486
+
{
3487
+
"a": "AW",
3488
+
"b": "S",
3489
+
"id": "AW_S",
3490
+
"weight": 0.8
3491
+
},
3492
+
{
3493
+
"a": "AW",
3494
+
"b": "SH",
3495
+
"id": "AW_SH",
3496
+
"weight": 0.8
3497
+
},
3498
+
{
3499
+
"a": "AW",
3500
+
"b": "TH",
3501
+
"id": "AW_TH",
3502
+
"weight": 0.8
3503
+
},
3504
+
{
3505
+
"a": "AW",
3506
+
"b": "V",
3507
+
"id": "AW_V",
3508
+
"weight": 0.8
3509
+
},
3510
+
{
3511
+
"a": "AW",
3512
+
"b": "Z",
3513
+
"id": "AW_Z",
3514
+
"weight": 0.8
3515
+
},
3516
+
{
3517
+
"a": "AW",
3518
+
"b": "ZH",
3519
+
"id": "AW_ZH",
3520
+
"weight": 0.8
3521
+
},
3522
+
{
3523
+
"a": "AY",
3524
+
"b": "DH",
3525
+
"id": "AY_DH",
3526
+
"weight": 0.8
3527
+
},
3528
+
{
3529
+
"a": "AY",
3530
+
"b": "F",
3531
+
"id": "AY_F",
3532
+
"weight": 0.8
3533
+
},
3534
+
{
3535
+
"a": "AY",
3536
+
"b": "HH",
3537
+
"id": "AY_HH",
3538
+
"weight": 0.8
3539
+
},
3540
+
{
3541
+
"a": "AY",
3542
+
"b": "S",
3543
+
"id": "AY_S",
3544
+
"weight": 0.8
3545
+
},
3546
+
{
3547
+
"a": "AY",
3548
+
"b": "SH",
3549
+
"id": "AY_SH",
3550
+
"weight": 0.8
3551
+
},
3552
+
{
3553
+
"a": "AY",
3554
+
"b": "TH",
3555
+
"id": "AY_TH",
3556
+
"weight": 0.8
3557
+
},
3558
+
{
3559
+
"a": "AY",
3560
+
"b": "V",
3561
+
"id": "AY_V",
3562
+
"weight": 0.8
3563
+
},
3564
+
{
3565
+
"a": "AY",
3566
+
"b": "Z",
3567
+
"id": "AY_Z",
3568
+
"weight": 0.8
3569
+
},
3570
+
{
3571
+
"a": "AY",
3572
+
"b": "ZH",
3573
+
"id": "AY_ZH",
3574
+
"weight": 0.8
3575
+
},
3576
+
{
3577
+
"a": "DH",
3578
+
"b": "AA",
3579
+
"id": "DH_AA",
3580
+
"weight": 0.8
3581
+
},
3582
+
{
3583
+
"a": "DH",
3584
+
"b": "AE",
3585
+
"id": "DH_AE",
3586
+
"weight": 0.8
3587
+
},
3588
+
{
3589
+
"a": "DH",
3590
+
"b": "AH",
3591
+
"id": "DH_AH",
3592
+
"weight": 0.8
3593
+
},
3594
+
{
3595
+
"a": "DH",
3596
+
"b": "AO",
3597
+
"id": "DH_AO",
3598
+
"weight": 0.8
3599
+
},
3600
+
{
3601
+
"a": "DH",
3602
+
"b": "AW",
3603
+
"id": "DH_AW",
3604
+
"weight": 0.8
3605
+
},
3606
+
{
3607
+
"a": "DH",
3608
+
"b": "AY",
3609
+
"id": "DH_AY",
3610
+
"weight": 0.8
3611
+
},
3612
+
{
3613
+
"a": "DH",
3614
+
"b": "EH",
3615
+
"id": "DH_EH",
3616
+
"weight": 0.8
3617
+
},
3618
+
{
3619
+
"a": "DH",
3620
+
"b": "ER",
3621
+
"id": "DH_ER",
3622
+
"weight": 0.8
3623
+
},
3624
+
{
3625
+
"a": "DH",
3626
+
"b": "EY",
3627
+
"id": "DH_EY",
3628
+
"weight": 0.8
3629
+
},
3630
+
{
3631
+
"a": "DH",
3632
+
"b": "IH",
3633
+
"id": "DH_IH",
3634
+
"weight": 0.8
3635
+
},
3636
+
{
3637
+
"a": "DH",
3638
+
"b": "IY",
3639
+
"id": "DH_IY",
3640
+
"weight": 0.8
3641
+
},
3642
+
{
3643
+
"a": "DH",
3644
+
"b": "OW",
3645
+
"id": "DH_OW",
3646
+
"weight": 0.8
3647
+
},
3648
+
{
3649
+
"a": "DH",
3650
+
"b": "OY",
3651
+
"id": "DH_OY",
3652
+
"weight": 0.8
3653
+
},
3654
+
{
3655
+
"a": "DH",
3656
+
"b": "UH",
3657
+
"id": "DH_UH",
3658
+
"weight": 0.8
3659
+
},
3660
+
{
3661
+
"a": "DH",
3662
+
"b": "UW",
3663
+
"id": "DH_UW",
3664
+
"weight": 0.8
3665
+
},
3666
+
{
3667
+
"a": "EH",
3668
+
"b": "DH",
3669
+
"id": "EH_DH",
3670
+
"weight": 0.8
3671
+
},
3672
+
{
3673
+
"a": "EH",
3674
+
"b": "F",
3675
+
"id": "EH_F",
3676
+
"weight": 0.8
3677
+
},
3678
+
{
3679
+
"a": "EH",
3680
+
"b": "HH",
3681
+
"id": "EH_HH",
3682
+
"weight": 0.8
3683
+
},
3684
+
{
3685
+
"a": "EH",
3686
+
"b": "S",
3687
+
"id": "EH_S",
3688
+
"weight": 0.8
3689
+
},
3690
+
{
3691
+
"a": "EH",
3692
+
"b": "SH",
3693
+
"id": "EH_SH",
3694
+
"weight": 0.8
3695
+
},
3696
+
{
3697
+
"a": "EH",
3698
+
"b": "TH",
3699
+
"id": "EH_TH",
3700
+
"weight": 0.8
3701
+
},
3702
+
{
3703
+
"a": "EH",
3704
+
"b": "V",
3705
+
"id": "EH_V",
3706
+
"weight": 0.8
3707
+
},
3708
+
{
3709
+
"a": "EH",
3710
+
"b": "Z",
3711
+
"id": "EH_Z",
3712
+
"weight": 0.8
3713
+
},
3714
+
{
3715
+
"a": "EH",
3716
+
"b": "ZH",
3717
+
"id": "EH_ZH",
3718
+
"weight": 0.8
3719
+
},
3720
+
{
3721
+
"a": "ER",
3722
+
"b": "DH",
3723
+
"id": "ER_DH",
3724
+
"weight": 0.8
3725
+
},
3726
+
{
3727
+
"a": "ER",
3728
+
"b": "F",
3729
+
"id": "ER_F",
3730
+
"weight": 0.8
3731
+
},
3732
+
{
3733
+
"a": "ER",
3734
+
"b": "HH",
3735
+
"id": "ER_HH",
3736
+
"weight": 0.8
3737
+
},
3738
+
{
3739
+
"a": "ER",
3740
+
"b": "S",
3741
+
"id": "ER_S",
3742
+
"weight": 0.8
3743
+
},
3744
+
{
3745
+
"a": "ER",
3746
+
"b": "SH",
3747
+
"id": "ER_SH",
3748
+
"weight": 0.8
3749
+
},
3750
+
{
3751
+
"a": "ER",
3752
+
"b": "TH",
3753
+
"id": "ER_TH",
3754
+
"weight": 0.8
3755
+
},
3756
+
{
3757
+
"a": "ER",
3758
+
"b": "V",
3759
+
"id": "ER_V",
3760
+
"weight": 0.8
3761
+
},
3762
+
{
3763
+
"a": "ER",
3764
+
"b": "Z",
3765
+
"id": "ER_Z",
3766
+
"weight": 0.8
3767
+
},
3768
+
{
3769
+
"a": "ER",
3770
+
"b": "ZH",
3771
+
"id": "ER_ZH",
3772
+
"weight": 0.8
3773
+
},
3774
+
{
3775
+
"a": "EY",
3776
+
"b": "DH",
3777
+
"id": "EY_DH",
3778
+
"weight": 0.8
3779
+
},
3780
+
{
3781
+
"a": "EY",
3782
+
"b": "F",
3783
+
"id": "EY_F",
3784
+
"weight": 0.8
3785
+
},
3786
+
{
3787
+
"a": "EY",
3788
+
"b": "HH",
3789
+
"id": "EY_HH",
3790
+
"weight": 0.8
3791
+
},
3792
+
{
3793
+
"a": "EY",
3794
+
"b": "S",
3795
+
"id": "EY_S",
3796
+
"weight": 0.8
3797
+
},
3798
+
{
3799
+
"a": "EY",
3800
+
"b": "SH",
3801
+
"id": "EY_SH",
3802
+
"weight": 0.8
3803
+
},
3804
+
{
3805
+
"a": "EY",
3806
+
"b": "TH",
3807
+
"id": "EY_TH",
3808
+
"weight": 0.8
3809
+
},
3810
+
{
3811
+
"a": "EY",
3812
+
"b": "V",
3813
+
"id": "EY_V",
3814
+
"weight": 0.8
3815
+
},
3816
+
{
3817
+
"a": "EY",
3818
+
"b": "Z",
3819
+
"id": "EY_Z",
3820
+
"weight": 0.8
3821
+
},
3822
+
{
3823
+
"a": "EY",
3824
+
"b": "ZH",
3825
+
"id": "EY_ZH",
3826
+
"weight": 0.8
3827
+
},
3828
+
{
3829
+
"a": "F",
3830
+
"b": "AA",
3831
+
"id": "F_AA",
3832
+
"weight": 0.8
3833
+
},
3834
+
{
3835
+
"a": "F",
3836
+
"b": "AE",
3837
+
"id": "F_AE",
3838
+
"weight": 0.8
3839
+
},
3840
+
{
3841
+
"a": "F",
3842
+
"b": "AH",
3843
+
"id": "F_AH",
3844
+
"weight": 0.8
3845
+
},
3846
+
{
3847
+
"a": "F",
3848
+
"b": "AO",
3849
+
"id": "F_AO",
3850
+
"weight": 0.8
3851
+
},
3852
+
{
3853
+
"a": "F",
3854
+
"b": "AW",
3855
+
"id": "F_AW",
3856
+
"weight": 0.8
3857
+
},
3858
+
{
3859
+
"a": "F",
3860
+
"b": "AY",
3861
+
"id": "F_AY",
3862
+
"weight": 0.8
3863
+
},
3864
+
{
3865
+
"a": "F",
3866
+
"b": "EH",
3867
+
"id": "F_EH",
3868
+
"weight": 0.8
3869
+
},
3870
+
{
3871
+
"a": "F",
3872
+
"b": "ER",
3873
+
"id": "F_ER",
3874
+
"weight": 0.8
3875
+
},
3876
+
{
3877
+
"a": "F",
3878
+
"b": "EY",
3879
+
"id": "F_EY",
3880
+
"weight": 0.8
3881
+
},
3882
+
{
3883
+
"a": "F",
3884
+
"b": "IH",
3885
+
"id": "F_IH",
3886
+
"weight": 0.8
3887
+
},
3888
+
{
3889
+
"a": "F",
3890
+
"b": "IY",
3891
+
"id": "F_IY",
3892
+
"weight": 0.8
3893
+
},
3894
+
{
3895
+
"a": "F",
3896
+
"b": "OW",
3897
+
"id": "F_OW",
3898
+
"weight": 0.8
3899
+
},
3900
+
{
3901
+
"a": "F",
3902
+
"b": "OY",
3903
+
"id": "F_OY",
3904
+
"weight": 0.8
3905
+
},
3906
+
{
3907
+
"a": "F",
3908
+
"b": "UH",
3909
+
"id": "F_UH",
3910
+
"weight": 0.8
3911
+
},
3912
+
{
3913
+
"a": "F",
3914
+
"b": "UW",
3915
+
"id": "F_UW",
3916
+
"weight": 0.8
3917
+
},
3918
+
{
3919
+
"a": "HH",
3920
+
"b": "AA",
3921
+
"id": "HH_AA",
3922
+
"weight": 0.8
3923
+
},
3924
+
{
3925
+
"a": "HH",
3926
+
"b": "AE",
3927
+
"id": "HH_AE",
3928
+
"weight": 0.8
3929
+
},
3930
+
{
3931
+
"a": "HH",
3932
+
"b": "AH",
3933
+
"id": "HH_AH",
3934
+
"weight": 0.8
3935
+
},
3936
+
{
3937
+
"a": "HH",
3938
+
"b": "AO",
3939
+
"id": "HH_AO",
3940
+
"weight": 0.8
3941
+
},
3942
+
{
3943
+
"a": "HH",
3944
+
"b": "AW",
3945
+
"id": "HH_AW",
3946
+
"weight": 0.8
3947
+
},
3948
+
{
3949
+
"a": "HH",
3950
+
"b": "AY",
3951
+
"id": "HH_AY",
3952
+
"weight": 0.8
3953
+
},
3954
+
{
3955
+
"a": "HH",
3956
+
"b": "EH",
3957
+
"id": "HH_EH",
3958
+
"weight": 0.8
3959
+
},
3960
+
{
3961
+
"a": "HH",
3962
+
"b": "ER",
3963
+
"id": "HH_ER",
3964
+
"weight": 0.8
3965
+
},
3966
+
{
3967
+
"a": "HH",
3968
+
"b": "EY",
3969
+
"id": "HH_EY",
3970
+
"weight": 0.8
3971
+
},
3972
+
{
3973
+
"a": "HH",
3974
+
"b": "IH",
3975
+
"id": "HH_IH",
3976
+
"weight": 0.8
3977
+
},
3978
+
{
3979
+
"a": "HH",
3980
+
"b": "IY",
3981
+
"id": "HH_IY",
3982
+
"weight": 0.8
3983
+
},
3984
+
{
3985
+
"a": "HH",
3986
+
"b": "OW",
3987
+
"id": "HH_OW",
3988
+
"weight": 0.8
3989
+
},
3990
+
{
3991
+
"a": "HH",
3992
+
"b": "OY",
3993
+
"id": "HH_OY",
3994
+
"weight": 0.8
3995
+
},
3996
+
{
3997
+
"a": "HH",
3998
+
"b": "UH",
3999
+
"id": "HH_UH",
4000
+
"weight": 0.8
4001
+
},
4002
+
{
4003
+
"a": "HH",
4004
+
"b": "UW",
4005
+
"id": "HH_UW",
4006
+
"weight": 0.8
4007
+
},
4008
+
{
4009
+
"a": "IH",
4010
+
"b": "DH",
4011
+
"id": "IH_DH",
4012
+
"weight": 0.8
4013
+
},
4014
+
{
4015
+
"a": "IH",
4016
+
"b": "F",
4017
+
"id": "IH_F",
4018
+
"weight": 0.8
4019
+
},
4020
+
{
4021
+
"a": "IH",
4022
+
"b": "HH",
4023
+
"id": "IH_HH",
4024
+
"weight": 0.8
4025
+
},
4026
+
{
4027
+
"a": "IH",
4028
+
"b": "S",
4029
+
"id": "IH_S",
4030
+
"weight": 0.8
4031
+
},
4032
+
{
4033
+
"a": "IH",
4034
+
"b": "SH",
4035
+
"id": "IH_SH",
4036
+
"weight": 0.8
4037
+
},
4038
+
{
4039
+
"a": "IH",
4040
+
"b": "TH",
4041
+
"id": "IH_TH",
4042
+
"weight": 0.8
4043
+
},
4044
+
{
4045
+
"a": "IH",
4046
+
"b": "V",
4047
+
"id": "IH_V",
4048
+
"weight": 0.8
4049
+
},
4050
+
{
4051
+
"a": "IH",
4052
+
"b": "Z",
4053
+
"id": "IH_Z",
4054
+
"weight": 0.8
4055
+
},
4056
+
{
4057
+
"a": "IH",
4058
+
"b": "ZH",
4059
+
"id": "IH_ZH",
4060
+
"weight": 0.8
4061
+
},
4062
+
{
4063
+
"a": "IY",
4064
+
"b": "DH",
4065
+
"id": "IY_DH",
4066
+
"weight": 0.8
4067
+
},
4068
+
{
4069
+
"a": "IY",
4070
+
"b": "F",
4071
+
"id": "IY_F",
4072
+
"weight": 0.8
4073
+
},
4074
+
{
4075
+
"a": "IY",
4076
+
"b": "HH",
4077
+
"id": "IY_HH",
4078
+
"weight": 0.8
4079
+
},
4080
+
{
4081
+
"a": "IY",
4082
+
"b": "S",
4083
+
"id": "IY_S",
4084
+
"weight": 0.8
4085
+
},
4086
+
{
4087
+
"a": "IY",
4088
+
"b": "SH",
4089
+
"id": "IY_SH",
4090
+
"weight": 0.8
4091
+
},
4092
+
{
4093
+
"a": "IY",
4094
+
"b": "TH",
4095
+
"id": "IY_TH",
4096
+
"weight": 0.8
4097
+
},
4098
+
{
4099
+
"a": "IY",
4100
+
"b": "V",
4101
+
"id": "IY_V",
4102
+
"weight": 0.8
4103
+
},
4104
+
{
4105
+
"a": "IY",
4106
+
"b": "Z",
4107
+
"id": "IY_Z",
4108
+
"weight": 0.8
4109
+
},
4110
+
{
4111
+
"a": "IY",
4112
+
"b": "ZH",
4113
+
"id": "IY_ZH",
4114
+
"weight": 0.8
4115
+
},
4116
+
{
4117
+
"a": "OW",
4118
+
"b": "DH",
4119
+
"id": "OW_DH",
4120
+
"weight": 0.8
4121
+
},
4122
+
{
4123
+
"a": "OW",
4124
+
"b": "F",
4125
+
"id": "OW_F",
4126
+
"weight": 0.8
4127
+
},
4128
+
{
4129
+
"a": "OW",
4130
+
"b": "HH",
4131
+
"id": "OW_HH",
4132
+
"weight": 0.8
4133
+
},
4134
+
{
4135
+
"a": "OW",
4136
+
"b": "S",
4137
+
"id": "OW_S",
4138
+
"weight": 0.8
4139
+
},
4140
+
{
4141
+
"a": "OW",
4142
+
"b": "SH",
4143
+
"id": "OW_SH",
4144
+
"weight": 0.8
4145
+
},
4146
+
{
4147
+
"a": "OW",
4148
+
"b": "TH",
4149
+
"id": "OW_TH",
4150
+
"weight": 0.8
4151
+
},
4152
+
{
4153
+
"a": "OW",
4154
+
"b": "V",
4155
+
"id": "OW_V",
4156
+
"weight": 0.8
4157
+
},
4158
+
{
4159
+
"a": "OW",
4160
+
"b": "Z",
4161
+
"id": "OW_Z",
4162
+
"weight": 0.8
4163
+
},
4164
+
{
4165
+
"a": "OW",
4166
+
"b": "ZH",
4167
+
"id": "OW_ZH",
4168
+
"weight": 0.8
4169
+
},
4170
+
{
4171
+
"a": "OY",
4172
+
"b": "DH",
4173
+
"id": "OY_DH",
4174
+
"weight": 0.8
4175
+
},
4176
+
{
4177
+
"a": "OY",
4178
+
"b": "F",
4179
+
"id": "OY_F",
4180
+
"weight": 0.8
4181
+
},
4182
+
{
4183
+
"a": "OY",
4184
+
"b": "HH",
4185
+
"id": "OY_HH",
4186
+
"weight": 0.8
4187
+
},
4188
+
{
4189
+
"a": "OY",
4190
+
"b": "S",
4191
+
"id": "OY_S",
4192
+
"weight": 0.8
4193
+
},
4194
+
{
4195
+
"a": "OY",
4196
+
"b": "SH",
4197
+
"id": "OY_SH",
4198
+
"weight": 0.8
4199
+
},
4200
+
{
4201
+
"a": "OY",
4202
+
"b": "TH",
4203
+
"id": "OY_TH",
4204
+
"weight": 0.8
4205
+
},
4206
+
{
4207
+
"a": "OY",
4208
+
"b": "V",
4209
+
"id": "OY_V",
4210
+
"weight": 0.8
4211
+
},
4212
+
{
4213
+
"a": "OY",
4214
+
"b": "Z",
4215
+
"id": "OY_Z",
4216
+
"weight": 0.8
4217
+
},
4218
+
{
4219
+
"a": "OY",
4220
+
"b": "ZH",
4221
+
"id": "OY_ZH",
4222
+
"weight": 0.8
4223
+
},
4224
+
{
4225
+
"a": "S",
4226
+
"b": "AA",
4227
+
"id": "S_AA",
4228
+
"weight": 0.8
4229
+
},
4230
+
{
4231
+
"a": "S",
4232
+
"b": "AE",
4233
+
"id": "S_AE",
4234
+
"weight": 0.8
4235
+
},
4236
+
{
4237
+
"a": "S",
4238
+
"b": "AH",
4239
+
"id": "S_AH",
4240
+
"weight": 0.8
4241
+
},
4242
+
{
4243
+
"a": "S",
4244
+
"b": "AO",
4245
+
"id": "S_AO",
4246
+
"weight": 0.8
4247
+
},
4248
+
{
4249
+
"a": "S",
4250
+
"b": "AW",
4251
+
"id": "S_AW",
4252
+
"weight": 0.8
4253
+
},
4254
+
{
4255
+
"a": "S",
4256
+
"b": "AY",
4257
+
"id": "S_AY",
4258
+
"weight": 0.8
4259
+
},
4260
+
{
4261
+
"a": "S",
4262
+
"b": "EH",
4263
+
"id": "S_EH",
4264
+
"weight": 0.8
4265
+
},
4266
+
{
4267
+
"a": "S",
4268
+
"b": "ER",
4269
+
"id": "S_ER",
4270
+
"weight": 0.8
4271
+
},
4272
+
{
4273
+
"a": "S",
4274
+
"b": "EY",
4275
+
"id": "S_EY",
4276
+
"weight": 0.8
4277
+
},
4278
+
{
4279
+
"a": "S",
4280
+
"b": "IH",
4281
+
"id": "S_IH",
4282
+
"weight": 0.8
4283
+
},
4284
+
{
4285
+
"a": "S",
4286
+
"b": "IY",
4287
+
"id": "S_IY",
4288
+
"weight": 0.8
4289
+
},
4290
+
{
4291
+
"a": "S",
4292
+
"b": "OW",
4293
+
"id": "S_OW",
4294
+
"weight": 0.8
4295
+
},
4296
+
{
4297
+
"a": "S",
4298
+
"b": "OY",
4299
+
"id": "S_OY",
4300
+
"weight": 0.8
4301
+
},
4302
+
{
4303
+
"a": "S",
4304
+
"b": "UH",
4305
+
"id": "S_UH",
4306
+
"weight": 0.8
4307
+
},
4308
+
{
4309
+
"a": "S",
4310
+
"b": "UW",
4311
+
"id": "S_UW",
4312
+
"weight": 0.8
4313
+
},
4314
+
{
4315
+
"a": "SH",
4316
+
"b": "AA",
4317
+
"id": "SH_AA",
4318
+
"weight": 0.8
4319
+
},
4320
+
{
4321
+
"a": "SH",
4322
+
"b": "AE",
4323
+
"id": "SH_AE",
4324
+
"weight": 0.8
4325
+
},
4326
+
{
4327
+
"a": "SH",
4328
+
"b": "AH",
4329
+
"id": "SH_AH",
4330
+
"weight": 0.8
4331
+
},
4332
+
{
4333
+
"a": "SH",
4334
+
"b": "AO",
4335
+
"id": "SH_AO",
4336
+
"weight": 0.8
4337
+
},
4338
+
{
4339
+
"a": "SH",
4340
+
"b": "AW",
4341
+
"id": "SH_AW",
4342
+
"weight": 0.8
4343
+
},
4344
+
{
4345
+
"a": "SH",
4346
+
"b": "AY",
4347
+
"id": "SH_AY",
4348
+
"weight": 0.8
4349
+
},
4350
+
{
4351
+
"a": "SH",
4352
+
"b": "EH",
4353
+
"id": "SH_EH",
4354
+
"weight": 0.8
4355
+
},
4356
+
{
4357
+
"a": "SH",
4358
+
"b": "ER",
4359
+
"id": "SH_ER",
4360
+
"weight": 0.8
4361
+
},
4362
+
{
4363
+
"a": "SH",
4364
+
"b": "EY",
4365
+
"id": "SH_EY",
4366
+
"weight": 0.8
4367
+
},
4368
+
{
4369
+
"a": "SH",
4370
+
"b": "IH",
4371
+
"id": "SH_IH",
4372
+
"weight": 0.8
4373
+
},
4374
+
{
4375
+
"a": "SH",
4376
+
"b": "IY",
4377
+
"id": "SH_IY",
4378
+
"weight": 0.8
4379
+
},
4380
+
{
4381
+
"a": "SH",
4382
+
"b": "OW",
4383
+
"id": "SH_OW",
4384
+
"weight": 0.8
4385
+
},
4386
+
{
4387
+
"a": "SH",
4388
+
"b": "OY",
4389
+
"id": "SH_OY",
4390
+
"weight": 0.8
4391
+
},
4392
+
{
4393
+
"a": "SH",
4394
+
"b": "UH",
4395
+
"id": "SH_UH",
4396
+
"weight": 0.8
4397
+
},
4398
+
{
4399
+
"a": "SH",
4400
+
"b": "UW",
4401
+
"id": "SH_UW",
4402
+
"weight": 0.8
4403
+
},
4404
+
{
4405
+
"a": "TH",
4406
+
"b": "AA",
4407
+
"id": "TH_AA",
4408
+
"weight": 0.8
4409
+
},
4410
+
{
4411
+
"a": "TH",
4412
+
"b": "AE",
4413
+
"id": "TH_AE",
4414
+
"weight": 0.8
4415
+
},
4416
+
{
4417
+
"a": "TH",
4418
+
"b": "AH",
4419
+
"id": "TH_AH",
4420
+
"weight": 0.8
4421
+
},
4422
+
{
4423
+
"a": "TH",
4424
+
"b": "AO",
4425
+
"id": "TH_AO",
4426
+
"weight": 0.8
4427
+
},
4428
+
{
4429
+
"a": "TH",
4430
+
"b": "AW",
4431
+
"id": "TH_AW",
4432
+
"weight": 0.8
4433
+
},
4434
+
{
4435
+
"a": "TH",
4436
+
"b": "AY",
4437
+
"id": "TH_AY",
4438
+
"weight": 0.8
4439
+
},
4440
+
{
4441
+
"a": "TH",
4442
+
"b": "EH",
4443
+
"id": "TH_EH",
4444
+
"weight": 0.8
4445
+
},
4446
+
{
4447
+
"a": "TH",
4448
+
"b": "ER",
4449
+
"id": "TH_ER",
4450
+
"weight": 0.8
4451
+
},
4452
+
{
4453
+
"a": "TH",
4454
+
"b": "EY",
4455
+
"id": "TH_EY",
4456
+
"weight": 0.8
4457
+
},
4458
+
{
4459
+
"a": "TH",
4460
+
"b": "IH",
4461
+
"id": "TH_IH",
4462
+
"weight": 0.8
4463
+
},
4464
+
{
4465
+
"a": "TH",
4466
+
"b": "IY",
4467
+
"id": "TH_IY",
4468
+
"weight": 0.8
4469
+
},
4470
+
{
4471
+
"a": "TH",
4472
+
"b": "OW",
4473
+
"id": "TH_OW",
4474
+
"weight": 0.8
4475
+
},
4476
+
{
4477
+
"a": "TH",
4478
+
"b": "OY",
4479
+
"id": "TH_OY",
4480
+
"weight": 0.8
4481
+
},
4482
+
{
4483
+
"a": "TH",
4484
+
"b": "UH",
4485
+
"id": "TH_UH",
4486
+
"weight": 0.8
4487
+
},
4488
+
{
4489
+
"a": "TH",
4490
+
"b": "UW",
4491
+
"id": "TH_UW",
4492
+
"weight": 0.8
4493
+
},
4494
+
{
4495
+
"a": "UH",
4496
+
"b": "DH",
4497
+
"id": "UH_DH",
4498
+
"weight": 0.8
4499
+
},
4500
+
{
4501
+
"a": "UH",
4502
+
"b": "F",
4503
+
"id": "UH_F",
4504
+
"weight": 0.8
4505
+
},
4506
+
{
4507
+
"a": "UH",
4508
+
"b": "HH",
4509
+
"id": "UH_HH",
4510
+
"weight": 0.8
4511
+
},
4512
+
{
4513
+
"a": "UH",
4514
+
"b": "S",
4515
+
"id": "UH_S",
4516
+
"weight": 0.8
4517
+
},
4518
+
{
4519
+
"a": "UH",
4520
+
"b": "SH",
4521
+
"id": "UH_SH",
4522
+
"weight": 0.8
4523
+
},
4524
+
{
4525
+
"a": "UH",
4526
+
"b": "TH",
4527
+
"id": "UH_TH",
4528
+
"weight": 0.8
4529
+
},
4530
+
{
4531
+
"a": "UH",
4532
+
"b": "V",
4533
+
"id": "UH_V",
4534
+
"weight": 0.8
4535
+
},
4536
+
{
4537
+
"a": "UH",
4538
+
"b": "Z",
4539
+
"id": "UH_Z",
4540
+
"weight": 0.8
4541
+
},
4542
+
{
4543
+
"a": "UH",
4544
+
"b": "ZH",
4545
+
"id": "UH_ZH",
4546
+
"weight": 0.8
4547
+
},
4548
+
{
4549
+
"a": "UW",
4550
+
"b": "DH",
4551
+
"id": "UW_DH",
4552
+
"weight": 0.8
4553
+
},
4554
+
{
4555
+
"a": "UW",
4556
+
"b": "F",
4557
+
"id": "UW_F",
4558
+
"weight": 0.8
4559
+
},
4560
+
{
4561
+
"a": "UW",
4562
+
"b": "HH",
4563
+
"id": "UW_HH",
4564
+
"weight": 0.8
4565
+
},
4566
+
{
4567
+
"a": "UW",
4568
+
"b": "S",
4569
+
"id": "UW_S",
4570
+
"weight": 0.8
4571
+
},
4572
+
{
4573
+
"a": "UW",
4574
+
"b": "SH",
4575
+
"id": "UW_SH",
4576
+
"weight": 0.8
4577
+
},
4578
+
{
4579
+
"a": "UW",
4580
+
"b": "TH",
4581
+
"id": "UW_TH",
4582
+
"weight": 0.8
4583
+
},
4584
+
{
4585
+
"a": "UW",
4586
+
"b": "V",
4587
+
"id": "UW_V",
4588
+
"weight": 0.8
4589
+
},
4590
+
{
4591
+
"a": "UW",
4592
+
"b": "Z",
4593
+
"id": "UW_Z",
4594
+
"weight": 0.8
4595
+
},
4596
+
{
4597
+
"a": "UW",
4598
+
"b": "ZH",
4599
+
"id": "UW_ZH",
4600
+
"weight": 0.8
4601
+
},
4602
+
{
4603
+
"a": "V",
4604
+
"b": "AA",
4605
+
"id": "V_AA",
4606
+
"weight": 0.8
4607
+
},
4608
+
{
4609
+
"a": "V",
4610
+
"b": "AE",
4611
+
"id": "V_AE",
4612
+
"weight": 0.8
4613
+
},
4614
+
{
4615
+
"a": "V",
4616
+
"b": "AH",
4617
+
"id": "V_AH",
4618
+
"weight": 0.8
4619
+
},
4620
+
{
4621
+
"a": "V",
4622
+
"b": "AO",
4623
+
"id": "V_AO",
4624
+
"weight": 0.8
4625
+
},
4626
+
{
4627
+
"a": "V",
4628
+
"b": "AW",
4629
+
"id": "V_AW",
4630
+
"weight": 0.8
4631
+
},
4632
+
{
4633
+
"a": "V",
4634
+
"b": "AY",
4635
+
"id": "V_AY",
4636
+
"weight": 0.8
4637
+
},
4638
+
{
4639
+
"a": "V",
4640
+
"b": "EH",
4641
+
"id": "V_EH",
4642
+
"weight": 0.8
4643
+
},
4644
+
{
4645
+
"a": "V",
4646
+
"b": "ER",
4647
+
"id": "V_ER",
4648
+
"weight": 0.8
4649
+
},
4650
+
{
4651
+
"a": "V",
4652
+
"b": "EY",
4653
+
"id": "V_EY",
4654
+
"weight": 0.8
4655
+
},
4656
+
{
4657
+
"a": "V",
4658
+
"b": "IH",
4659
+
"id": "V_IH",
4660
+
"weight": 0.8
4661
+
},
4662
+
{
4663
+
"a": "V",
4664
+
"b": "IY",
4665
+
"id": "V_IY",
4666
+
"weight": 0.8
4667
+
},
4668
+
{
4669
+
"a": "V",
4670
+
"b": "OW",
4671
+
"id": "V_OW",
4672
+
"weight": 0.8
4673
+
},
4674
+
{
4675
+
"a": "V",
4676
+
"b": "OY",
4677
+
"id": "V_OY",
4678
+
"weight": 0.8
4679
+
},
4680
+
{
4681
+
"a": "V",
4682
+
"b": "UH",
4683
+
"id": "V_UH",
4684
+
"weight": 0.8
4685
+
},
4686
+
{
4687
+
"a": "V",
4688
+
"b": "UW",
4689
+
"id": "V_UW",
4690
+
"weight": 0.8
4691
+
},
4692
+
{
4693
+
"a": "Z",
4694
+
"b": "AA",
4695
+
"id": "Z_AA",
4696
+
"weight": 0.8
4697
+
},
4698
+
{
4699
+
"a": "Z",
4700
+
"b": "AE",
4701
+
"id": "Z_AE",
4702
+
"weight": 0.8
4703
+
},
4704
+
{
4705
+
"a": "Z",
4706
+
"b": "AH",
4707
+
"id": "Z_AH",
4708
+
"weight": 0.8
4709
+
},
4710
+
{
4711
+
"a": "Z",
4712
+
"b": "AO",
4713
+
"id": "Z_AO",
4714
+
"weight": 0.8
4715
+
},
4716
+
{
4717
+
"a": "Z",
4718
+
"b": "AW",
4719
+
"id": "Z_AW",
4720
+
"weight": 0.8
4721
+
},
4722
+
{
4723
+
"a": "Z",
4724
+
"b": "AY",
4725
+
"id": "Z_AY",
4726
+
"weight": 0.8
4727
+
},
4728
+
{
4729
+
"a": "Z",
4730
+
"b": "EH",
4731
+
"id": "Z_EH",
4732
+
"weight": 0.8
4733
+
},
4734
+
{
4735
+
"a": "Z",
4736
+
"b": "ER",
4737
+
"id": "Z_ER",
4738
+
"weight": 0.8
4739
+
},
4740
+
{
4741
+
"a": "Z",
4742
+
"b": "EY",
4743
+
"id": "Z_EY",
4744
+
"weight": 0.8
4745
+
},
4746
+
{
4747
+
"a": "Z",
4748
+
"b": "IH",
4749
+
"id": "Z_IH",
4750
+
"weight": 0.8
4751
+
},
4752
+
{
4753
+
"a": "Z",
4754
+
"b": "IY",
4755
+
"id": "Z_IY",
4756
+
"weight": 0.8
4757
+
},
4758
+
{
4759
+
"a": "Z",
4760
+
"b": "OW",
4761
+
"id": "Z_OW",
4762
+
"weight": 0.8
4763
+
},
4764
+
{
4765
+
"a": "Z",
4766
+
"b": "OY",
4767
+
"id": "Z_OY",
4768
+
"weight": 0.8
4769
+
},
4770
+
{
4771
+
"a": "Z",
4772
+
"b": "UH",
4773
+
"id": "Z_UH",
4774
+
"weight": 0.8
4775
+
},
4776
+
{
4777
+
"a": "Z",
4778
+
"b": "UW",
4779
+
"id": "Z_UW",
4780
+
"weight": 0.8
4781
+
},
4782
+
{
4783
+
"a": "ZH",
4784
+
"b": "AA",
4785
+
"id": "ZH_AA",
4786
+
"weight": 0.8
4787
+
},
4788
+
{
4789
+
"a": "ZH",
4790
+
"b": "AE",
4791
+
"id": "ZH_AE",
4792
+
"weight": 0.8
4793
+
},
4794
+
{
4795
+
"a": "ZH",
4796
+
"b": "AH",
4797
+
"id": "ZH_AH",
4798
+
"weight": 0.8
4799
+
},
4800
+
{
4801
+
"a": "ZH",
4802
+
"b": "AO",
4803
+
"id": "ZH_AO",
4804
+
"weight": 0.8
4805
+
},
4806
+
{
4807
+
"a": "ZH",
4808
+
"b": "AW",
4809
+
"id": "ZH_AW",
4810
+
"weight": 0.8
4811
+
},
4812
+
{
4813
+
"a": "ZH",
4814
+
"b": "AY",
4815
+
"id": "ZH_AY",
4816
+
"weight": 0.8
4817
+
},
4818
+
{
4819
+
"a": "ZH",
4820
+
"b": "EH",
4821
+
"id": "ZH_EH",
4822
+
"weight": 0.8
4823
+
},
4824
+
{
4825
+
"a": "ZH",
4826
+
"b": "ER",
4827
+
"id": "ZH_ER",
4828
+
"weight": 0.8
4829
+
},
4830
+
{
4831
+
"a": "ZH",
4832
+
"b": "EY",
4833
+
"id": "ZH_EY",
4834
+
"weight": 0.8
4835
+
},
4836
+
{
4837
+
"a": "ZH",
4838
+
"b": "IH",
4839
+
"id": "ZH_IH",
4840
+
"weight": 0.8
4841
+
},
4842
+
{
4843
+
"a": "ZH",
4844
+
"b": "IY",
4845
+
"id": "ZH_IY",
4846
+
"weight": 0.8
4847
+
},
4848
+
{
4849
+
"a": "ZH",
4850
+
"b": "OW",
4851
+
"id": "ZH_OW",
4852
+
"weight": 0.8
4853
+
},
4854
+
{
4855
+
"a": "ZH",
4856
+
"b": "OY",
4857
+
"id": "ZH_OY",
4858
+
"weight": 0.8
4859
+
},
4860
+
{
4861
+
"a": "ZH",
4862
+
"b": "UH",
4863
+
"id": "ZH_UH",
4864
+
"weight": 0.8
4865
+
},
4866
+
{
4867
+
"a": "ZH",
4868
+
"b": "UW",
4869
+
"id": "ZH_UW",
4870
+
"weight": 0.8
4871
+
},
4872
+
{
4873
+
"a": "AA",
4874
+
"b": "#E",
4875
+
"id": "AA_#E",
4876
+
"weight": 0.7
4877
+
},
4878
+
{
4879
+
"a": "AE",
4880
+
"b": "#E",
4881
+
"id": "AE_#E",
4882
+
"weight": 0.7
4883
+
},
4884
+
{
4885
+
"a": "AH",
4886
+
"b": "#E",
4887
+
"id": "AH_#E",
4888
+
"weight": 0.7
4889
+
},
4890
+
{
4891
+
"a": "AO",
4892
+
"b": "#E",
4893
+
"id": "AO_#E",
4894
+
"weight": 0.7
4895
+
},
4896
+
{
4897
+
"a": "AW",
4898
+
"b": "#E",
4899
+
"id": "AW_#E",
4900
+
"weight": 0.7
4901
+
},
4902
+
{
4903
+
"a": "AY",
4904
+
"b": "#E",
4905
+
"id": "AY_#E",
4906
+
"weight": 0.7
4907
+
},
4908
+
{
4909
+
"a": "B",
4910
+
"b": "#E",
4911
+
"id": "B_#E",
4912
+
"weight": 0.7
4913
+
},
4914
+
{
4915
+
"a": "CH",
4916
+
"b": "#E",
4917
+
"id": "CH_#E",
4918
+
"weight": 0.7
4919
+
},
4920
+
{
4921
+
"a": "D",
4922
+
"b": "#E",
4923
+
"id": "D_#E",
4924
+
"weight": 0.7
4925
+
},
4926
+
{
4927
+
"a": "DH",
4928
+
"b": "#E",
4929
+
"id": "DH_#E",
4930
+
"weight": 0.7
4931
+
},
4932
+
{
4933
+
"a": "EH",
4934
+
"b": "#E",
4935
+
"id": "EH_#E",
4936
+
"weight": 0.7
4937
+
},
4938
+
{
4939
+
"a": "ER",
4940
+
"b": "#E",
4941
+
"id": "ER_#E",
4942
+
"weight": 0.7
4943
+
},
4944
+
{
4945
+
"a": "EY",
4946
+
"b": "#E",
4947
+
"id": "EY_#E",
4948
+
"weight": 0.7
4949
+
},
4950
+
{
4951
+
"a": "F",
4952
+
"b": "#E",
4953
+
"id": "F_#E",
4954
+
"weight": 0.7
4955
+
},
4956
+
{
4957
+
"a": "G",
4958
+
"b": "#E",
4959
+
"id": "G_#E",
4960
+
"weight": 0.7
4961
+
},
4962
+
{
4963
+
"a": "IH",
4964
+
"b": "#E",
4965
+
"id": "IH_#E",
4966
+
"weight": 0.7
4967
+
},
4968
+
{
4969
+
"a": "IY",
4970
+
"b": "#E",
4971
+
"id": "IY_#E",
4972
+
"weight": 0.7
4973
+
},
4974
+
{
4975
+
"a": "JH",
4976
+
"b": "#E",
4977
+
"id": "JH_#E",
4978
+
"weight": 0.7
4979
+
},
4980
+
{
4981
+
"a": "K",
4982
+
"b": "#E",
4983
+
"id": "K_#E",
4984
+
"weight": 0.7
4985
+
},
4986
+
{
4987
+
"a": "L",
4988
+
"b": "#E",
4989
+
"id": "L_#E",
4990
+
"weight": 0.7
4991
+
},
4992
+
{
4993
+
"a": "M",
4994
+
"b": "#E",
4995
+
"id": "M_#E",
4996
+
"weight": 0.7
4997
+
},
4998
+
{
4999
+
"a": "N",
5000
+
"b": "#E",
5001
+
"id": "N_#E",
5002
+
"weight": 0.7
5003
+
},
5004
+
{
5005
+
"a": "NG",
5006
+
"b": "#E",
5007
+
"id": "NG_#E",
5008
+
"weight": 0.7
5009
+
},
5010
+
{
5011
+
"a": "OW",
5012
+
"b": "#E",
5013
+
"id": "OW_#E",
5014
+
"weight": 0.7
5015
+
},
5016
+
{
5017
+
"a": "OY",
5018
+
"b": "#E",
5019
+
"id": "OY_#E",
5020
+
"weight": 0.7
5021
+
},
5022
+
{
5023
+
"a": "P",
5024
+
"b": "#E",
5025
+
"id": "P_#E",
5026
+
"weight": 0.7
5027
+
},
5028
+
{
5029
+
"a": "R",
5030
+
"b": "#E",
5031
+
"id": "R_#E",
5032
+
"weight": 0.7
5033
+
},
5034
+
{
5035
+
"a": "S",
5036
+
"b": "#E",
5037
+
"id": "S_#E",
5038
+
"weight": 0.7
5039
+
},
5040
+
{
5041
+
"a": "SH",
5042
+
"b": "#E",
5043
+
"id": "SH_#E",
5044
+
"weight": 0.7
5045
+
},
5046
+
{
5047
+
"a": "T",
5048
+
"b": "#E",
5049
+
"id": "T_#E",
5050
+
"weight": 0.7
5051
+
},
5052
+
{
5053
+
"a": "TH",
5054
+
"b": "#E",
5055
+
"id": "TH_#E",
5056
+
"weight": 0.7
5057
+
},
5058
+
{
5059
+
"a": "UH",
5060
+
"b": "#E",
5061
+
"id": "UH_#E",
5062
+
"weight": 0.7
5063
+
},
5064
+
{
5065
+
"a": "UW",
5066
+
"b": "#E",
5067
+
"id": "UW_#E",
5068
+
"weight": 0.7
5069
+
},
5070
+
{
5071
+
"a": "V",
5072
+
"b": "#E",
5073
+
"id": "V_#E",
5074
+
"weight": 0.7
5075
+
},
5076
+
{
5077
+
"a": "Z",
5078
+
"b": "#E",
5079
+
"id": "Z_#E",
5080
+
"weight": 0.7
5081
+
},
5082
+
{
5083
+
"a": "ZH",
5084
+
"b": "#E",
5085
+
"id": "ZH_#E",
5086
+
"weight": 0.7
5087
+
},
5088
+
{
5089
+
"a": "AA",
5090
+
"b": "CH",
5091
+
"id": "AA_CH",
5092
+
"weight": 0.5
5093
+
},
5094
+
{
5095
+
"a": "AA",
5096
+
"b": "JH",
5097
+
"id": "AA_JH",
5098
+
"weight": 0.5
5099
+
},
5100
+
{
5101
+
"a": "AA",
5102
+
"b": "L",
5103
+
"id": "AA_L",
5104
+
"weight": 0.5
5105
+
},
5106
+
{
5107
+
"a": "AA",
5108
+
"b": "R",
5109
+
"id": "AA_R",
5110
+
"weight": 0.5
5111
+
},
5112
+
{
5113
+
"a": "AA",
5114
+
"b": "W",
5115
+
"id": "AA_W",
5116
+
"weight": 0.5
5117
+
},
5118
+
{
5119
+
"a": "AA",
5120
+
"b": "Y",
5121
+
"id": "AA_Y",
5122
+
"weight": 0.5
5123
+
},
5124
+
{
5125
+
"a": "AE",
5126
+
"b": "CH",
5127
+
"id": "AE_CH",
5128
+
"weight": 0.5
5129
+
},
5130
+
{
5131
+
"a": "AE",
5132
+
"b": "JH",
5133
+
"id": "AE_JH",
5134
+
"weight": 0.5
5135
+
},
5136
+
{
5137
+
"a": "AE",
5138
+
"b": "L",
5139
+
"id": "AE_L",
5140
+
"weight": 0.5
5141
+
},
5142
+
{
5143
+
"a": "AE",
5144
+
"b": "R",
5145
+
"id": "AE_R",
5146
+
"weight": 0.5
5147
+
},
5148
+
{
5149
+
"a": "AE",
5150
+
"b": "W",
5151
+
"id": "AE_W",
5152
+
"weight": 0.5
5153
+
},
5154
+
{
5155
+
"a": "AE",
5156
+
"b": "Y",
5157
+
"id": "AE_Y",
5158
+
"weight": 0.5
5159
+
},
5160
+
{
5161
+
"a": "AH",
5162
+
"b": "CH",
5163
+
"id": "AH_CH",
5164
+
"weight": 0.5
5165
+
},
5166
+
{
5167
+
"a": "AH",
5168
+
"b": "JH",
5169
+
"id": "AH_JH",
5170
+
"weight": 0.5
5171
+
},
5172
+
{
5173
+
"a": "AH",
5174
+
"b": "L",
5175
+
"id": "AH_L",
5176
+
"weight": 0.5
5177
+
},
5178
+
{
5179
+
"a": "AH",
5180
+
"b": "R",
5181
+
"id": "AH_R",
5182
+
"weight": 0.5
5183
+
},
5184
+
{
5185
+
"a": "AH",
5186
+
"b": "W",
5187
+
"id": "AH_W",
5188
+
"weight": 0.5
5189
+
},
5190
+
{
5191
+
"a": "AH",
5192
+
"b": "Y",
5193
+
"id": "AH_Y",
5194
+
"weight": 0.5
5195
+
},
5196
+
{
5197
+
"a": "AO",
5198
+
"b": "CH",
5199
+
"id": "AO_CH",
5200
+
"weight": 0.5
5201
+
},
5202
+
{
5203
+
"a": "AO",
5204
+
"b": "JH",
5205
+
"id": "AO_JH",
5206
+
"weight": 0.5
5207
+
},
5208
+
{
5209
+
"a": "AO",
5210
+
"b": "L",
5211
+
"id": "AO_L",
5212
+
"weight": 0.5
5213
+
},
5214
+
{
5215
+
"a": "AO",
5216
+
"b": "R",
5217
+
"id": "AO_R",
5218
+
"weight": 0.5
5219
+
},
5220
+
{
5221
+
"a": "AO",
5222
+
"b": "W",
5223
+
"id": "AO_W",
5224
+
"weight": 0.5
5225
+
},
5226
+
{
5227
+
"a": "AO",
5228
+
"b": "Y",
5229
+
"id": "AO_Y",
5230
+
"weight": 0.5
5231
+
},
5232
+
{
5233
+
"a": "AW",
5234
+
"b": "CH",
5235
+
"id": "AW_CH",
5236
+
"weight": 0.5
5237
+
},
5238
+
{
5239
+
"a": "AW",
5240
+
"b": "JH",
5241
+
"id": "AW_JH",
5242
+
"weight": 0.5
5243
+
},
5244
+
{
5245
+
"a": "AW",
5246
+
"b": "L",
5247
+
"id": "AW_L",
5248
+
"weight": 0.5
5249
+
},
5250
+
{
5251
+
"a": "AW",
5252
+
"b": "R",
5253
+
"id": "AW_R",
5254
+
"weight": 0.5
5255
+
},
5256
+
{
5257
+
"a": "AW",
5258
+
"b": "W",
5259
+
"id": "AW_W",
5260
+
"weight": 0.5
5261
+
},
5262
+
{
5263
+
"a": "AW",
5264
+
"b": "Y",
5265
+
"id": "AW_Y",
5266
+
"weight": 0.5
5267
+
},
5268
+
{
5269
+
"a": "AY",
5270
+
"b": "CH",
5271
+
"id": "AY_CH",
5272
+
"weight": 0.5
5273
+
},
5274
+
{
5275
+
"a": "AY",
5276
+
"b": "JH",
5277
+
"id": "AY_JH",
5278
+
"weight": 0.5
5279
+
},
5280
+
{
5281
+
"a": "AY",
5282
+
"b": "L",
5283
+
"id": "AY_L",
5284
+
"weight": 0.5
5285
+
},
5286
+
{
5287
+
"a": "AY",
5288
+
"b": "R",
5289
+
"id": "AY_R",
5290
+
"weight": 0.5
5291
+
},
5292
+
{
5293
+
"a": "AY",
5294
+
"b": "W",
5295
+
"id": "AY_W",
5296
+
"weight": 0.5
5297
+
},
5298
+
{
5299
+
"a": "AY",
5300
+
"b": "Y",
5301
+
"id": "AY_Y",
5302
+
"weight": 0.5
5303
+
},
5304
+
{
5305
+
"a": "B",
5306
+
"b": "B",
5307
+
"id": "B_B",
5308
+
"weight": 0.5
5309
+
},
5310
+
{
5311
+
"a": "B",
5312
+
"b": "CH",
5313
+
"id": "B_CH",
5314
+
"weight": 0.5
5315
+
},
5316
+
{
5317
+
"a": "B",
5318
+
"b": "D",
5319
+
"id": "B_D",
5320
+
"weight": 0.5
5321
+
},
5322
+
{
5323
+
"a": "B",
5324
+
"b": "DH",
5325
+
"id": "B_DH",
5326
+
"weight": 0.5
5327
+
},
5328
+
{
5329
+
"a": "B",
5330
+
"b": "F",
5331
+
"id": "B_F",
5332
+
"weight": 0.5
5333
+
},
5334
+
{
5335
+
"a": "B",
5336
+
"b": "G",
5337
+
"id": "B_G",
5338
+
"weight": 0.5
5339
+
},
5340
+
{
5341
+
"a": "B",
5342
+
"b": "JH",
5343
+
"id": "B_JH",
5344
+
"weight": 0.5
5345
+
},
5346
+
{
5347
+
"a": "B",
5348
+
"b": "K",
5349
+
"id": "B_K",
5350
+
"weight": 0.5
5351
+
},
5352
+
{
5353
+
"a": "B",
5354
+
"b": "L",
5355
+
"id": "B_L",
5356
+
"weight": 0.5
5357
+
},
5358
+
{
5359
+
"a": "B",
5360
+
"b": "M",
5361
+
"id": "B_M",
5362
+
"weight": 0.5
5363
+
},
5364
+
{
5365
+
"a": "B",
5366
+
"b": "N",
5367
+
"id": "B_N",
5368
+
"weight": 0.5
5369
+
},
5370
+
{
5371
+
"a": "B",
5372
+
"b": "NG",
5373
+
"id": "B_NG",
5374
+
"weight": 0.5
5375
+
},
5376
+
{
5377
+
"a": "B",
5378
+
"b": "P",
5379
+
"id": "B_P",
5380
+
"weight": 0.5
5381
+
},
5382
+
{
5383
+
"a": "B",
5384
+
"b": "R",
5385
+
"id": "B_R",
5386
+
"weight": 0.5
5387
+
},
5388
+
{
5389
+
"a": "B",
5390
+
"b": "S",
5391
+
"id": "B_S",
5392
+
"weight": 0.5
5393
+
},
5394
+
{
5395
+
"a": "B",
5396
+
"b": "SH",
5397
+
"id": "B_SH",
5398
+
"weight": 0.5
5399
+
},
5400
+
{
5401
+
"a": "B",
5402
+
"b": "T",
5403
+
"id": "B_T",
5404
+
"weight": 0.5
5405
+
},
5406
+
{
5407
+
"a": "B",
5408
+
"b": "TH",
5409
+
"id": "B_TH",
5410
+
"weight": 0.5
5411
+
},
5412
+
{
5413
+
"a": "B",
5414
+
"b": "V",
5415
+
"id": "B_V",
5416
+
"weight": 0.5
5417
+
},
5418
+
{
5419
+
"a": "B",
5420
+
"b": "W",
5421
+
"id": "B_W",
5422
+
"weight": 0.5
5423
+
},
5424
+
{
5425
+
"a": "B",
5426
+
"b": "Y",
5427
+
"id": "B_Y",
5428
+
"weight": 0.5
5429
+
},
5430
+
{
5431
+
"a": "B",
5432
+
"b": "Z",
5433
+
"id": "B_Z",
5434
+
"weight": 0.5
5435
+
},
5436
+
{
5437
+
"a": "CH",
5438
+
"b": "AA",
5439
+
"id": "CH_AA",
5440
+
"weight": 0.5
5441
+
},
5442
+
{
5443
+
"a": "CH",
5444
+
"b": "AE",
5445
+
"id": "CH_AE",
5446
+
"weight": 0.5
5447
+
},
5448
+
{
5449
+
"a": "CH",
5450
+
"b": "AH",
5451
+
"id": "CH_AH",
5452
+
"weight": 0.5
5453
+
},
5454
+
{
5455
+
"a": "CH",
5456
+
"b": "AO",
5457
+
"id": "CH_AO",
5458
+
"weight": 0.5
5459
+
},
5460
+
{
5461
+
"a": "CH",
5462
+
"b": "AW",
5463
+
"id": "CH_AW",
5464
+
"weight": 0.5
5465
+
},
5466
+
{
5467
+
"a": "CH",
5468
+
"b": "AY",
5469
+
"id": "CH_AY",
5470
+
"weight": 0.5
5471
+
},
5472
+
{
5473
+
"a": "CH",
5474
+
"b": "B",
5475
+
"id": "CH_B",
5476
+
"weight": 0.5
5477
+
},
5478
+
{
5479
+
"a": "CH",
5480
+
"b": "CH",
5481
+
"id": "CH_CH",
5482
+
"weight": 0.5
5483
+
},
5484
+
{
5485
+
"a": "CH",
5486
+
"b": "D",
5487
+
"id": "CH_D",
5488
+
"weight": 0.5
5489
+
},
5490
+
{
5491
+
"a": "CH",
5492
+
"b": "DH",
5493
+
"id": "CH_DH",
5494
+
"weight": 0.5
5495
+
},
5496
+
{
5497
+
"a": "CH",
5498
+
"b": "EH",
5499
+
"id": "CH_EH",
5500
+
"weight": 0.5
5501
+
},
5502
+
{
5503
+
"a": "CH",
5504
+
"b": "ER",
5505
+
"id": "CH_ER",
5506
+
"weight": 0.5
5507
+
},
5508
+
{
5509
+
"a": "CH",
5510
+
"b": "EY",
5511
+
"id": "CH_EY",
5512
+
"weight": 0.5
5513
+
},
5514
+
{
5515
+
"a": "CH",
5516
+
"b": "F",
5517
+
"id": "CH_F",
5518
+
"weight": 0.5
5519
+
},
5520
+
{
5521
+
"a": "CH",
5522
+
"b": "G",
5523
+
"id": "CH_G",
5524
+
"weight": 0.5
5525
+
},
5526
+
{
5527
+
"a": "CH",
5528
+
"b": "IH",
5529
+
"id": "CH_IH",
5530
+
"weight": 0.5
5531
+
},
5532
+
{
5533
+
"a": "CH",
5534
+
"b": "IY",
5535
+
"id": "CH_IY",
5536
+
"weight": 0.5
5537
+
},
5538
+
{
5539
+
"a": "CH",
5540
+
"b": "JH",
5541
+
"id": "CH_JH",
5542
+
"weight": 0.5
5543
+
},
5544
+
{
5545
+
"a": "CH",
5546
+
"b": "K",
5547
+
"id": "CH_K",
5548
+
"weight": 0.5
5549
+
},
5550
+
{
5551
+
"a": "CH",
5552
+
"b": "L",
5553
+
"id": "CH_L",
5554
+
"weight": 0.5
5555
+
},
5556
+
{
5557
+
"a": "CH",
5558
+
"b": "M",
5559
+
"id": "CH_M",
5560
+
"weight": 0.5
5561
+
},
5562
+
{
5563
+
"a": "CH",
5564
+
"b": "N",
5565
+
"id": "CH_N",
5566
+
"weight": 0.5
5567
+
},
5568
+
{
5569
+
"a": "CH",
5570
+
"b": "NG",
5571
+
"id": "CH_NG",
5572
+
"weight": 0.5
5573
+
},
5574
+
{
5575
+
"a": "CH",
5576
+
"b": "OW",
5577
+
"id": "CH_OW",
5578
+
"weight": 0.5
5579
+
},
5580
+
{
5581
+
"a": "CH",
5582
+
"b": "OY",
5583
+
"id": "CH_OY",
5584
+
"weight": 0.5
5585
+
},
5586
+
{
5587
+
"a": "CH",
5588
+
"b": "P",
5589
+
"id": "CH_P",
5590
+
"weight": 0.5
5591
+
},
5592
+
{
5593
+
"a": "CH",
5594
+
"b": "R",
5595
+
"id": "CH_R",
5596
+
"weight": 0.5
5597
+
},
5598
+
{
5599
+
"a": "CH",
5600
+
"b": "S",
5601
+
"id": "CH_S",
5602
+
"weight": 0.5
5603
+
},
5604
+
{
5605
+
"a": "CH",
5606
+
"b": "SH",
5607
+
"id": "CH_SH",
5608
+
"weight": 0.5
5609
+
},
5610
+
{
5611
+
"a": "CH",
5612
+
"b": "T",
5613
+
"id": "CH_T",
5614
+
"weight": 0.5
5615
+
},
5616
+
{
5617
+
"a": "CH",
5618
+
"b": "TH",
5619
+
"id": "CH_TH",
5620
+
"weight": 0.5
5621
+
},
5622
+
{
5623
+
"a": "CH",
5624
+
"b": "UH",
5625
+
"id": "CH_UH",
5626
+
"weight": 0.5
5627
+
},
5628
+
{
5629
+
"a": "CH",
5630
+
"b": "UW",
5631
+
"id": "CH_UW",
5632
+
"weight": 0.5
5633
+
},
5634
+
{
5635
+
"a": "CH",
5636
+
"b": "V",
5637
+
"id": "CH_V",
5638
+
"weight": 0.5
5639
+
},
5640
+
{
5641
+
"a": "CH",
5642
+
"b": "W",
5643
+
"id": "CH_W",
5644
+
"weight": 0.5
5645
+
},
5646
+
{
5647
+
"a": "CH",
5648
+
"b": "Y",
5649
+
"id": "CH_Y",
5650
+
"weight": 0.5
5651
+
},
5652
+
{
5653
+
"a": "CH",
5654
+
"b": "Z",
5655
+
"id": "CH_Z",
5656
+
"weight": 0.5
5657
+
},
5658
+
{
5659
+
"a": "D",
5660
+
"b": "B",
5661
+
"id": "D_B",
5662
+
"weight": 0.5
5663
+
},
5664
+
{
5665
+
"a": "D",
5666
+
"b": "CH",
5667
+
"id": "D_CH",
5668
+
"weight": 0.5
5669
+
},
5670
+
{
5671
+
"a": "D",
5672
+
"b": "D",
5673
+
"id": "D_D",
5674
+
"weight": 0.5
5675
+
},
5676
+
{
5677
+
"a": "D",
5678
+
"b": "DH",
5679
+
"id": "D_DH",
5680
+
"weight": 0.5
5681
+
},
5682
+
{
5683
+
"a": "D",
5684
+
"b": "F",
5685
+
"id": "D_F",
5686
+
"weight": 0.5
5687
+
},
5688
+
{
5689
+
"a": "D",
5690
+
"b": "G",
5691
+
"id": "D_G",
5692
+
"weight": 0.5
5693
+
},
5694
+
{
5695
+
"a": "D",
5696
+
"b": "JH",
5697
+
"id": "D_JH",
5698
+
"weight": 0.5
5699
+
},
5700
+
{
5701
+
"a": "D",
5702
+
"b": "K",
5703
+
"id": "D_K",
5704
+
"weight": 0.5
5705
+
},
5706
+
{
5707
+
"a": "D",
5708
+
"b": "L",
5709
+
"id": "D_L",
5710
+
"weight": 0.5
5711
+
},
5712
+
{
5713
+
"a": "D",
5714
+
"b": "M",
5715
+
"id": "D_M",
5716
+
"weight": 0.5
5717
+
},
5718
+
{
5719
+
"a": "D",
5720
+
"b": "N",
5721
+
"id": "D_N",
5722
+
"weight": 0.5
5723
+
},
5724
+
{
5725
+
"a": "D",
5726
+
"b": "NG",
5727
+
"id": "D_NG",
5728
+
"weight": 0.5
5729
+
},
5730
+
{
5731
+
"a": "D",
5732
+
"b": "P",
5733
+
"id": "D_P",
5734
+
"weight": 0.5
5735
+
},
5736
+
{
5737
+
"a": "D",
5738
+
"b": "R",
5739
+
"id": "D_R",
5740
+
"weight": 0.5
5741
+
},
5742
+
{
5743
+
"a": "D",
5744
+
"b": "S",
5745
+
"id": "D_S",
5746
+
"weight": 0.5
5747
+
},
5748
+
{
5749
+
"a": "D",
5750
+
"b": "SH",
5751
+
"id": "D_SH",
5752
+
"weight": 0.5
5753
+
},
5754
+
{
5755
+
"a": "D",
5756
+
"b": "T",
5757
+
"id": "D_T",
5758
+
"weight": 0.5
5759
+
},
5760
+
{
5761
+
"a": "D",
5762
+
"b": "TH",
5763
+
"id": "D_TH",
5764
+
"weight": 0.5
5765
+
},
5766
+
{
5767
+
"a": "D",
5768
+
"b": "V",
5769
+
"id": "D_V",
5770
+
"weight": 0.5
5771
+
},
5772
+
{
5773
+
"a": "D",
5774
+
"b": "W",
5775
+
"id": "D_W",
5776
+
"weight": 0.5
5777
+
},
5778
+
{
5779
+
"a": "D",
5780
+
"b": "Y",
5781
+
"id": "D_Y",
5782
+
"weight": 0.5
5783
+
},
5784
+
{
5785
+
"a": "D",
5786
+
"b": "Z",
5787
+
"id": "D_Z",
5788
+
"weight": 0.5
5789
+
},
5790
+
{
5791
+
"a": "DH",
5792
+
"b": "B",
5793
+
"id": "DH_B",
5794
+
"weight": 0.5
5795
+
},
5796
+
{
5797
+
"a": "DH",
5798
+
"b": "CH",
5799
+
"id": "DH_CH",
5800
+
"weight": 0.5
5801
+
},
5802
+
{
5803
+
"a": "DH",
5804
+
"b": "D",
5805
+
"id": "DH_D",
5806
+
"weight": 0.5
5807
+
},
5808
+
{
5809
+
"a": "DH",
5810
+
"b": "DH",
5811
+
"id": "DH_DH",
5812
+
"weight": 0.5
5813
+
},
5814
+
{
5815
+
"a": "DH",
5816
+
"b": "F",
5817
+
"id": "DH_F",
5818
+
"weight": 0.5
5819
+
},
5820
+
{
5821
+
"a": "DH",
5822
+
"b": "G",
5823
+
"id": "DH_G",
5824
+
"weight": 0.5
5825
+
},
5826
+
{
5827
+
"a": "DH",
5828
+
"b": "JH",
5829
+
"id": "DH_JH",
5830
+
"weight": 0.5
5831
+
},
5832
+
{
5833
+
"a": "DH",
5834
+
"b": "K",
5835
+
"id": "DH_K",
5836
+
"weight": 0.5
5837
+
},
5838
+
{
5839
+
"a": "DH",
5840
+
"b": "L",
5841
+
"id": "DH_L",
5842
+
"weight": 0.5
5843
+
},
5844
+
{
5845
+
"a": "DH",
5846
+
"b": "M",
5847
+
"id": "DH_M",
5848
+
"weight": 0.5
5849
+
},
5850
+
{
5851
+
"a": "DH",
5852
+
"b": "N",
5853
+
"id": "DH_N",
5854
+
"weight": 0.5
5855
+
},
5856
+
{
5857
+
"a": "DH",
5858
+
"b": "NG",
5859
+
"id": "DH_NG",
5860
+
"weight": 0.5
5861
+
},
5862
+
{
5863
+
"a": "DH",
5864
+
"b": "P",
5865
+
"id": "DH_P",
5866
+
"weight": 0.5
5867
+
},
5868
+
{
5869
+
"a": "DH",
5870
+
"b": "R",
5871
+
"id": "DH_R",
5872
+
"weight": 0.5
5873
+
},
5874
+
{
5875
+
"a": "DH",
5876
+
"b": "S",
5877
+
"id": "DH_S",
5878
+
"weight": 0.5
5879
+
},
5880
+
{
5881
+
"a": "DH",
5882
+
"b": "SH",
5883
+
"id": "DH_SH",
5884
+
"weight": 0.5
5885
+
},
5886
+
{
5887
+
"a": "DH",
5888
+
"b": "T",
5889
+
"id": "DH_T",
5890
+
"weight": 0.5
5891
+
},
5892
+
{
5893
+
"a": "DH",
5894
+
"b": "TH",
5895
+
"id": "DH_TH",
5896
+
"weight": 0.5
5897
+
},
5898
+
{
5899
+
"a": "DH",
5900
+
"b": "V",
5901
+
"id": "DH_V",
5902
+
"weight": 0.5
5903
+
},
5904
+
{
5905
+
"a": "DH",
5906
+
"b": "W",
5907
+
"id": "DH_W",
5908
+
"weight": 0.5
5909
+
},
5910
+
{
5911
+
"a": "DH",
5912
+
"b": "Y",
5913
+
"id": "DH_Y",
5914
+
"weight": 0.5
5915
+
},
5916
+
{
5917
+
"a": "DH",
5918
+
"b": "Z",
5919
+
"id": "DH_Z",
5920
+
"weight": 0.5
5921
+
},
5922
+
{
5923
+
"a": "EH",
5924
+
"b": "CH",
5925
+
"id": "EH_CH",
5926
+
"weight": 0.5
5927
+
},
5928
+
{
5929
+
"a": "EH",
5930
+
"b": "JH",
5931
+
"id": "EH_JH",
5932
+
"weight": 0.5
5933
+
},
5934
+
{
5935
+
"a": "EH",
5936
+
"b": "L",
5937
+
"id": "EH_L",
5938
+
"weight": 0.5
5939
+
},
5940
+
{
5941
+
"a": "EH",
5942
+
"b": "R",
5943
+
"id": "EH_R",
5944
+
"weight": 0.5
5945
+
},
5946
+
{
5947
+
"a": "EH",
5948
+
"b": "W",
5949
+
"id": "EH_W",
5950
+
"weight": 0.5
5951
+
},
5952
+
{
5953
+
"a": "EH",
5954
+
"b": "Y",
5955
+
"id": "EH_Y",
5956
+
"weight": 0.5
5957
+
},
5958
+
{
5959
+
"a": "ER",
5960
+
"b": "CH",
5961
+
"id": "ER_CH",
5962
+
"weight": 0.5
5963
+
},
5964
+
{
5965
+
"a": "ER",
5966
+
"b": "JH",
5967
+
"id": "ER_JH",
5968
+
"weight": 0.5
5969
+
},
5970
+
{
5971
+
"a": "ER",
5972
+
"b": "L",
5973
+
"id": "ER_L",
5974
+
"weight": 0.5
5975
+
},
5976
+
{
5977
+
"a": "ER",
5978
+
"b": "R",
5979
+
"id": "ER_R",
5980
+
"weight": 0.5
5981
+
},
5982
+
{
5983
+
"a": "ER",
5984
+
"b": "W",
5985
+
"id": "ER_W",
5986
+
"weight": 0.5
5987
+
},
5988
+
{
5989
+
"a": "ER",
5990
+
"b": "Y",
5991
+
"id": "ER_Y",
5992
+
"weight": 0.5
5993
+
},
5994
+
{
5995
+
"a": "EY",
5996
+
"b": "CH",
5997
+
"id": "EY_CH",
5998
+
"weight": 0.5
5999
+
},
6000
+
{
6001
+
"a": "EY",
6002
+
"b": "JH",
6003
+
"id": "EY_JH",
6004
+
"weight": 0.5
6005
+
},
6006
+
{
6007
+
"a": "EY",
6008
+
"b": "L",
6009
+
"id": "EY_L",
6010
+
"weight": 0.5
6011
+
},
6012
+
{
6013
+
"a": "EY",
6014
+
"b": "R",
6015
+
"id": "EY_R",
6016
+
"weight": 0.5
6017
+
},
6018
+
{
6019
+
"a": "EY",
6020
+
"b": "W",
6021
+
"id": "EY_W",
6022
+
"weight": 0.5
6023
+
},
6024
+
{
6025
+
"a": "EY",
6026
+
"b": "Y",
6027
+
"id": "EY_Y",
6028
+
"weight": 0.5
6029
+
},
6030
+
{
6031
+
"a": "F",
6032
+
"b": "B",
6033
+
"id": "F_B",
6034
+
"weight": 0.5
6035
+
},
6036
+
{
6037
+
"a": "F",
6038
+
"b": "CH",
6039
+
"id": "F_CH",
6040
+
"weight": 0.5
6041
+
},
6042
+
{
6043
+
"a": "F",
6044
+
"b": "D",
6045
+
"id": "F_D",
6046
+
"weight": 0.5
6047
+
},
6048
+
{
6049
+
"a": "F",
6050
+
"b": "DH",
6051
+
"id": "F_DH",
6052
+
"weight": 0.5
6053
+
},
6054
+
{
6055
+
"a": "F",
6056
+
"b": "F",
6057
+
"id": "F_F",
6058
+
"weight": 0.5
6059
+
},
6060
+
{
6061
+
"a": "F",
6062
+
"b": "G",
6063
+
"id": "F_G",
6064
+
"weight": 0.5
6065
+
},
6066
+
{
6067
+
"a": "F",
6068
+
"b": "JH",
6069
+
"id": "F_JH",
6070
+
"weight": 0.5
6071
+
},
6072
+
{
6073
+
"a": "F",
6074
+
"b": "K",
6075
+
"id": "F_K",
6076
+
"weight": 0.5
6077
+
},
6078
+
{
6079
+
"a": "F",
6080
+
"b": "L",
6081
+
"id": "F_L",
6082
+
"weight": 0.5
6083
+
},
6084
+
{
6085
+
"a": "F",
6086
+
"b": "M",
6087
+
"id": "F_M",
6088
+
"weight": 0.5
6089
+
},
6090
+
{
6091
+
"a": "F",
6092
+
"b": "N",
6093
+
"id": "F_N",
6094
+
"weight": 0.5
6095
+
},
6096
+
{
6097
+
"a": "F",
6098
+
"b": "NG",
6099
+
"id": "F_NG",
6100
+
"weight": 0.5
6101
+
},
6102
+
{
6103
+
"a": "F",
6104
+
"b": "P",
6105
+
"id": "F_P",
6106
+
"weight": 0.5
6107
+
},
6108
+
{
6109
+
"a": "F",
6110
+
"b": "R",
6111
+
"id": "F_R",
6112
+
"weight": 0.5
6113
+
},
6114
+
{
6115
+
"a": "F",
6116
+
"b": "S",
6117
+
"id": "F_S",
6118
+
"weight": 0.5
6119
+
},
6120
+
{
6121
+
"a": "F",
6122
+
"b": "SH",
6123
+
"id": "F_SH",
6124
+
"weight": 0.5
6125
+
},
6126
+
{
6127
+
"a": "F",
6128
+
"b": "T",
6129
+
"id": "F_T",
6130
+
"weight": 0.5
6131
+
},
6132
+
{
6133
+
"a": "F",
6134
+
"b": "TH",
6135
+
"id": "F_TH",
6136
+
"weight": 0.5
6137
+
},
6138
+
{
6139
+
"a": "F",
6140
+
"b": "V",
6141
+
"id": "F_V",
6142
+
"weight": 0.5
6143
+
},
6144
+
{
6145
+
"a": "F",
6146
+
"b": "W",
6147
+
"id": "F_W",
6148
+
"weight": 0.5
6149
+
},
6150
+
{
6151
+
"a": "F",
6152
+
"b": "Y",
6153
+
"id": "F_Y",
6154
+
"weight": 0.5
6155
+
},
6156
+
{
6157
+
"a": "F",
6158
+
"b": "Z",
6159
+
"id": "F_Z",
6160
+
"weight": 0.5
6161
+
},
6162
+
{
6163
+
"a": "G",
6164
+
"b": "B",
6165
+
"id": "G_B",
6166
+
"weight": 0.5
6167
+
},
6168
+
{
6169
+
"a": "G",
6170
+
"b": "CH",
6171
+
"id": "G_CH",
6172
+
"weight": 0.5
6173
+
},
6174
+
{
6175
+
"a": "G",
6176
+
"b": "D",
6177
+
"id": "G_D",
6178
+
"weight": 0.5
6179
+
},
6180
+
{
6181
+
"a": "G",
6182
+
"b": "DH",
6183
+
"id": "G_DH",
6184
+
"weight": 0.5
6185
+
},
6186
+
{
6187
+
"a": "G",
6188
+
"b": "F",
6189
+
"id": "G_F",
6190
+
"weight": 0.5
6191
+
},
6192
+
{
6193
+
"a": "G",
6194
+
"b": "G",
6195
+
"id": "G_G",
6196
+
"weight": 0.5
6197
+
},
6198
+
{
6199
+
"a": "G",
6200
+
"b": "JH",
6201
+
"id": "G_JH",
6202
+
"weight": 0.5
6203
+
},
6204
+
{
6205
+
"a": "G",
6206
+
"b": "K",
6207
+
"id": "G_K",
6208
+
"weight": 0.5
6209
+
},
6210
+
{
6211
+
"a": "G",
6212
+
"b": "L",
6213
+
"id": "G_L",
6214
+
"weight": 0.5
6215
+
},
6216
+
{
6217
+
"a": "G",
6218
+
"b": "M",
6219
+
"id": "G_M",
6220
+
"weight": 0.5
6221
+
},
6222
+
{
6223
+
"a": "G",
6224
+
"b": "N",
6225
+
"id": "G_N",
6226
+
"weight": 0.5
6227
+
},
6228
+
{
6229
+
"a": "G",
6230
+
"b": "NG",
6231
+
"id": "G_NG",
6232
+
"weight": 0.5
6233
+
},
6234
+
{
6235
+
"a": "G",
6236
+
"b": "P",
6237
+
"id": "G_P",
6238
+
"weight": 0.5
6239
+
},
6240
+
{
6241
+
"a": "G",
6242
+
"b": "R",
6243
+
"id": "G_R",
6244
+
"weight": 0.5
6245
+
},
6246
+
{
6247
+
"a": "G",
6248
+
"b": "S",
6249
+
"id": "G_S",
6250
+
"weight": 0.5
6251
+
},
6252
+
{
6253
+
"a": "G",
6254
+
"b": "SH",
6255
+
"id": "G_SH",
6256
+
"weight": 0.5
6257
+
},
6258
+
{
6259
+
"a": "G",
6260
+
"b": "T",
6261
+
"id": "G_T",
6262
+
"weight": 0.5
6263
+
},
6264
+
{
6265
+
"a": "G",
6266
+
"b": "TH",
6267
+
"id": "G_TH",
6268
+
"weight": 0.5
6269
+
},
6270
+
{
6271
+
"a": "G",
6272
+
"b": "V",
6273
+
"id": "G_V",
6274
+
"weight": 0.5
6275
+
},
6276
+
{
6277
+
"a": "G",
6278
+
"b": "W",
6279
+
"id": "G_W",
6280
+
"weight": 0.5
6281
+
},
6282
+
{
6283
+
"a": "G",
6284
+
"b": "Y",
6285
+
"id": "G_Y",
6286
+
"weight": 0.5
6287
+
},
6288
+
{
6289
+
"a": "G",
6290
+
"b": "Z",
6291
+
"id": "G_Z",
6292
+
"weight": 0.5
6293
+
},
6294
+
{
6295
+
"a": "HH",
6296
+
"b": "B",
6297
+
"id": "HH_B",
6298
+
"weight": 0.5
6299
+
},
6300
+
{
6301
+
"a": "HH",
6302
+
"b": "CH",
6303
+
"id": "HH_CH",
6304
+
"weight": 0.5
6305
+
},
6306
+
{
6307
+
"a": "HH",
6308
+
"b": "D",
6309
+
"id": "HH_D",
6310
+
"weight": 0.5
6311
+
},
6312
+
{
6313
+
"a": "HH",
6314
+
"b": "DH",
6315
+
"id": "HH_DH",
6316
+
"weight": 0.5
6317
+
},
6318
+
{
6319
+
"a": "HH",
6320
+
"b": "F",
6321
+
"id": "HH_F",
6322
+
"weight": 0.5
6323
+
},
6324
+
{
6325
+
"a": "HH",
6326
+
"b": "G",
6327
+
"id": "HH_G",
6328
+
"weight": 0.5
6329
+
},
6330
+
{
6331
+
"a": "HH",
6332
+
"b": "JH",
6333
+
"id": "HH_JH",
6334
+
"weight": 0.5
6335
+
},
6336
+
{
6337
+
"a": "HH",
6338
+
"b": "K",
6339
+
"id": "HH_K",
6340
+
"weight": 0.5
6341
+
},
6342
+
{
6343
+
"a": "HH",
6344
+
"b": "L",
6345
+
"id": "HH_L",
6346
+
"weight": 0.5
6347
+
},
6348
+
{
6349
+
"a": "HH",
6350
+
"b": "M",
6351
+
"id": "HH_M",
6352
+
"weight": 0.5
6353
+
},
6354
+
{
6355
+
"a": "HH",
6356
+
"b": "N",
6357
+
"id": "HH_N",
6358
+
"weight": 0.5
6359
+
},
6360
+
{
6361
+
"a": "HH",
6362
+
"b": "NG",
6363
+
"id": "HH_NG",
6364
+
"weight": 0.5
6365
+
},
6366
+
{
6367
+
"a": "HH",
6368
+
"b": "P",
6369
+
"id": "HH_P",
6370
+
"weight": 0.5
6371
+
},
6372
+
{
6373
+
"a": "HH",
6374
+
"b": "R",
6375
+
"id": "HH_R",
6376
+
"weight": 0.5
6377
+
},
6378
+
{
6379
+
"a": "HH",
6380
+
"b": "S",
6381
+
"id": "HH_S",
6382
+
"weight": 0.5
6383
+
},
6384
+
{
6385
+
"a": "HH",
6386
+
"b": "SH",
6387
+
"id": "HH_SH",
6388
+
"weight": 0.5
6389
+
},
6390
+
{
6391
+
"a": "HH",
6392
+
"b": "T",
6393
+
"id": "HH_T",
6394
+
"weight": 0.5
6395
+
},
6396
+
{
6397
+
"a": "HH",
6398
+
"b": "TH",
6399
+
"id": "HH_TH",
6400
+
"weight": 0.5
6401
+
},
6402
+
{
6403
+
"a": "HH",
6404
+
"b": "V",
6405
+
"id": "HH_V",
6406
+
"weight": 0.5
6407
+
},
6408
+
{
6409
+
"a": "HH",
6410
+
"b": "W",
6411
+
"id": "HH_W",
6412
+
"weight": 0.5
6413
+
},
6414
+
{
6415
+
"a": "HH",
6416
+
"b": "Y",
6417
+
"id": "HH_Y",
6418
+
"weight": 0.5
6419
+
},
6420
+
{
6421
+
"a": "HH",
6422
+
"b": "Z",
6423
+
"id": "HH_Z",
6424
+
"weight": 0.5
6425
+
},
6426
+
{
6427
+
"a": "IH",
6428
+
"b": "CH",
6429
+
"id": "IH_CH",
6430
+
"weight": 0.5
6431
+
},
6432
+
{
6433
+
"a": "IH",
6434
+
"b": "JH",
6435
+
"id": "IH_JH",
6436
+
"weight": 0.5
6437
+
},
6438
+
{
6439
+
"a": "IH",
6440
+
"b": "L",
6441
+
"id": "IH_L",
6442
+
"weight": 0.5
6443
+
},
6444
+
{
6445
+
"a": "IH",
6446
+
"b": "R",
6447
+
"id": "IH_R",
6448
+
"weight": 0.5
6449
+
},
6450
+
{
6451
+
"a": "IH",
6452
+
"b": "W",
6453
+
"id": "IH_W",
6454
+
"weight": 0.5
6455
+
},
6456
+
{
6457
+
"a": "IH",
6458
+
"b": "Y",
6459
+
"id": "IH_Y",
6460
+
"weight": 0.5
6461
+
},
6462
+
{
6463
+
"a": "IY",
6464
+
"b": "CH",
6465
+
"id": "IY_CH",
6466
+
"weight": 0.5
6467
+
},
6468
+
{
6469
+
"a": "IY",
6470
+
"b": "JH",
6471
+
"id": "IY_JH",
6472
+
"weight": 0.5
6473
+
},
6474
+
{
6475
+
"a": "IY",
6476
+
"b": "L",
6477
+
"id": "IY_L",
6478
+
"weight": 0.5
6479
+
},
6480
+
{
6481
+
"a": "IY",
6482
+
"b": "R",
6483
+
"id": "IY_R",
6484
+
"weight": 0.5
6485
+
},
6486
+
{
6487
+
"a": "IY",
6488
+
"b": "W",
6489
+
"id": "IY_W",
6490
+
"weight": 0.5
6491
+
},
6492
+
{
6493
+
"a": "IY",
6494
+
"b": "Y",
6495
+
"id": "IY_Y",
6496
+
"weight": 0.5
6497
+
},
6498
+
{
6499
+
"a": "JH",
6500
+
"b": "AA",
6501
+
"id": "JH_AA",
6502
+
"weight": 0.5
6503
+
},
6504
+
{
6505
+
"a": "JH",
6506
+
"b": "AE",
6507
+
"id": "JH_AE",
6508
+
"weight": 0.5
6509
+
},
6510
+
{
6511
+
"a": "JH",
6512
+
"b": "AH",
6513
+
"id": "JH_AH",
6514
+
"weight": 0.5
6515
+
},
6516
+
{
6517
+
"a": "JH",
6518
+
"b": "AO",
6519
+
"id": "JH_AO",
6520
+
"weight": 0.5
6521
+
},
6522
+
{
6523
+
"a": "JH",
6524
+
"b": "AW",
6525
+
"id": "JH_AW",
6526
+
"weight": 0.5
6527
+
},
6528
+
{
6529
+
"a": "JH",
6530
+
"b": "AY",
6531
+
"id": "JH_AY",
6532
+
"weight": 0.5
6533
+
},
6534
+
{
6535
+
"a": "JH",
6536
+
"b": "B",
6537
+
"id": "JH_B",
6538
+
"weight": 0.5
6539
+
},
6540
+
{
6541
+
"a": "JH",
6542
+
"b": "CH",
6543
+
"id": "JH_CH",
6544
+
"weight": 0.5
6545
+
},
6546
+
{
6547
+
"a": "JH",
6548
+
"b": "D",
6549
+
"id": "JH_D",
6550
+
"weight": 0.5
6551
+
},
6552
+
{
6553
+
"a": "JH",
6554
+
"b": "DH",
6555
+
"id": "JH_DH",
6556
+
"weight": 0.5
6557
+
},
6558
+
{
6559
+
"a": "JH",
6560
+
"b": "EH",
6561
+
"id": "JH_EH",
6562
+
"weight": 0.5
6563
+
},
6564
+
{
6565
+
"a": "JH",
6566
+
"b": "ER",
6567
+
"id": "JH_ER",
6568
+
"weight": 0.5
6569
+
},
6570
+
{
6571
+
"a": "JH",
6572
+
"b": "EY",
6573
+
"id": "JH_EY",
6574
+
"weight": 0.5
6575
+
},
6576
+
{
6577
+
"a": "JH",
6578
+
"b": "F",
6579
+
"id": "JH_F",
6580
+
"weight": 0.5
6581
+
},
6582
+
{
6583
+
"a": "JH",
6584
+
"b": "G",
6585
+
"id": "JH_G",
6586
+
"weight": 0.5
6587
+
},
6588
+
{
6589
+
"a": "JH",
6590
+
"b": "IH",
6591
+
"id": "JH_IH",
6592
+
"weight": 0.5
6593
+
},
6594
+
{
6595
+
"a": "JH",
6596
+
"b": "IY",
6597
+
"id": "JH_IY",
6598
+
"weight": 0.5
6599
+
},
6600
+
{
6601
+
"a": "JH",
6602
+
"b": "JH",
6603
+
"id": "JH_JH",
6604
+
"weight": 0.5
6605
+
},
6606
+
{
6607
+
"a": "JH",
6608
+
"b": "K",
6609
+
"id": "JH_K",
6610
+
"weight": 0.5
6611
+
},
6612
+
{
6613
+
"a": "JH",
6614
+
"b": "L",
6615
+
"id": "JH_L",
6616
+
"weight": 0.5
6617
+
},
6618
+
{
6619
+
"a": "JH",
6620
+
"b": "M",
6621
+
"id": "JH_M",
6622
+
"weight": 0.5
6623
+
},
6624
+
{
6625
+
"a": "JH",
6626
+
"b": "N",
6627
+
"id": "JH_N",
6628
+
"weight": 0.5
6629
+
},
6630
+
{
6631
+
"a": "JH",
6632
+
"b": "NG",
6633
+
"id": "JH_NG",
6634
+
"weight": 0.5
6635
+
},
6636
+
{
6637
+
"a": "JH",
6638
+
"b": "OW",
6639
+
"id": "JH_OW",
6640
+
"weight": 0.5
6641
+
},
6642
+
{
6643
+
"a": "JH",
6644
+
"b": "OY",
6645
+
"id": "JH_OY",
6646
+
"weight": 0.5
6647
+
},
6648
+
{
6649
+
"a": "JH",
6650
+
"b": "P",
6651
+
"id": "JH_P",
6652
+
"weight": 0.5
6653
+
},
6654
+
{
6655
+
"a": "JH",
6656
+
"b": "R",
6657
+
"id": "JH_R",
6658
+
"weight": 0.5
6659
+
},
6660
+
{
6661
+
"a": "JH",
6662
+
"b": "S",
6663
+
"id": "JH_S",
6664
+
"weight": 0.5
6665
+
},
6666
+
{
6667
+
"a": "JH",
6668
+
"b": "SH",
6669
+
"id": "JH_SH",
6670
+
"weight": 0.5
6671
+
},
6672
+
{
6673
+
"a": "JH",
6674
+
"b": "T",
6675
+
"id": "JH_T",
6676
+
"weight": 0.5
6677
+
},
6678
+
{
6679
+
"a": "JH",
6680
+
"b": "TH",
6681
+
"id": "JH_TH",
6682
+
"weight": 0.5
6683
+
},
6684
+
{
6685
+
"a": "JH",
6686
+
"b": "UH",
6687
+
"id": "JH_UH",
6688
+
"weight": 0.5
6689
+
},
6690
+
{
6691
+
"a": "JH",
6692
+
"b": "UW",
6693
+
"id": "JH_UW",
6694
+
"weight": 0.5
6695
+
},
6696
+
{
6697
+
"a": "JH",
6698
+
"b": "V",
6699
+
"id": "JH_V",
6700
+
"weight": 0.5
6701
+
},
6702
+
{
6703
+
"a": "JH",
6704
+
"b": "W",
6705
+
"id": "JH_W",
6706
+
"weight": 0.5
6707
+
},
6708
+
{
6709
+
"a": "JH",
6710
+
"b": "Y",
6711
+
"id": "JH_Y",
6712
+
"weight": 0.5
6713
+
},
6714
+
{
6715
+
"a": "JH",
6716
+
"b": "Z",
6717
+
"id": "JH_Z",
6718
+
"weight": 0.5
6719
+
},
6720
+
{
6721
+
"a": "K",
6722
+
"b": "B",
6723
+
"id": "K_B",
6724
+
"weight": 0.5
6725
+
},
6726
+
{
6727
+
"a": "K",
6728
+
"b": "CH",
6729
+
"id": "K_CH",
6730
+
"weight": 0.5
6731
+
},
6732
+
{
6733
+
"a": "K",
6734
+
"b": "D",
6735
+
"id": "K_D",
6736
+
"weight": 0.5
6737
+
},
6738
+
{
6739
+
"a": "K",
6740
+
"b": "DH",
6741
+
"id": "K_DH",
6742
+
"weight": 0.5
6743
+
},
6744
+
{
6745
+
"a": "K",
6746
+
"b": "F",
6747
+
"id": "K_F",
6748
+
"weight": 0.5
6749
+
},
6750
+
{
6751
+
"a": "K",
6752
+
"b": "G",
6753
+
"id": "K_G",
6754
+
"weight": 0.5
6755
+
},
6756
+
{
6757
+
"a": "K",
6758
+
"b": "JH",
6759
+
"id": "K_JH",
6760
+
"weight": 0.5
6761
+
},
6762
+
{
6763
+
"a": "K",
6764
+
"b": "K",
6765
+
"id": "K_K",
6766
+
"weight": 0.5
6767
+
},
6768
+
{
6769
+
"a": "K",
6770
+
"b": "L",
6771
+
"id": "K_L",
6772
+
"weight": 0.5
6773
+
},
6774
+
{
6775
+
"a": "K",
6776
+
"b": "M",
6777
+
"id": "K_M",
6778
+
"weight": 0.5
6779
+
},
6780
+
{
6781
+
"a": "K",
6782
+
"b": "N",
6783
+
"id": "K_N",
6784
+
"weight": 0.5
6785
+
},
6786
+
{
6787
+
"a": "K",
6788
+
"b": "NG",
6789
+
"id": "K_NG",
6790
+
"weight": 0.5
6791
+
},
6792
+
{
6793
+
"a": "K",
6794
+
"b": "P",
6795
+
"id": "K_P",
6796
+
"weight": 0.5
6797
+
},
6798
+
{
6799
+
"a": "K",
6800
+
"b": "R",
6801
+
"id": "K_R",
6802
+
"weight": 0.5
6803
+
},
6804
+
{
6805
+
"a": "K",
6806
+
"b": "S",
6807
+
"id": "K_S",
6808
+
"weight": 0.5
6809
+
},
6810
+
{
6811
+
"a": "K",
6812
+
"b": "SH",
6813
+
"id": "K_SH",
6814
+
"weight": 0.5
6815
+
},
6816
+
{
6817
+
"a": "K",
6818
+
"b": "T",
6819
+
"id": "K_T",
6820
+
"weight": 0.5
6821
+
},
6822
+
{
6823
+
"a": "K",
6824
+
"b": "TH",
6825
+
"id": "K_TH",
6826
+
"weight": 0.5
6827
+
},
6828
+
{
6829
+
"a": "K",
6830
+
"b": "V",
6831
+
"id": "K_V",
6832
+
"weight": 0.5
6833
+
},
6834
+
{
6835
+
"a": "K",
6836
+
"b": "W",
6837
+
"id": "K_W",
6838
+
"weight": 0.5
6839
+
},
6840
+
{
6841
+
"a": "K",
6842
+
"b": "Y",
6843
+
"id": "K_Y",
6844
+
"weight": 0.5
6845
+
},
6846
+
{
6847
+
"a": "K",
6848
+
"b": "Z",
6849
+
"id": "K_Z",
6850
+
"weight": 0.5
6851
+
},
6852
+
{
6853
+
"a": "L",
6854
+
"b": "AA",
6855
+
"id": "L_AA",
6856
+
"weight": 0.5
6857
+
},
6858
+
{
6859
+
"a": "L",
6860
+
"b": "AE",
6861
+
"id": "L_AE",
6862
+
"weight": 0.5
6863
+
},
6864
+
{
6865
+
"a": "L",
6866
+
"b": "AH",
6867
+
"id": "L_AH",
6868
+
"weight": 0.5
6869
+
},
6870
+
{
6871
+
"a": "L",
6872
+
"b": "AO",
6873
+
"id": "L_AO",
6874
+
"weight": 0.5
6875
+
},
6876
+
{
6877
+
"a": "L",
6878
+
"b": "AW",
6879
+
"id": "L_AW",
6880
+
"weight": 0.5
6881
+
},
6882
+
{
6883
+
"a": "L",
6884
+
"b": "AY",
6885
+
"id": "L_AY",
6886
+
"weight": 0.5
6887
+
},
6888
+
{
6889
+
"a": "L",
6890
+
"b": "B",
6891
+
"id": "L_B",
6892
+
"weight": 0.5
6893
+
},
6894
+
{
6895
+
"a": "L",
6896
+
"b": "CH",
6897
+
"id": "L_CH",
6898
+
"weight": 0.5
6899
+
},
6900
+
{
6901
+
"a": "L",
6902
+
"b": "D",
6903
+
"id": "L_D",
6904
+
"weight": 0.5
6905
+
},
6906
+
{
6907
+
"a": "L",
6908
+
"b": "DH",
6909
+
"id": "L_DH",
6910
+
"weight": 0.5
6911
+
},
6912
+
{
6913
+
"a": "L",
6914
+
"b": "EH",
6915
+
"id": "L_EH",
6916
+
"weight": 0.5
6917
+
},
6918
+
{
6919
+
"a": "L",
6920
+
"b": "ER",
6921
+
"id": "L_ER",
6922
+
"weight": 0.5
6923
+
},
6924
+
{
6925
+
"a": "L",
6926
+
"b": "EY",
6927
+
"id": "L_EY",
6928
+
"weight": 0.5
6929
+
},
6930
+
{
6931
+
"a": "L",
6932
+
"b": "F",
6933
+
"id": "L_F",
6934
+
"weight": 0.5
6935
+
},
6936
+
{
6937
+
"a": "L",
6938
+
"b": "G",
6939
+
"id": "L_G",
6940
+
"weight": 0.5
6941
+
},
6942
+
{
6943
+
"a": "L",
6944
+
"b": "IH",
6945
+
"id": "L_IH",
6946
+
"weight": 0.5
6947
+
},
6948
+
{
6949
+
"a": "L",
6950
+
"b": "IY",
6951
+
"id": "L_IY",
6952
+
"weight": 0.5
6953
+
},
6954
+
{
6955
+
"a": "L",
6956
+
"b": "JH",
6957
+
"id": "L_JH",
6958
+
"weight": 0.5
6959
+
},
6960
+
{
6961
+
"a": "L",
6962
+
"b": "K",
6963
+
"id": "L_K",
6964
+
"weight": 0.5
6965
+
},
6966
+
{
6967
+
"a": "L",
6968
+
"b": "L",
6969
+
"id": "L_L",
6970
+
"weight": 0.5
6971
+
},
6972
+
{
6973
+
"a": "L",
6974
+
"b": "M",
6975
+
"id": "L_M",
6976
+
"weight": 0.5
6977
+
},
6978
+
{
6979
+
"a": "L",
6980
+
"b": "N",
6981
+
"id": "L_N",
6982
+
"weight": 0.5
6983
+
},
6984
+
{
6985
+
"a": "L",
6986
+
"b": "NG",
6987
+
"id": "L_NG",
6988
+
"weight": 0.5
6989
+
},
6990
+
{
6991
+
"a": "L",
6992
+
"b": "OW",
6993
+
"id": "L_OW",
6994
+
"weight": 0.5
6995
+
},
6996
+
{
6997
+
"a": "L",
6998
+
"b": "OY",
6999
+
"id": "L_OY",
7000
+
"weight": 0.5
7001
+
},
7002
+
{
7003
+
"a": "L",
7004
+
"b": "P",
7005
+
"id": "L_P",
7006
+
"weight": 0.5
7007
+
},
7008
+
{
7009
+
"a": "L",
7010
+
"b": "R",
7011
+
"id": "L_R",
7012
+
"weight": 0.5
7013
+
},
7014
+
{
7015
+
"a": "L",
7016
+
"b": "S",
7017
+
"id": "L_S",
7018
+
"weight": 0.5
7019
+
},
7020
+
{
7021
+
"a": "L",
7022
+
"b": "SH",
7023
+
"id": "L_SH",
7024
+
"weight": 0.5
7025
+
},
7026
+
{
7027
+
"a": "L",
7028
+
"b": "T",
7029
+
"id": "L_T",
7030
+
"weight": 0.5
7031
+
},
7032
+
{
7033
+
"a": "L",
7034
+
"b": "TH",
7035
+
"id": "L_TH",
7036
+
"weight": 0.5
7037
+
},
7038
+
{
7039
+
"a": "L",
7040
+
"b": "UH",
7041
+
"id": "L_UH",
7042
+
"weight": 0.5
7043
+
},
7044
+
{
7045
+
"a": "L",
7046
+
"b": "UW",
7047
+
"id": "L_UW",
7048
+
"weight": 0.5
7049
+
},
7050
+
{
7051
+
"a": "L",
7052
+
"b": "V",
7053
+
"id": "L_V",
7054
+
"weight": 0.5
7055
+
},
7056
+
{
7057
+
"a": "L",
7058
+
"b": "W",
7059
+
"id": "L_W",
7060
+
"weight": 0.5
7061
+
},
7062
+
{
7063
+
"a": "L",
7064
+
"b": "Y",
7065
+
"id": "L_Y",
7066
+
"weight": 0.5
7067
+
},
7068
+
{
7069
+
"a": "L",
7070
+
"b": "Z",
7071
+
"id": "L_Z",
7072
+
"weight": 0.5
7073
+
},
7074
+
{
7075
+
"a": "L",
7076
+
"b": "ZH",
7077
+
"id": "L_ZH",
7078
+
"weight": 0.5
7079
+
},
7080
+
{
7081
+
"a": "M",
7082
+
"b": "B",
7083
+
"id": "M_B",
7084
+
"weight": 0.5
7085
+
},
7086
+
{
7087
+
"a": "M",
7088
+
"b": "CH",
7089
+
"id": "M_CH",
7090
+
"weight": 0.5
7091
+
},
7092
+
{
7093
+
"a": "M",
7094
+
"b": "D",
7095
+
"id": "M_D",
7096
+
"weight": 0.5
7097
+
},
7098
+
{
7099
+
"a": "M",
7100
+
"b": "DH",
7101
+
"id": "M_DH",
7102
+
"weight": 0.5
7103
+
},
7104
+
{
7105
+
"a": "M",
7106
+
"b": "F",
7107
+
"id": "M_F",
7108
+
"weight": 0.5
7109
+
},
7110
+
{
7111
+
"a": "M",
7112
+
"b": "G",
7113
+
"id": "M_G",
7114
+
"weight": 0.5
7115
+
},
7116
+
{
7117
+
"a": "M",
7118
+
"b": "JH",
7119
+
"id": "M_JH",
7120
+
"weight": 0.5
7121
+
},
7122
+
{
7123
+
"a": "M",
7124
+
"b": "K",
7125
+
"id": "M_K",
7126
+
"weight": 0.5
7127
+
},
7128
+
{
7129
+
"a": "M",
7130
+
"b": "L",
7131
+
"id": "M_L",
7132
+
"weight": 0.5
7133
+
},
7134
+
{
7135
+
"a": "M",
7136
+
"b": "M",
7137
+
"id": "M_M",
7138
+
"weight": 0.5
7139
+
},
7140
+
{
7141
+
"a": "M",
7142
+
"b": "N",
7143
+
"id": "M_N",
7144
+
"weight": 0.5
7145
+
},
7146
+
{
7147
+
"a": "M",
7148
+
"b": "NG",
7149
+
"id": "M_NG",
7150
+
"weight": 0.5
7151
+
},
7152
+
{
7153
+
"a": "M",
7154
+
"b": "P",
7155
+
"id": "M_P",
7156
+
"weight": 0.5
7157
+
},
7158
+
{
7159
+
"a": "M",
7160
+
"b": "R",
7161
+
"id": "M_R",
7162
+
"weight": 0.5
7163
+
},
7164
+
{
7165
+
"a": "M",
7166
+
"b": "S",
7167
+
"id": "M_S",
7168
+
"weight": 0.5
7169
+
},
7170
+
{
7171
+
"a": "M",
7172
+
"b": "SH",
7173
+
"id": "M_SH",
7174
+
"weight": 0.5
7175
+
},
7176
+
{
7177
+
"a": "M",
7178
+
"b": "T",
7179
+
"id": "M_T",
7180
+
"weight": 0.5
7181
+
},
7182
+
{
7183
+
"a": "M",
7184
+
"b": "TH",
7185
+
"id": "M_TH",
7186
+
"weight": 0.5
7187
+
},
7188
+
{
7189
+
"a": "M",
7190
+
"b": "V",
7191
+
"id": "M_V",
7192
+
"weight": 0.5
7193
+
},
7194
+
{
7195
+
"a": "M",
7196
+
"b": "W",
7197
+
"id": "M_W",
7198
+
"weight": 0.5
7199
+
},
7200
+
{
7201
+
"a": "M",
7202
+
"b": "Y",
7203
+
"id": "M_Y",
7204
+
"weight": 0.5
7205
+
},
7206
+
{
7207
+
"a": "M",
7208
+
"b": "Z",
7209
+
"id": "M_Z",
7210
+
"weight": 0.5
7211
+
},
7212
+
{
7213
+
"a": "N",
7214
+
"b": "B",
7215
+
"id": "N_B",
7216
+
"weight": 0.5
7217
+
},
7218
+
{
7219
+
"a": "N",
7220
+
"b": "CH",
7221
+
"id": "N_CH",
7222
+
"weight": 0.5
7223
+
},
7224
+
{
7225
+
"a": "N",
7226
+
"b": "D",
7227
+
"id": "N_D",
7228
+
"weight": 0.5
7229
+
},
7230
+
{
7231
+
"a": "N",
7232
+
"b": "DH",
7233
+
"id": "N_DH",
7234
+
"weight": 0.5
7235
+
},
7236
+
{
7237
+
"a": "N",
7238
+
"b": "F",
7239
+
"id": "N_F",
7240
+
"weight": 0.5
7241
+
},
7242
+
{
7243
+
"a": "N",
7244
+
"b": "G",
7245
+
"id": "N_G",
7246
+
"weight": 0.5
7247
+
},
7248
+
{
7249
+
"a": "N",
7250
+
"b": "JH",
7251
+
"id": "N_JH",
7252
+
"weight": 0.5
7253
+
},
7254
+
{
7255
+
"a": "N",
7256
+
"b": "K",
7257
+
"id": "N_K",
7258
+
"weight": 0.5
7259
+
},
7260
+
{
7261
+
"a": "N",
7262
+
"b": "L",
7263
+
"id": "N_L",
7264
+
"weight": 0.5
7265
+
},
7266
+
{
7267
+
"a": "N",
7268
+
"b": "M",
7269
+
"id": "N_M",
7270
+
"weight": 0.5
7271
+
},
7272
+
{
7273
+
"a": "N",
7274
+
"b": "N",
7275
+
"id": "N_N",
7276
+
"weight": 0.5
7277
+
},
7278
+
{
7279
+
"a": "N",
7280
+
"b": "NG",
7281
+
"id": "N_NG",
7282
+
"weight": 0.5
7283
+
},
7284
+
{
7285
+
"a": "N",
7286
+
"b": "P",
7287
+
"id": "N_P",
7288
+
"weight": 0.5
7289
+
},
7290
+
{
7291
+
"a": "N",
7292
+
"b": "R",
7293
+
"id": "N_R",
7294
+
"weight": 0.5
7295
+
},
7296
+
{
7297
+
"a": "N",
7298
+
"b": "S",
7299
+
"id": "N_S",
7300
+
"weight": 0.5
7301
+
},
7302
+
{
7303
+
"a": "N",
7304
+
"b": "SH",
7305
+
"id": "N_SH",
7306
+
"weight": 0.5
7307
+
},
7308
+
{
7309
+
"a": "N",
7310
+
"b": "T",
7311
+
"id": "N_T",
7312
+
"weight": 0.5
7313
+
},
7314
+
{
7315
+
"a": "N",
7316
+
"b": "TH",
7317
+
"id": "N_TH",
7318
+
"weight": 0.5
7319
+
},
7320
+
{
7321
+
"a": "N",
7322
+
"b": "V",
7323
+
"id": "N_V",
7324
+
"weight": 0.5
7325
+
},
7326
+
{
7327
+
"a": "N",
7328
+
"b": "W",
7329
+
"id": "N_W",
7330
+
"weight": 0.5
7331
+
},
7332
+
{
7333
+
"a": "N",
7334
+
"b": "Y",
7335
+
"id": "N_Y",
7336
+
"weight": 0.5
7337
+
},
7338
+
{
7339
+
"a": "N",
7340
+
"b": "Z",
7341
+
"id": "N_Z",
7342
+
"weight": 0.5
7343
+
},
7344
+
{
7345
+
"a": "NG",
7346
+
"b": "G",
7347
+
"id": "NG_G",
7348
+
"weight": 0.5
7349
+
},
7350
+
{
7351
+
"a": "NG",
7352
+
"b": "K",
7353
+
"id": "NG_K",
7354
+
"weight": 0.5
7355
+
},
7356
+
{
7357
+
"a": "OW",
7358
+
"b": "CH",
7359
+
"id": "OW_CH",
7360
+
"weight": 0.5
7361
+
},
7362
+
{
7363
+
"a": "OW",
7364
+
"b": "JH",
7365
+
"id": "OW_JH",
7366
+
"weight": 0.5
7367
+
},
7368
+
{
7369
+
"a": "OW",
7370
+
"b": "L",
7371
+
"id": "OW_L",
7372
+
"weight": 0.5
7373
+
},
7374
+
{
7375
+
"a": "OW",
7376
+
"b": "R",
7377
+
"id": "OW_R",
7378
+
"weight": 0.5
7379
+
},
7380
+
{
7381
+
"a": "OW",
7382
+
"b": "W",
7383
+
"id": "OW_W",
7384
+
"weight": 0.5
7385
+
},
7386
+
{
7387
+
"a": "OW",
7388
+
"b": "Y",
7389
+
"id": "OW_Y",
7390
+
"weight": 0.5
7391
+
},
7392
+
{
7393
+
"a": "OY",
7394
+
"b": "CH",
7395
+
"id": "OY_CH",
7396
+
"weight": 0.5
7397
+
},
7398
+
{
7399
+
"a": "OY",
7400
+
"b": "JH",
7401
+
"id": "OY_JH",
7402
+
"weight": 0.5
7403
+
},
7404
+
{
7405
+
"a": "OY",
7406
+
"b": "L",
7407
+
"id": "OY_L",
7408
+
"weight": 0.5
7409
+
},
7410
+
{
7411
+
"a": "OY",
7412
+
"b": "R",
7413
+
"id": "OY_R",
7414
+
"weight": 0.5
7415
+
},
7416
+
{
7417
+
"a": "OY",
7418
+
"b": "W",
7419
+
"id": "OY_W",
7420
+
"weight": 0.5
7421
+
},
7422
+
{
7423
+
"a": "OY",
7424
+
"b": "Y",
7425
+
"id": "OY_Y",
7426
+
"weight": 0.5
7427
+
},
7428
+
{
7429
+
"a": "P",
7430
+
"b": "B",
7431
+
"id": "P_B",
7432
+
"weight": 0.5
7433
+
},
7434
+
{
7435
+
"a": "P",
7436
+
"b": "CH",
7437
+
"id": "P_CH",
7438
+
"weight": 0.5
7439
+
},
7440
+
{
7441
+
"a": "P",
7442
+
"b": "D",
7443
+
"id": "P_D",
7444
+
"weight": 0.5
7445
+
},
7446
+
{
7447
+
"a": "P",
7448
+
"b": "DH",
7449
+
"id": "P_DH",
7450
+
"weight": 0.5
7451
+
},
7452
+
{
7453
+
"a": "P",
7454
+
"b": "F",
7455
+
"id": "P_F",
7456
+
"weight": 0.5
7457
+
},
7458
+
{
7459
+
"a": "P",
7460
+
"b": "G",
7461
+
"id": "P_G",
7462
+
"weight": 0.5
7463
+
},
7464
+
{
7465
+
"a": "P",
7466
+
"b": "JH",
7467
+
"id": "P_JH",
7468
+
"weight": 0.5
7469
+
},
7470
+
{
7471
+
"a": "P",
7472
+
"b": "K",
7473
+
"id": "P_K",
7474
+
"weight": 0.5
7475
+
},
7476
+
{
7477
+
"a": "P",
7478
+
"b": "L",
7479
+
"id": "P_L",
7480
+
"weight": 0.5
7481
+
},
7482
+
{
7483
+
"a": "P",
7484
+
"b": "M",
7485
+
"id": "P_M",
7486
+
"weight": 0.5
7487
+
},
7488
+
{
7489
+
"a": "P",
7490
+
"b": "N",
7491
+
"id": "P_N",
7492
+
"weight": 0.5
7493
+
},
7494
+
{
7495
+
"a": "P",
7496
+
"b": "NG",
7497
+
"id": "P_NG",
7498
+
"weight": 0.5
7499
+
},
7500
+
{
7501
+
"a": "P",
7502
+
"b": "P",
7503
+
"id": "P_P",
7504
+
"weight": 0.5
7505
+
},
7506
+
{
7507
+
"a": "P",
7508
+
"b": "R",
7509
+
"id": "P_R",
7510
+
"weight": 0.5
7511
+
},
7512
+
{
7513
+
"a": "P",
7514
+
"b": "S",
7515
+
"id": "P_S",
7516
+
"weight": 0.5
7517
+
},
7518
+
{
7519
+
"a": "P",
7520
+
"b": "SH",
7521
+
"id": "P_SH",
7522
+
"weight": 0.5
7523
+
},
7524
+
{
7525
+
"a": "P",
7526
+
"b": "T",
7527
+
"id": "P_T",
7528
+
"weight": 0.5
7529
+
},
7530
+
{
7531
+
"a": "P",
7532
+
"b": "TH",
7533
+
"id": "P_TH",
7534
+
"weight": 0.5
7535
+
},
7536
+
{
7537
+
"a": "P",
7538
+
"b": "V",
7539
+
"id": "P_V",
7540
+
"weight": 0.5
7541
+
},
7542
+
{
7543
+
"a": "P",
7544
+
"b": "W",
7545
+
"id": "P_W",
7546
+
"weight": 0.5
7547
+
},
7548
+
{
7549
+
"a": "P",
7550
+
"b": "Y",
7551
+
"id": "P_Y",
7552
+
"weight": 0.5
7553
+
},
7554
+
{
7555
+
"a": "P",
7556
+
"b": "Z",
7557
+
"id": "P_Z",
7558
+
"weight": 0.5
7559
+
},
7560
+
{
7561
+
"a": "R",
7562
+
"b": "AA",
7563
+
"id": "R_AA",
7564
+
"weight": 0.5
7565
+
},
7566
+
{
7567
+
"a": "R",
7568
+
"b": "AE",
7569
+
"id": "R_AE",
7570
+
"weight": 0.5
7571
+
},
7572
+
{
7573
+
"a": "R",
7574
+
"b": "AH",
7575
+
"id": "R_AH",
7576
+
"weight": 0.5
7577
+
},
7578
+
{
7579
+
"a": "R",
7580
+
"b": "AO",
7581
+
"id": "R_AO",
7582
+
"weight": 0.5
7583
+
},
7584
+
{
7585
+
"a": "R",
7586
+
"b": "AW",
7587
+
"id": "R_AW",
7588
+
"weight": 0.5
7589
+
},
7590
+
{
7591
+
"a": "R",
7592
+
"b": "AY",
7593
+
"id": "R_AY",
7594
+
"weight": 0.5
7595
+
},
7596
+
{
7597
+
"a": "R",
7598
+
"b": "B",
7599
+
"id": "R_B",
7600
+
"weight": 0.5
7601
+
},
7602
+
{
7603
+
"a": "R",
7604
+
"b": "CH",
7605
+
"id": "R_CH",
7606
+
"weight": 0.5
7607
+
},
7608
+
{
7609
+
"a": "R",
7610
+
"b": "D",
7611
+
"id": "R_D",
7612
+
"weight": 0.5
7613
+
},
7614
+
{
7615
+
"a": "R",
7616
+
"b": "DH",
7617
+
"id": "R_DH",
7618
+
"weight": 0.5
7619
+
},
7620
+
{
7621
+
"a": "R",
7622
+
"b": "EH",
7623
+
"id": "R_EH",
7624
+
"weight": 0.5
7625
+
},
7626
+
{
7627
+
"a": "R",
7628
+
"b": "ER",
7629
+
"id": "R_ER",
7630
+
"weight": 0.5
7631
+
},
7632
+
{
7633
+
"a": "R",
7634
+
"b": "EY",
7635
+
"id": "R_EY",
7636
+
"weight": 0.5
7637
+
},
7638
+
{
7639
+
"a": "R",
7640
+
"b": "F",
7641
+
"id": "R_F",
7642
+
"weight": 0.5
7643
+
},
7644
+
{
7645
+
"a": "R",
7646
+
"b": "G",
7647
+
"id": "R_G",
7648
+
"weight": 0.5
7649
+
},
7650
+
{
7651
+
"a": "R",
7652
+
"b": "IH",
7653
+
"id": "R_IH",
7654
+
"weight": 0.5
7655
+
},
7656
+
{
7657
+
"a": "R",
7658
+
"b": "IY",
7659
+
"id": "R_IY",
7660
+
"weight": 0.5
7661
+
},
7662
+
{
7663
+
"a": "R",
7664
+
"b": "JH",
7665
+
"id": "R_JH",
7666
+
"weight": 0.5
7667
+
},
7668
+
{
7669
+
"a": "R",
7670
+
"b": "K",
7671
+
"id": "R_K",
7672
+
"weight": 0.5
7673
+
},
7674
+
{
7675
+
"a": "R",
7676
+
"b": "L",
7677
+
"id": "R_L",
7678
+
"weight": 0.5
7679
+
},
7680
+
{
7681
+
"a": "R",
7682
+
"b": "M",
7683
+
"id": "R_M",
7684
+
"weight": 0.5
7685
+
},
7686
+
{
7687
+
"a": "R",
7688
+
"b": "N",
7689
+
"id": "R_N",
7690
+
"weight": 0.5
7691
+
},
7692
+
{
7693
+
"a": "R",
7694
+
"b": "NG",
7695
+
"id": "R_NG",
7696
+
"weight": 0.5
7697
+
},
7698
+
{
7699
+
"a": "R",
7700
+
"b": "OW",
7701
+
"id": "R_OW",
7702
+
"weight": 0.5
7703
+
},
7704
+
{
7705
+
"a": "R",
7706
+
"b": "OY",
7707
+
"id": "R_OY",
7708
+
"weight": 0.5
7709
+
},
7710
+
{
7711
+
"a": "R",
7712
+
"b": "P",
7713
+
"id": "R_P",
7714
+
"weight": 0.5
7715
+
},
7716
+
{
7717
+
"a": "R",
7718
+
"b": "R",
7719
+
"id": "R_R",
7720
+
"weight": 0.5
7721
+
},
7722
+
{
7723
+
"a": "R",
7724
+
"b": "S",
7725
+
"id": "R_S",
7726
+
"weight": 0.5
7727
+
},
7728
+
{
7729
+
"a": "R",
7730
+
"b": "SH",
7731
+
"id": "R_SH",
7732
+
"weight": 0.5
7733
+
},
7734
+
{
7735
+
"a": "R",
7736
+
"b": "T",
7737
+
"id": "R_T",
7738
+
"weight": 0.5
7739
+
},
7740
+
{
7741
+
"a": "R",
7742
+
"b": "TH",
7743
+
"id": "R_TH",
7744
+
"weight": 0.5
7745
+
},
7746
+
{
7747
+
"a": "R",
7748
+
"b": "UH",
7749
+
"id": "R_UH",
7750
+
"weight": 0.5
7751
+
},
7752
+
{
7753
+
"a": "R",
7754
+
"b": "UW",
7755
+
"id": "R_UW",
7756
+
"weight": 0.5
7757
+
},
7758
+
{
7759
+
"a": "R",
7760
+
"b": "V",
7761
+
"id": "R_V",
7762
+
"weight": 0.5
7763
+
},
7764
+
{
7765
+
"a": "R",
7766
+
"b": "W",
7767
+
"id": "R_W",
7768
+
"weight": 0.5
7769
+
},
7770
+
{
7771
+
"a": "R",
7772
+
"b": "Y",
7773
+
"id": "R_Y",
7774
+
"weight": 0.5
7775
+
},
7776
+
{
7777
+
"a": "R",
7778
+
"b": "Z",
7779
+
"id": "R_Z",
7780
+
"weight": 0.5
7781
+
},
7782
+
{
7783
+
"a": "S",
7784
+
"b": "B",
7785
+
"id": "S_B",
7786
+
"weight": 0.5
7787
+
},
7788
+
{
7789
+
"a": "S",
7790
+
"b": "CH",
7791
+
"id": "S_CH",
7792
+
"weight": 0.5
7793
+
},
7794
+
{
7795
+
"a": "S",
7796
+
"b": "D",
7797
+
"id": "S_D",
7798
+
"weight": 0.5
7799
+
},
7800
+
{
7801
+
"a": "S",
7802
+
"b": "DH",
7803
+
"id": "S_DH",
7804
+
"weight": 0.5
7805
+
},
7806
+
{
7807
+
"a": "S",
7808
+
"b": "F",
7809
+
"id": "S_F",
7810
+
"weight": 0.5
7811
+
},
7812
+
{
7813
+
"a": "S",
7814
+
"b": "G",
7815
+
"id": "S_G",
7816
+
"weight": 0.5
7817
+
},
7818
+
{
7819
+
"a": "S",
7820
+
"b": "JH",
7821
+
"id": "S_JH",
7822
+
"weight": 0.5
7823
+
},
7824
+
{
7825
+
"a": "S",
7826
+
"b": "K",
7827
+
"id": "S_K",
7828
+
"weight": 0.5
7829
+
},
7830
+
{
7831
+
"a": "S",
7832
+
"b": "L",
7833
+
"id": "S_L",
7834
+
"weight": 0.5
7835
+
},
7836
+
{
7837
+
"a": "S",
7838
+
"b": "M",
7839
+
"id": "S_M",
7840
+
"weight": 0.5
7841
+
},
7842
+
{
7843
+
"a": "S",
7844
+
"b": "N",
7845
+
"id": "S_N",
7846
+
"weight": 0.5
7847
+
},
7848
+
{
7849
+
"a": "S",
7850
+
"b": "NG",
7851
+
"id": "S_NG",
7852
+
"weight": 0.5
7853
+
},
7854
+
{
7855
+
"a": "S",
7856
+
"b": "P",
7857
+
"id": "S_P",
7858
+
"weight": 0.5
7859
+
},
7860
+
{
7861
+
"a": "S",
7862
+
"b": "R",
7863
+
"id": "S_R",
7864
+
"weight": 0.5
7865
+
},
7866
+
{
7867
+
"a": "S",
7868
+
"b": "S",
7869
+
"id": "S_S",
7870
+
"weight": 0.5
7871
+
},
7872
+
{
7873
+
"a": "S",
7874
+
"b": "SH",
7875
+
"id": "S_SH",
7876
+
"weight": 0.5
7877
+
},
7878
+
{
7879
+
"a": "S",
7880
+
"b": "T",
7881
+
"id": "S_T",
7882
+
"weight": 0.5
7883
+
},
7884
+
{
7885
+
"a": "S",
7886
+
"b": "TH",
7887
+
"id": "S_TH",
7888
+
"weight": 0.5
7889
+
},
7890
+
{
7891
+
"a": "S",
7892
+
"b": "V",
7893
+
"id": "S_V",
7894
+
"weight": 0.5
7895
+
},
7896
+
{
7897
+
"a": "S",
7898
+
"b": "W",
7899
+
"id": "S_W",
7900
+
"weight": 0.5
7901
+
},
7902
+
{
7903
+
"a": "S",
7904
+
"b": "Y",
7905
+
"id": "S_Y",
7906
+
"weight": 0.5
7907
+
},
7908
+
{
7909
+
"a": "S",
7910
+
"b": "Z",
7911
+
"id": "S_Z",
7912
+
"weight": 0.5
7913
+
},
7914
+
{
7915
+
"a": "SH",
7916
+
"b": "B",
7917
+
"id": "SH_B",
7918
+
"weight": 0.5
7919
+
},
7920
+
{
7921
+
"a": "SH",
7922
+
"b": "CH",
7923
+
"id": "SH_CH",
7924
+
"weight": 0.5
7925
+
},
7926
+
{
7927
+
"a": "SH",
7928
+
"b": "D",
7929
+
"id": "SH_D",
7930
+
"weight": 0.5
7931
+
},
7932
+
{
7933
+
"a": "SH",
7934
+
"b": "DH",
7935
+
"id": "SH_DH",
7936
+
"weight": 0.5
7937
+
},
7938
+
{
7939
+
"a": "SH",
7940
+
"b": "F",
7941
+
"id": "SH_F",
7942
+
"weight": 0.5
7943
+
},
7944
+
{
7945
+
"a": "SH",
7946
+
"b": "G",
7947
+
"id": "SH_G",
7948
+
"weight": 0.5
7949
+
},
7950
+
{
7951
+
"a": "SH",
7952
+
"b": "JH",
7953
+
"id": "SH_JH",
7954
+
"weight": 0.5
7955
+
},
7956
+
{
7957
+
"a": "SH",
7958
+
"b": "K",
7959
+
"id": "SH_K",
7960
+
"weight": 0.5
7961
+
},
7962
+
{
7963
+
"a": "SH",
7964
+
"b": "L",
7965
+
"id": "SH_L",
7966
+
"weight": 0.5
7967
+
},
7968
+
{
7969
+
"a": "SH",
7970
+
"b": "M",
7971
+
"id": "SH_M",
7972
+
"weight": 0.5
7973
+
},
7974
+
{
7975
+
"a": "SH",
7976
+
"b": "N",
7977
+
"id": "SH_N",
7978
+
"weight": 0.5
7979
+
},
7980
+
{
7981
+
"a": "SH",
7982
+
"b": "NG",
7983
+
"id": "SH_NG",
7984
+
"weight": 0.5
7985
+
},
7986
+
{
7987
+
"a": "SH",
7988
+
"b": "P",
7989
+
"id": "SH_P",
7990
+
"weight": 0.5
7991
+
},
7992
+
{
7993
+
"a": "SH",
7994
+
"b": "R",
7995
+
"id": "SH_R",
7996
+
"weight": 0.5
7997
+
},
7998
+
{
7999
+
"a": "SH",
8000
+
"b": "S",
8001
+
"id": "SH_S",
8002
+
"weight": 0.5
8003
+
},
8004
+
{
8005
+
"a": "SH",
8006
+
"b": "SH",
8007
+
"id": "SH_SH",
8008
+
"weight": 0.5
8009
+
},
8010
+
{
8011
+
"a": "SH",
8012
+
"b": "T",
8013
+
"id": "SH_T",
8014
+
"weight": 0.5
8015
+
},
8016
+
{
8017
+
"a": "SH",
8018
+
"b": "TH",
8019
+
"id": "SH_TH",
8020
+
"weight": 0.5
8021
+
},
8022
+
{
8023
+
"a": "SH",
8024
+
"b": "V",
8025
+
"id": "SH_V",
8026
+
"weight": 0.5
8027
+
},
8028
+
{
8029
+
"a": "SH",
8030
+
"b": "W",
8031
+
"id": "SH_W",
8032
+
"weight": 0.5
8033
+
},
8034
+
{
8035
+
"a": "SH",
8036
+
"b": "Y",
8037
+
"id": "SH_Y",
8038
+
"weight": 0.5
8039
+
},
8040
+
{
8041
+
"a": "SH",
8042
+
"b": "Z",
8043
+
"id": "SH_Z",
8044
+
"weight": 0.5
8045
+
},
8046
+
{
8047
+
"a": "T",
8048
+
"b": "B",
8049
+
"id": "T_B",
8050
+
"weight": 0.5
8051
+
},
8052
+
{
8053
+
"a": "T",
8054
+
"b": "CH",
8055
+
"id": "T_CH",
8056
+
"weight": 0.5
8057
+
},
8058
+
{
8059
+
"a": "T",
8060
+
"b": "D",
8061
+
"id": "T_D",
8062
+
"weight": 0.5
8063
+
},
8064
+
{
8065
+
"a": "T",
8066
+
"b": "DH",
8067
+
"id": "T_DH",
8068
+
"weight": 0.5
8069
+
},
8070
+
{
8071
+
"a": "T",
8072
+
"b": "F",
8073
+
"id": "T_F",
8074
+
"weight": 0.5
8075
+
},
8076
+
{
8077
+
"a": "T",
8078
+
"b": "G",
8079
+
"id": "T_G",
8080
+
"weight": 0.5
8081
+
},
8082
+
{
8083
+
"a": "T",
8084
+
"b": "JH",
8085
+
"id": "T_JH",
8086
+
"weight": 0.5
8087
+
},
8088
+
{
8089
+
"a": "T",
8090
+
"b": "K",
8091
+
"id": "T_K",
8092
+
"weight": 0.5
8093
+
},
8094
+
{
8095
+
"a": "T",
8096
+
"b": "L",
8097
+
"id": "T_L",
8098
+
"weight": 0.5
8099
+
},
8100
+
{
8101
+
"a": "T",
8102
+
"b": "M",
8103
+
"id": "T_M",
8104
+
"weight": 0.5
8105
+
},
8106
+
{
8107
+
"a": "T",
8108
+
"b": "N",
8109
+
"id": "T_N",
8110
+
"weight": 0.5
8111
+
},
8112
+
{
8113
+
"a": "T",
8114
+
"b": "NG",
8115
+
"id": "T_NG",
8116
+
"weight": 0.5
8117
+
},
8118
+
{
8119
+
"a": "T",
8120
+
"b": "P",
8121
+
"id": "T_P",
8122
+
"weight": 0.5
8123
+
},
8124
+
{
8125
+
"a": "T",
8126
+
"b": "R",
8127
+
"id": "T_R",
8128
+
"weight": 0.5
8129
+
},
8130
+
{
8131
+
"a": "T",
8132
+
"b": "S",
8133
+
"id": "T_S",
8134
+
"weight": 0.5
8135
+
},
8136
+
{
8137
+
"a": "T",
8138
+
"b": "SH",
8139
+
"id": "T_SH",
8140
+
"weight": 0.5
8141
+
},
8142
+
{
8143
+
"a": "T",
8144
+
"b": "T",
8145
+
"id": "T_T",
8146
+
"weight": 0.5
8147
+
},
8148
+
{
8149
+
"a": "T",
8150
+
"b": "TH",
8151
+
"id": "T_TH",
8152
+
"weight": 0.5
8153
+
},
8154
+
{
8155
+
"a": "T",
8156
+
"b": "V",
8157
+
"id": "T_V",
8158
+
"weight": 0.5
8159
+
},
8160
+
{
8161
+
"a": "T",
8162
+
"b": "W",
8163
+
"id": "T_W",
8164
+
"weight": 0.5
8165
+
},
8166
+
{
8167
+
"a": "T",
8168
+
"b": "Y",
8169
+
"id": "T_Y",
8170
+
"weight": 0.5
8171
+
},
8172
+
{
8173
+
"a": "T",
8174
+
"b": "Z",
8175
+
"id": "T_Z",
8176
+
"weight": 0.5
8177
+
},
8178
+
{
8179
+
"a": "TH",
8180
+
"b": "B",
8181
+
"id": "TH_B",
8182
+
"weight": 0.5
8183
+
},
8184
+
{
8185
+
"a": "TH",
8186
+
"b": "CH",
8187
+
"id": "TH_CH",
8188
+
"weight": 0.5
8189
+
},
8190
+
{
8191
+
"a": "TH",
8192
+
"b": "D",
8193
+
"id": "TH_D",
8194
+
"weight": 0.5
8195
+
},
8196
+
{
8197
+
"a": "TH",
8198
+
"b": "DH",
8199
+
"id": "TH_DH",
8200
+
"weight": 0.5
8201
+
},
8202
+
{
8203
+
"a": "TH",
8204
+
"b": "F",
8205
+
"id": "TH_F",
8206
+
"weight": 0.5
8207
+
},
8208
+
{
8209
+
"a": "TH",
8210
+
"b": "G",
8211
+
"id": "TH_G",
8212
+
"weight": 0.5
8213
+
},
8214
+
{
8215
+
"a": "TH",
8216
+
"b": "JH",
8217
+
"id": "TH_JH",
8218
+
"weight": 0.5
8219
+
},
8220
+
{
8221
+
"a": "TH",
8222
+
"b": "K",
8223
+
"id": "TH_K",
8224
+
"weight": 0.5
8225
+
},
8226
+
{
8227
+
"a": "TH",
8228
+
"b": "L",
8229
+
"id": "TH_L",
8230
+
"weight": 0.5
8231
+
},
8232
+
{
8233
+
"a": "TH",
8234
+
"b": "M",
8235
+
"id": "TH_M",
8236
+
"weight": 0.5
8237
+
},
8238
+
{
8239
+
"a": "TH",
8240
+
"b": "N",
8241
+
"id": "TH_N",
8242
+
"weight": 0.5
8243
+
},
8244
+
{
8245
+
"a": "TH",
8246
+
"b": "NG",
8247
+
"id": "TH_NG",
8248
+
"weight": 0.5
8249
+
},
8250
+
{
8251
+
"a": "TH",
8252
+
"b": "P",
8253
+
"id": "TH_P",
8254
+
"weight": 0.5
8255
+
},
8256
+
{
8257
+
"a": "TH",
8258
+
"b": "R",
8259
+
"id": "TH_R",
8260
+
"weight": 0.5
8261
+
},
8262
+
{
8263
+
"a": "TH",
8264
+
"b": "S",
8265
+
"id": "TH_S",
8266
+
"weight": 0.5
8267
+
},
8268
+
{
8269
+
"a": "TH",
8270
+
"b": "SH",
8271
+
"id": "TH_SH",
8272
+
"weight": 0.5
8273
+
},
8274
+
{
8275
+
"a": "TH",
8276
+
"b": "T",
8277
+
"id": "TH_T",
8278
+
"weight": 0.5
8279
+
},
8280
+
{
8281
+
"a": "TH",
8282
+
"b": "TH",
8283
+
"id": "TH_TH",
8284
+
"weight": 0.5
8285
+
},
8286
+
{
8287
+
"a": "TH",
8288
+
"b": "V",
8289
+
"id": "TH_V",
8290
+
"weight": 0.5
8291
+
},
8292
+
{
8293
+
"a": "TH",
8294
+
"b": "W",
8295
+
"id": "TH_W",
8296
+
"weight": 0.5
8297
+
},
8298
+
{
8299
+
"a": "TH",
8300
+
"b": "Y",
8301
+
"id": "TH_Y",
8302
+
"weight": 0.5
8303
+
},
8304
+
{
8305
+
"a": "TH",
8306
+
"b": "Z",
8307
+
"id": "TH_Z",
8308
+
"weight": 0.5
8309
+
},
8310
+
{
8311
+
"a": "UH",
8312
+
"b": "CH",
8313
+
"id": "UH_CH",
8314
+
"weight": 0.5
8315
+
},
8316
+
{
8317
+
"a": "UH",
8318
+
"b": "JH",
8319
+
"id": "UH_JH",
8320
+
"weight": 0.5
8321
+
},
8322
+
{
8323
+
"a": "UH",
8324
+
"b": "L",
8325
+
"id": "UH_L",
8326
+
"weight": 0.5
8327
+
},
8328
+
{
8329
+
"a": "UH",
8330
+
"b": "R",
8331
+
"id": "UH_R",
8332
+
"weight": 0.5
8333
+
},
8334
+
{
8335
+
"a": "UH",
8336
+
"b": "W",
8337
+
"id": "UH_W",
8338
+
"weight": 0.5
8339
+
},
8340
+
{
8341
+
"a": "UH",
8342
+
"b": "Y",
8343
+
"id": "UH_Y",
8344
+
"weight": 0.5
8345
+
},
8346
+
{
8347
+
"a": "UW",
8348
+
"b": "CH",
8349
+
"id": "UW_CH",
8350
+
"weight": 0.5
8351
+
},
8352
+
{
8353
+
"a": "UW",
8354
+
"b": "JH",
8355
+
"id": "UW_JH",
8356
+
"weight": 0.5
8357
+
},
8358
+
{
8359
+
"a": "UW",
8360
+
"b": "L",
8361
+
"id": "UW_L",
8362
+
"weight": 0.5
8363
+
},
8364
+
{
8365
+
"a": "UW",
8366
+
"b": "R",
8367
+
"id": "UW_R",
8368
+
"weight": 0.5
8369
+
},
8370
+
{
8371
+
"a": "UW",
8372
+
"b": "W",
8373
+
"id": "UW_W",
8374
+
"weight": 0.5
8375
+
},
8376
+
{
8377
+
"a": "UW",
8378
+
"b": "Y",
8379
+
"id": "UW_Y",
8380
+
"weight": 0.5
8381
+
},
8382
+
{
8383
+
"a": "V",
8384
+
"b": "B",
8385
+
"id": "V_B",
8386
+
"weight": 0.5
8387
+
},
8388
+
{
8389
+
"a": "V",
8390
+
"b": "CH",
8391
+
"id": "V_CH",
8392
+
"weight": 0.5
8393
+
},
8394
+
{
8395
+
"a": "V",
8396
+
"b": "D",
8397
+
"id": "V_D",
8398
+
"weight": 0.5
8399
+
},
8400
+
{
8401
+
"a": "V",
8402
+
"b": "DH",
8403
+
"id": "V_DH",
8404
+
"weight": 0.5
8405
+
},
8406
+
{
8407
+
"a": "V",
8408
+
"b": "F",
8409
+
"id": "V_F",
8410
+
"weight": 0.5
8411
+
},
8412
+
{
8413
+
"a": "V",
8414
+
"b": "G",
8415
+
"id": "V_G",
8416
+
"weight": 0.5
8417
+
},
8418
+
{
8419
+
"a": "V",
8420
+
"b": "JH",
8421
+
"id": "V_JH",
8422
+
"weight": 0.5
8423
+
},
8424
+
{
8425
+
"a": "V",
8426
+
"b": "K",
8427
+
"id": "V_K",
8428
+
"weight": 0.5
8429
+
},
8430
+
{
8431
+
"a": "V",
8432
+
"b": "L",
8433
+
"id": "V_L",
8434
+
"weight": 0.5
8435
+
},
8436
+
{
8437
+
"a": "V",
8438
+
"b": "M",
8439
+
"id": "V_M",
8440
+
"weight": 0.5
8441
+
},
8442
+
{
8443
+
"a": "V",
8444
+
"b": "N",
8445
+
"id": "V_N",
8446
+
"weight": 0.5
8447
+
},
8448
+
{
8449
+
"a": "V",
8450
+
"b": "NG",
8451
+
"id": "V_NG",
8452
+
"weight": 0.5
8453
+
},
8454
+
{
8455
+
"a": "V",
8456
+
"b": "P",
8457
+
"id": "V_P",
8458
+
"weight": 0.5
8459
+
},
8460
+
{
8461
+
"a": "V",
8462
+
"b": "R",
8463
+
"id": "V_R",
8464
+
"weight": 0.5
8465
+
},
8466
+
{
8467
+
"a": "V",
8468
+
"b": "S",
8469
+
"id": "V_S",
8470
+
"weight": 0.5
8471
+
},
8472
+
{
8473
+
"a": "V",
8474
+
"b": "SH",
8475
+
"id": "V_SH",
8476
+
"weight": 0.5
8477
+
},
8478
+
{
8479
+
"a": "V",
8480
+
"b": "T",
8481
+
"id": "V_T",
8482
+
"weight": 0.5
8483
+
},
8484
+
{
8485
+
"a": "V",
8486
+
"b": "TH",
8487
+
"id": "V_TH",
8488
+
"weight": 0.5
8489
+
},
8490
+
{
8491
+
"a": "V",
8492
+
"b": "V",
8493
+
"id": "V_V",
8494
+
"weight": 0.5
8495
+
},
8496
+
{
8497
+
"a": "V",
8498
+
"b": "W",
8499
+
"id": "V_W",
8500
+
"weight": 0.5
8501
+
},
8502
+
{
8503
+
"a": "V",
8504
+
"b": "Y",
8505
+
"id": "V_Y",
8506
+
"weight": 0.5
8507
+
},
8508
+
{
8509
+
"a": "V",
8510
+
"b": "Z",
8511
+
"id": "V_Z",
8512
+
"weight": 0.5
8513
+
},
8514
+
{
8515
+
"a": "W",
8516
+
"b": "AA",
8517
+
"id": "W_AA",
8518
+
"weight": 0.5
8519
+
},
8520
+
{
8521
+
"a": "W",
8522
+
"b": "AE",
8523
+
"id": "W_AE",
8524
+
"weight": 0.5
8525
+
},
8526
+
{
8527
+
"a": "W",
8528
+
"b": "AH",
8529
+
"id": "W_AH",
8530
+
"weight": 0.5
8531
+
},
8532
+
{
8533
+
"a": "W",
8534
+
"b": "AO",
8535
+
"id": "W_AO",
8536
+
"weight": 0.5
8537
+
},
8538
+
{
8539
+
"a": "W",
8540
+
"b": "AW",
8541
+
"id": "W_AW",
8542
+
"weight": 0.5
8543
+
},
8544
+
{
8545
+
"a": "W",
8546
+
"b": "AY",
8547
+
"id": "W_AY",
8548
+
"weight": 0.5
8549
+
},
8550
+
{
8551
+
"a": "W",
8552
+
"b": "B",
8553
+
"id": "W_B",
8554
+
"weight": 0.5
8555
+
},
8556
+
{
8557
+
"a": "W",
8558
+
"b": "CH",
8559
+
"id": "W_CH",
8560
+
"weight": 0.5
8561
+
},
8562
+
{
8563
+
"a": "W",
8564
+
"b": "D",
8565
+
"id": "W_D",
8566
+
"weight": 0.5
8567
+
},
8568
+
{
8569
+
"a": "W",
8570
+
"b": "DH",
8571
+
"id": "W_DH",
8572
+
"weight": 0.5
8573
+
},
8574
+
{
8575
+
"a": "W",
8576
+
"b": "EH",
8577
+
"id": "W_EH",
8578
+
"weight": 0.5
8579
+
},
8580
+
{
8581
+
"a": "W",
8582
+
"b": "ER",
8583
+
"id": "W_ER",
8584
+
"weight": 0.5
8585
+
},
8586
+
{
8587
+
"a": "W",
8588
+
"b": "EY",
8589
+
"id": "W_EY",
8590
+
"weight": 0.5
8591
+
},
8592
+
{
8593
+
"a": "W",
8594
+
"b": "F",
8595
+
"id": "W_F",
8596
+
"weight": 0.5
8597
+
},
8598
+
{
8599
+
"a": "W",
8600
+
"b": "G",
8601
+
"id": "W_G",
8602
+
"weight": 0.5
8603
+
},
8604
+
{
8605
+
"a": "W",
8606
+
"b": "IH",
8607
+
"id": "W_IH",
8608
+
"weight": 0.5
8609
+
},
8610
+
{
8611
+
"a": "W",
8612
+
"b": "IY",
8613
+
"id": "W_IY",
8614
+
"weight": 0.5
8615
+
},
8616
+
{
8617
+
"a": "W",
8618
+
"b": "JH",
8619
+
"id": "W_JH",
8620
+
"weight": 0.5
8621
+
},
8622
+
{
8623
+
"a": "W",
8624
+
"b": "K",
8625
+
"id": "W_K",
8626
+
"weight": 0.5
8627
+
},
8628
+
{
8629
+
"a": "W",
8630
+
"b": "L",
8631
+
"id": "W_L",
8632
+
"weight": 0.5
8633
+
},
8634
+
{
8635
+
"a": "W",
8636
+
"b": "M",
8637
+
"id": "W_M",
8638
+
"weight": 0.5
8639
+
},
8640
+
{
8641
+
"a": "W",
8642
+
"b": "N",
8643
+
"id": "W_N",
8644
+
"weight": 0.5
8645
+
},
8646
+
{
8647
+
"a": "W",
8648
+
"b": "NG",
8649
+
"id": "W_NG",
8650
+
"weight": 0.5
8651
+
},
8652
+
{
8653
+
"a": "W",
8654
+
"b": "OW",
8655
+
"id": "W_OW",
8656
+
"weight": 0.5
8657
+
},
8658
+
{
8659
+
"a": "W",
8660
+
"b": "OY",
8661
+
"id": "W_OY",
8662
+
"weight": 0.5
8663
+
},
8664
+
{
8665
+
"a": "W",
8666
+
"b": "P",
8667
+
"id": "W_P",
8668
+
"weight": 0.5
8669
+
},
8670
+
{
8671
+
"a": "W",
8672
+
"b": "R",
8673
+
"id": "W_R",
8674
+
"weight": 0.5
8675
+
},
8676
+
{
8677
+
"a": "W",
8678
+
"b": "S",
8679
+
"id": "W_S",
8680
+
"weight": 0.5
8681
+
},
8682
+
{
8683
+
"a": "W",
8684
+
"b": "SH",
8685
+
"id": "W_SH",
8686
+
"weight": 0.5
8687
+
},
8688
+
{
8689
+
"a": "W",
8690
+
"b": "T",
8691
+
"id": "W_T",
8692
+
"weight": 0.5
8693
+
},
8694
+
{
8695
+
"a": "W",
8696
+
"b": "TH",
8697
+
"id": "W_TH",
8698
+
"weight": 0.5
8699
+
},
8700
+
{
8701
+
"a": "W",
8702
+
"b": "UH",
8703
+
"id": "W_UH",
8704
+
"weight": 0.5
8705
+
},
8706
+
{
8707
+
"a": "W",
8708
+
"b": "UW",
8709
+
"id": "W_UW",
8710
+
"weight": 0.5
8711
+
},
8712
+
{
8713
+
"a": "W",
8714
+
"b": "V",
8715
+
"id": "W_V",
8716
+
"weight": 0.5
8717
+
},
8718
+
{
8719
+
"a": "W",
8720
+
"b": "W",
8721
+
"id": "W_W",
8722
+
"weight": 0.5
8723
+
},
8724
+
{
8725
+
"a": "W",
8726
+
"b": "Y",
8727
+
"id": "W_Y",
8728
+
"weight": 0.5
8729
+
},
8730
+
{
8731
+
"a": "W",
8732
+
"b": "Z",
8733
+
"id": "W_Z",
8734
+
"weight": 0.5
8735
+
},
8736
+
{
8737
+
"a": "Y",
8738
+
"b": "AA",
8739
+
"id": "Y_AA",
8740
+
"weight": 0.5
8741
+
},
8742
+
{
8743
+
"a": "Y",
8744
+
"b": "AE",
8745
+
"id": "Y_AE",
8746
+
"weight": 0.5
8747
+
},
8748
+
{
8749
+
"a": "Y",
8750
+
"b": "AH",
8751
+
"id": "Y_AH",
8752
+
"weight": 0.5
8753
+
},
8754
+
{
8755
+
"a": "Y",
8756
+
"b": "AO",
8757
+
"id": "Y_AO",
8758
+
"weight": 0.5
8759
+
},
8760
+
{
8761
+
"a": "Y",
8762
+
"b": "AW",
8763
+
"id": "Y_AW",
8764
+
"weight": 0.5
8765
+
},
8766
+
{
8767
+
"a": "Y",
8768
+
"b": "AY",
8769
+
"id": "Y_AY",
8770
+
"weight": 0.5
8771
+
},
8772
+
{
8773
+
"a": "Y",
8774
+
"b": "B",
8775
+
"id": "Y_B",
8776
+
"weight": 0.5
8777
+
},
8778
+
{
8779
+
"a": "Y",
8780
+
"b": "CH",
8781
+
"id": "Y_CH",
8782
+
"weight": 0.5
8783
+
},
8784
+
{
8785
+
"a": "Y",
8786
+
"b": "D",
8787
+
"id": "Y_D",
8788
+
"weight": 0.5
8789
+
},
8790
+
{
8791
+
"a": "Y",
8792
+
"b": "DH",
8793
+
"id": "Y_DH",
8794
+
"weight": 0.5
8795
+
},
8796
+
{
8797
+
"a": "Y",
8798
+
"b": "EH",
8799
+
"id": "Y_EH",
8800
+
"weight": 0.5
8801
+
},
8802
+
{
8803
+
"a": "Y",
8804
+
"b": "ER",
8805
+
"id": "Y_ER",
8806
+
"weight": 0.5
8807
+
},
8808
+
{
8809
+
"a": "Y",
8810
+
"b": "EY",
8811
+
"id": "Y_EY",
8812
+
"weight": 0.5
8813
+
},
8814
+
{
8815
+
"a": "Y",
8816
+
"b": "F",
8817
+
"id": "Y_F",
8818
+
"weight": 0.5
8819
+
},
8820
+
{
8821
+
"a": "Y",
8822
+
"b": "G",
8823
+
"id": "Y_G",
8824
+
"weight": 0.5
8825
+
},
8826
+
{
8827
+
"a": "Y",
8828
+
"b": "IH",
8829
+
"id": "Y_IH",
8830
+
"weight": 0.5
8831
+
},
8832
+
{
8833
+
"a": "Y",
8834
+
"b": "IY",
8835
+
"id": "Y_IY",
8836
+
"weight": 0.5
8837
+
},
8838
+
{
8839
+
"a": "Y",
8840
+
"b": "JH",
8841
+
"id": "Y_JH",
8842
+
"weight": 0.5
8843
+
},
8844
+
{
8845
+
"a": "Y",
8846
+
"b": "K",
8847
+
"id": "Y_K",
8848
+
"weight": 0.5
8849
+
},
8850
+
{
8851
+
"a": "Y",
8852
+
"b": "L",
8853
+
"id": "Y_L",
8854
+
"weight": 0.5
8855
+
},
8856
+
{
8857
+
"a": "Y",
8858
+
"b": "M",
8859
+
"id": "Y_M",
8860
+
"weight": 0.5
8861
+
},
8862
+
{
8863
+
"a": "Y",
8864
+
"b": "N",
8865
+
"id": "Y_N",
8866
+
"weight": 0.5
8867
+
},
8868
+
{
8869
+
"a": "Y",
8870
+
"b": "NG",
8871
+
"id": "Y_NG",
8872
+
"weight": 0.5
8873
+
},
8874
+
{
8875
+
"a": "Y",
8876
+
"b": "OW",
8877
+
"id": "Y_OW",
8878
+
"weight": 0.5
8879
+
},
8880
+
{
8881
+
"a": "Y",
8882
+
"b": "OY",
8883
+
"id": "Y_OY",
8884
+
"weight": 0.5
8885
+
},
8886
+
{
8887
+
"a": "Y",
8888
+
"b": "P",
8889
+
"id": "Y_P",
8890
+
"weight": 0.5
8891
+
},
8892
+
{
8893
+
"a": "Y",
8894
+
"b": "R",
8895
+
"id": "Y_R",
8896
+
"weight": 0.5
8897
+
},
8898
+
{
8899
+
"a": "Y",
8900
+
"b": "S",
8901
+
"id": "Y_S",
8902
+
"weight": 0.5
8903
+
},
8904
+
{
8905
+
"a": "Y",
8906
+
"b": "SH",
8907
+
"id": "Y_SH",
8908
+
"weight": 0.5
8909
+
},
8910
+
{
8911
+
"a": "Y",
8912
+
"b": "T",
8913
+
"id": "Y_T",
8914
+
"weight": 0.5
8915
+
},
8916
+
{
8917
+
"a": "Y",
8918
+
"b": "TH",
8919
+
"id": "Y_TH",
8920
+
"weight": 0.5
8921
+
},
8922
+
{
8923
+
"a": "Y",
8924
+
"b": "UH",
8925
+
"id": "Y_UH",
8926
+
"weight": 0.5
8927
+
},
8928
+
{
8929
+
"a": "Y",
8930
+
"b": "UW",
8931
+
"id": "Y_UW",
8932
+
"weight": 0.5
8933
+
},
8934
+
{
8935
+
"a": "Y",
8936
+
"b": "V",
8937
+
"id": "Y_V",
8938
+
"weight": 0.5
8939
+
},
8940
+
{
8941
+
"a": "Y",
8942
+
"b": "W",
8943
+
"id": "Y_W",
8944
+
"weight": 0.5
8945
+
},
8946
+
{
8947
+
"a": "Y",
8948
+
"b": "Y",
8949
+
"id": "Y_Y",
8950
+
"weight": 0.5
8951
+
},
8952
+
{
8953
+
"a": "Y",
8954
+
"b": "Z",
8955
+
"id": "Y_Z",
8956
+
"weight": 0.5
8957
+
},
8958
+
{
8959
+
"a": "Z",
8960
+
"b": "B",
8961
+
"id": "Z_B",
8962
+
"weight": 0.5
8963
+
},
8964
+
{
8965
+
"a": "Z",
8966
+
"b": "CH",
8967
+
"id": "Z_CH",
8968
+
"weight": 0.5
8969
+
},
8970
+
{
8971
+
"a": "Z",
8972
+
"b": "D",
8973
+
"id": "Z_D",
8974
+
"weight": 0.5
8975
+
},
8976
+
{
8977
+
"a": "Z",
8978
+
"b": "DH",
8979
+
"id": "Z_DH",
8980
+
"weight": 0.5
8981
+
},
8982
+
{
8983
+
"a": "Z",
8984
+
"b": "F",
8985
+
"id": "Z_F",
8986
+
"weight": 0.5
8987
+
},
8988
+
{
8989
+
"a": "Z",
8990
+
"b": "G",
8991
+
"id": "Z_G",
8992
+
"weight": 0.5
8993
+
},
8994
+
{
8995
+
"a": "Z",
8996
+
"b": "JH",
8997
+
"id": "Z_JH",
8998
+
"weight": 0.5
8999
+
},
9000
+
{
9001
+
"a": "Z",
9002
+
"b": "K",
9003
+
"id": "Z_K",
9004
+
"weight": 0.5
9005
+
},
9006
+
{
9007
+
"a": "Z",
9008
+
"b": "L",
9009
+
"id": "Z_L",
9010
+
"weight": 0.5
9011
+
},
9012
+
{
9013
+
"a": "Z",
9014
+
"b": "M",
9015
+
"id": "Z_M",
9016
+
"weight": 0.5
9017
+
},
9018
+
{
9019
+
"a": "Z",
9020
+
"b": "N",
9021
+
"id": "Z_N",
9022
+
"weight": 0.5
9023
+
},
9024
+
{
9025
+
"a": "Z",
9026
+
"b": "NG",
9027
+
"id": "Z_NG",
9028
+
"weight": 0.5
9029
+
},
9030
+
{
9031
+
"a": "Z",
9032
+
"b": "P",
9033
+
"id": "Z_P",
9034
+
"weight": 0.5
9035
+
},
9036
+
{
9037
+
"a": "Z",
9038
+
"b": "R",
9039
+
"id": "Z_R",
9040
+
"weight": 0.5
9041
+
},
9042
+
{
9043
+
"a": "Z",
9044
+
"b": "S",
9045
+
"id": "Z_S",
9046
+
"weight": 0.5
9047
+
},
9048
+
{
9049
+
"a": "Z",
9050
+
"b": "SH",
9051
+
"id": "Z_SH",
9052
+
"weight": 0.5
9053
+
},
9054
+
{
9055
+
"a": "Z",
9056
+
"b": "T",
9057
+
"id": "Z_T",
9058
+
"weight": 0.5
9059
+
},
9060
+
{
9061
+
"a": "Z",
9062
+
"b": "TH",
9063
+
"id": "Z_TH",
9064
+
"weight": 0.5
9065
+
},
9066
+
{
9067
+
"a": "Z",
9068
+
"b": "V",
9069
+
"id": "Z_V",
9070
+
"weight": 0.5
9071
+
},
9072
+
{
9073
+
"a": "Z",
9074
+
"b": "W",
9075
+
"id": "Z_W",
9076
+
"weight": 0.5
9077
+
},
9078
+
{
9079
+
"a": "Z",
9080
+
"b": "Y",
9081
+
"id": "Z_Y",
9082
+
"weight": 0.5
9083
+
},
9084
+
{
9085
+
"a": "Z",
9086
+
"b": "Z",
9087
+
"id": "Z_Z",
9088
+
"weight": 0.5
9089
+
},
9090
+
{
9091
+
"a": "ZH",
9092
+
"b": "B",
9093
+
"id": "ZH_B",
9094
+
"weight": 0.5
9095
+
},
9096
+
{
9097
+
"a": "ZH",
9098
+
"b": "CH",
9099
+
"id": "ZH_CH",
9100
+
"weight": 0.5
9101
+
},
9102
+
{
9103
+
"a": "ZH",
9104
+
"b": "D",
9105
+
"id": "ZH_D",
9106
+
"weight": 0.5
9107
+
},
9108
+
{
9109
+
"a": "ZH",
9110
+
"b": "DH",
9111
+
"id": "ZH_DH",
9112
+
"weight": 0.5
9113
+
},
9114
+
{
9115
+
"a": "ZH",
9116
+
"b": "F",
9117
+
"id": "ZH_F",
9118
+
"weight": 0.5
9119
+
},
9120
+
{
9121
+
"a": "ZH",
9122
+
"b": "G",
9123
+
"id": "ZH_G",
9124
+
"weight": 0.5
9125
+
},
9126
+
{
9127
+
"a": "ZH",
9128
+
"b": "JH",
9129
+
"id": "ZH_JH",
9130
+
"weight": 0.5
9131
+
},
9132
+
{
9133
+
"a": "ZH",
9134
+
"b": "K",
9135
+
"id": "ZH_K",
9136
+
"weight": 0.5
9137
+
},
9138
+
{
9139
+
"a": "ZH",
9140
+
"b": "L",
9141
+
"id": "ZH_L",
9142
+
"weight": 0.5
9143
+
},
9144
+
{
9145
+
"a": "ZH",
9146
+
"b": "M",
9147
+
"id": "ZH_M",
9148
+
"weight": 0.5
9149
+
},
9150
+
{
9151
+
"a": "ZH",
9152
+
"b": "N",
9153
+
"id": "ZH_N",
9154
+
"weight": 0.5
9155
+
},
9156
+
{
9157
+
"a": "ZH",
9158
+
"b": "NG",
9159
+
"id": "ZH_NG",
9160
+
"weight": 0.5
9161
+
},
9162
+
{
9163
+
"a": "ZH",
9164
+
"b": "P",
9165
+
"id": "ZH_P",
9166
+
"weight": 0.5
9167
+
},
9168
+
{
9169
+
"a": "ZH",
9170
+
"b": "R",
9171
+
"id": "ZH_R",
9172
+
"weight": 0.5
9173
+
},
9174
+
{
9175
+
"a": "ZH",
9176
+
"b": "S",
9177
+
"id": "ZH_S",
9178
+
"weight": 0.5
9179
+
},
9180
+
{
9181
+
"a": "ZH",
9182
+
"b": "SH",
9183
+
"id": "ZH_SH",
9184
+
"weight": 0.5
9185
+
},
9186
+
{
9187
+
"a": "ZH",
9188
+
"b": "T",
9189
+
"id": "ZH_T",
9190
+
"weight": 0.5
9191
+
},
9192
+
{
9193
+
"a": "ZH",
9194
+
"b": "TH",
9195
+
"id": "ZH_TH",
9196
+
"weight": 0.5
9197
+
},
9198
+
{
9199
+
"a": "ZH",
9200
+
"b": "V",
9201
+
"id": "ZH_V",
9202
+
"weight": 0.5
9203
+
},
9204
+
{
9205
+
"a": "ZH",
9206
+
"b": "W",
9207
+
"id": "ZH_W",
9208
+
"weight": 0.5
9209
+
},
9210
+
{
9211
+
"a": "ZH",
9212
+
"b": "Y",
9213
+
"id": "ZH_Y",
9214
+
"weight": 0.5
9215
+
},
9216
+
{
9217
+
"a": "ZH",
9218
+
"b": "Z",
9219
+
"id": "ZH_Z",
9220
+
"weight": 0.5
9221
+
}
9222
+
]
9223
+
}
+110
pop/voice/jeffrey-anthropometry.json
+110
pop/voice/jeffrey-anthropometry.json
···
1
+
{
2
+
"version": 1,
3
+
"generated": "2026-05-04",
4
+
"generator": "pop/voice/bin/measure-jeffrey.py v0.2",
5
+
"source_glob": "/Users/jas/aesthetic-computer/portraits/jeffrey/corpus/shoot/*.jpg",
6
+
"source_corpus": {
7
+
"whistlegraph": 521
8
+
},
9
+
"rejected": {
10
+
"no_face": 158
11
+
},
12
+
"measurements_cm": {
13
+
"forehead_to_chin_cm": {
14
+
"median": 15.82,
15
+
"iqr": [
16
+
15.18,
17
+
16.78
18
+
],
19
+
"n": 363
20
+
},
21
+
"bizygomatic_width_cm": {
22
+
"median": 12.31,
23
+
"iqr": [
24
+
11.94,
25
+
12.8
26
+
],
27
+
"n": 363
28
+
},
29
+
"mandible_length_cm": {
30
+
"median": 6.38,
31
+
"iqr": [
32
+
5.77,
33
+
7.43
34
+
],
35
+
"n": 363
36
+
},
37
+
"lip_width_cm": {
38
+
"median": 4.99,
39
+
"iqr": [
40
+
4.67,
41
+
5.41
42
+
],
43
+
"n": 363
44
+
},
45
+
"lip_aperture_cm": {
46
+
"median": 2.02,
47
+
"iqr": [
48
+
1.61,
49
+
2.68
50
+
],
51
+
"n": 363
52
+
},
53
+
"philtrum_cm": {
54
+
"median": 2.06,
55
+
"iqr": [
56
+
1.82,
57
+
2.25
58
+
],
59
+
"n": 363
60
+
},
61
+
"nose_to_chin_cm": {
62
+
"median": 6.72,
63
+
"iqr": [
64
+
6.23,
65
+
7.32
66
+
],
67
+
"n": 363
68
+
},
69
+
"nostril_width_cm": {
70
+
"median": 2.95,
71
+
"iqr": [
72
+
2.77,
73
+
3.16
74
+
],
75
+
"n": 363
76
+
}
77
+
},
78
+
"derived": {
79
+
"vtl_cm": {
80
+
"central": 17.5,
81
+
"ci95": [
82
+
15.93,
83
+
19.07
84
+
],
85
+
"model": "fitch-giedd-1999",
86
+
"input": "population-median-male"
87
+
},
88
+
"tract_segments_44": {
89
+
"segment_length_cm": 0.3977
90
+
},
91
+
"lip_aperture_max_cm": 2.68,
92
+
"lip_mul_upper_bound": 1.0,
93
+
"mandible_rotation_max_deg": 22,
94
+
"notes": [
95
+
"VTL central uses Fitch+Giedd 1999 body-height regression.",
96
+
"Photo-derived measurements (forehead-to-chin, bizygomatic, mandible) are not yet wired into a closed-form VTL \u2014 Lammert-Narayanan 2015 needs MRI-derived inputs we don't have.",
97
+
"Lip aperture upper bound comes from corpus 75th percentile (max would be too sensitive to selfie outliers)."
98
+
]
99
+
},
100
+
"scale_anchor": {
101
+
"method": "interpupillary distance",
102
+
"ipd_cm_used": 6.3,
103
+
"ipd_px_median": 120.8,
104
+
"rationale": "Adult-population mean (Dodgson 2004). Replace with stated height + Lammert 2015 if/when --height-cm is provided."
105
+
},
106
+
"notes": [
107
+
"Single-pass mediapipe FaceMesh on local AV-shoot cache.",
108
+
"Identity validation skipped (AV shoot is jeffrey by construction). Wire --identity-jsonl when running against IG archive."
109
+
]
110
+
}