hermann/expria-frontend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

Sprint 6b — Frontend audio capture + playback hooks

Browse source 
feat(audio): pcm-capture-processor.js AudioWorklet (16kHz resample, Int16 LE)
feat(hooks): useAudioCapture (getUserMedia + worklet + onChunk base64)
feat(hooks): useAudioPlayback (24kHz sequential scheduling, gap-free)
feat(hooks): useAudioRecording (chronological buffer, resample 16→24k, WAV export)
feat(lib): audio-utils (base64, int16/float32, resample, WAV header)
test: 12 audio-utils + 7 useAudioRecording = 238/238 green (+19)
This commit is contained in:
Hermann_Kitio 2026-04-26 20:08:45 +03:00
parent 5a31819bca
commit 7862f7c9f3
8 changed files with 862 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

20
docs/CHANGELOG.md
View file

@ -29,6 +29,26 @@ Chaque entrée suit ce format :
---
## [Unreleased] — 2026-04-26 — Sprint 6b — Frontend audio (T2 Live)
### Added
- `public/pcm-capture-processor.js` — AudioWorklet processor : capture micro, rééchantillonnage vers 16 kHz si `sampleRate` natif différent, conversion Float32 → Int16 LE, chunks de 4096 samples (~256 ms).
- `src/shared/lib/audio-utils.ts` — 6 helpers purs : `arrayBufferToBase64`, `base64ToArrayBuffer`, `int16ToFloat32`, `float32ToInt16`, `resample16kTo24k`, `buildWavHeader`.
- `src/features/t2-live/hooks/useAudioCapture.ts` — hook capture : `getUserMedia` (mono, echoCancellation, noiseSuppression) → AudioContext 16 kHz → AudioWorklet → callback `onChunk(base64)`. Cleanup au stop/unmount.
- `src/features/t2-live/hooks/useAudioPlayback.ts` — hook playback : AudioContext 24 kHz lazy-init, scheduling séquentiel via `start(max(currentTime, lastEndTime))` pour lecture sans gaps. Cleanup au stop/unmount.
- `src/features/t2-live/hooks/useAudioRecording.ts` — hook recording : buffer chronologique unique normalisé 24 kHz (chunks candidat rééchantillonnés 16→24k), `addAIChunk(base64)` décode en interne, `exportWAV()` → Blob `audio/wav` mono 24 kHz.
- 12 tests `audio-utils.test.ts` (round-trips base64/ArrayBuffer, clamping int16/float32, interpolation resample, header WAV).
- 7 tests `useAudioRecording.test.ts` (add candidat resample, add IA, alternance, header WAV, reset, export vide, chunks vides ignorés).
### Notes
- Tests frontend : 219 → 238 verts (+19).
- `useAudioCapture` et `useAudioPlayback` dépendent de AudioContext (API navigateur) — validation manuelle au Sprint 6c.
- AudioWorklet utilisé directement (pas ScriptProcessorNode) — FTD-06 ne s'applique plus pour T2 Live.
---
## [Unreleased] — 2026-04-26 — Sprint 5.5 Clean FTD
### Changed