Labyricorn/voicebox
1
0
Fork 0
mirror of https://github.com/jamiepine/voicebox.git synced 2026-09-15 04:40:40 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix(transcription): transcode uploads to WAV before STT (#957)

Browse Source 
The /transcribe endpoint passed the raw uploaded file straight to the STT
backend (mlx_audio.stt -> miniaudio), which only decodes WAV/FLAC/MP3/Vorbis.
Browser recordings arrive as WebM/Opus (Chrome/Firefox MediaRecorder), so
web-mode dictation failed with 500 "unsupported file format". The Tauri app
was unaffected because WebKit produces MP4.

librosa already fully decodes the upload to compute duration (falling back to
audioread/ffmpeg for exotic containers), so re-encode that PCM to a temp WAV
and hand it to Whisper. WAV inputs pass through unchanged; the temp file is
cleaned up in the finally block.

Co-authored-by: Claude Opus 4.8 <[email protected]>
This commit is contained in:
Ivan Corsetti
2026-07-26 23:31:27 -07:00
committed by GitHub
co-authored by Claude Opus 4.8
parent ed54347e81
commit a5773807a5
1 changed files with 16 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/routes/transcription.py
+16 -2
View File
@@ -35,13 +35,25 @@ async def transcribe_audio(
tmp.write(chunk)
tmp_path = tmp.name
stt_path = tmp_path
try:
from ..utils.audio import load_audio
from ..utils.audio import load_audio, save_audio
from ..backends import WHISPER_HF_REPOS
audio, sr = await asyncio.to_thread(load_audio, tmp_path)
duration = len(audio) / sr
# The STT backend (mlx_audio.stt -> miniaudio) only decodes
# WAV/FLAC/MP3/Vorbis, so browser recordings uploaded as WebM/Opus
# fail with "unsupported file format" (issue: web-mode dictation).
# librosa already decoded the file above (it falls back to
# audioread/ffmpeg for exotic containers), so re-encode that PCM to a
# temp WAV and hand *that* to Whisper. WAV inputs pass through
# unchanged.
if file_suffix != ".wav":
stt_path = f"{tmp_path}.stt.wav"
await asyncio.to_thread(save_audio, audio, stt_path, sr)
whisper_model = transcribe.get_whisper_model()
model_size = model if model else whisper_model.model_size
@@ -76,7 +88,7 @@ async def transcribe_audio(
},
)
text = await whisper_model.transcribe(tmp_path, language, model_size)
text = await whisper_model.transcribe(stt_path, language, model_size)
return models.TranscriptionResponse(
text=text,
@@ -89,3 +101,5 @@ async def transcribe_audio(
raise HTTPException(status_code=500, detail=str(e))
finally:
Path(tmp_path).unlink(missing_ok=True)
if stt_path != tmp_path:
Path(stt_path).unlink(missing_ok=True)