GameCult/AquaSynth
Watch
9
0
Fork
You've already forked AquaSynth
 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

Tract stops: one place table, burst injected at the closure site #48

Merged
aqua merged 12 commits from aqua/tract-stops-20261112 into main 2026-11-13 07:14:35 +00:00
Conversation 0 Commits 12 Files changed 27 +1182 -63
aqua commented 2026-11-12 17:51:33 +00:00
Member
Copy link

Tract stops: one place table, burst injected at the closure site; /t/ /k/ /d/ /l/ added; listen kits and clip images. Specs: cut-stop-burst-site.r1, cut-tract-stop-t.r1 (r3 landed), cut-tract-stop-d.r1, cut-tract-lateral-l.r1, cut-stop-listen-kit.r1, cut-listen-page-images.r1. Soul: aqua-body:verdict:cut-listen-page-images.s1, pass at 813ae0b.

Place table. PatchScript holds one table (place -> cm behind the lip plane). /t/ /k/ /d/ /l/ have entries; deleting any fails a test. Burst injected at the closure site, area-scaled; /apa/ /aba/ stay byte identical (pinned by hash).

Not met. /l/ F2 measures 725 Hz against the 1000-1300 design target (F1, level and dip met). /t/ burst peak sits below /p/'s (4011 vs 7087 Hz). Everything here is unheard: no listener verdict yet.

Listen kits (tools/aqua-listen): stops kit and weak kit with spectrogram and waveform PNG per clip; burst times in a separate note.

Merged main (#49, #50) into the branch; stress/release, place-table, sample-rate and nasal tests pass (39).

🤖 Generated with Claude Code

Tract stops: one place table, burst injected at the closure site; /t/ /k/ /d/ /l/ added; listen kits and clip images. Specs: cut-stop-burst-site.r1, cut-tract-stop-t.r1 (r3 landed), cut-tract-stop-d.r1, cut-tract-lateral-l.r1, cut-stop-listen-kit.r1, cut-listen-page-images.r1. Soul: aqua-body:verdict:cut-listen-page-images.s1, pass at 813ae0b. **Place table.** `PatchScript` holds one table (place -> cm behind the lip plane). /t/ /k/ /d/ /l/ have entries; deleting any fails a test. Burst injected at the closure site, area-scaled; /apa/ /aba/ stay byte identical (pinned by hash). **Not met.** /l/ F2 measures 725 Hz against the 1000-1300 design target (F1, level and dip met). /t/ burst peak sits below /p/'s (4011 vs 7087 Hz). Everything here is unheard: no listener verdict yet. **Listen kits** (`tools/aqua-listen`): stops kit and weak kit with spectrogram and waveform PNG per clip; burst times in a separate note. Merged main (#49, #50) into the branch; stress/release, place-table, sample-rate and nasal tests pass (39). 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
The place table in PatchScript (cm behind the lip plane, or fraction only) replaces PlaceTarget and
ConsonantPositionFromGlottis. A contact whose position has a control surface injects its burst into the
ladder at that place, scaled by cell area; the surface exists only once a non-labial plosive has driven it,
so /p b/ words keep their script and Faust source byte for byte. weak_spots.burst_spectrum measures the
isolated burst.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A /t/ closure at the nasal's 0.03 cm aperture leaked (-14.3 dB re vowels against /p/ -22.1, the lips seal
/p/); a non-labial plosive now holds a zero aperture until its contact opens at 0.65. Its injected burst
reached 0.76 of the vowel peak at the shared 0.06 gain; TongueStopBurstGain 0.03. /p b/ untouched.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Measured on /ata/: closure -24.9 dB re vowels (/p/ -20.4), burst 0.39 of the vowel peak, burst peak inside
3.5-7 kHz. 'Above /p/' dropped from the criterion (steward ruling): the pinned labial burst enters at the
glottal end.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The place table held velar as fraction only (0.44); a distance puts it with /p t/ so the closure index and the burst site come from one entry. 9.81 cm
is the same 0.44 of the 17.52 cm tract, between the /u/ dorsal (6.97 cm) and /a/ pharyngeal (12.14 cm) narrowest sections.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Measured on /aka/: burst 0.45 of the vowel peak, burst spectral peak 3100 Hz (Chodroff and Wilson 2014: 2868 Hz; /t/ 3950 Hz by the same measure),
burst only at its own release. Closure -14.9 dB re vowels against /p/ -20.4: NOT at or under /p/. The tract seals (glottal source off: -22.7 dB); the leak
is the voicing the voiceless source still carries, and silencing it would change /p t/. Reported as unmet, with a regression bound in the test.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
/d/ is /t/ with voicing: no new mechanism, the 'voiced' token already selects the voice bar. Its consonant splines are identical to /t/'s (7 of 7).
Measured: voice bar -28.9 dB re the following vowel (/b/ -29.2, /t/ -38.7, /p/ -40.1), burst peak 3950 Hz as /t/, burst only at its own release.
The /d/-over-/t/ test measures the voice bar by deviation, as the /b/ and /p/ test does: the closure RMS window read 2.8 dB apart because of the slow offset of a shutting tract.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The gate passed: a branch port joins the oral path at the ridge and a closed 3 cm stub notches 2-3 kHz by 13.4 dB against the same render with the port shut
(the shut port is the render without the stub). The ladder cannot hold a second oral channel, so the midline is narrowed to 0.06 cm instead of closed.
The scaffold declares the stub only for a word with a lateral; /apa aba ana ata/ keep script and Faust source byte for byte.
Measured on /ala/: level -5.0 dB re the flanking vowels, F1 263 Hz, burst none. F2 725 Hz against the 1000-1300 design target: NOT met (reported).
A lateral gesture without the side stub is refused, not approximated.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Predictions and measured figures sit in the hidden titles only; the /l/ title carries the F2 miss (725 Hz against a 1000-1300 design target).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
clip_images.py (numpy + matplotlib Agg, STFT by numpy.fft) writes a deterministic spec and wave PNG per clip and, from the with/without --no-burst renders, burst-notes.md/json. Both kits' render.sh end with it; build.mjs --images embeds the pictures in a collapsed block and widens the page policy to data: images only then. The notes never enter listen.html or kit.json.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
aqua merged commit 92d4282546 into main 2026-11-13 07:14:35 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
GameCult/AquaSynth!48
No description provided.