christian/voice-agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(mac-worker): handle MLX segments as list/tuple in addition to dict

Browse Source 
lightning-whisper-mlx returns segments as [start, end, text] lists
(not dicts) in some versions, which broke transcribe(). Normalize all
three observed shapes: str output, list/tuple segments, dict segments.
Also fall back to joining segment texts when top-level text is empty.
This commit is contained in:
root 2026-05-13 16:00:30 +00:00
parent 27f2f91ec6
commit cb60172b51
1 changed files with 26 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
mac-worker/app/core/whisper_engine.py
View File

@ -53,19 +53,34 @@ class WhisperEngine:
if self.engine == "mlx":
out = self._impl.transcribe(audio_path=str(audio_path), language=lang)
# MLX-Output: {"text": "...", "segments": [...]} oder String. Robust mappen.
# MLX-Output schwankt zwischen Versionen:
# - String
# - {"text": "...", "segments": [...]}
# - {"text": "...", "segments": [[start, end, text], ...]}
# - {"text": "...", "segments": [{"start":..,"end":..,"text":..}, ...]}
if isinstance(out, str):
return {"text": out, "language": lang, "segments": []}
segments = out.get("segments") or []
normalized = [
{
"start": float(s.get("start", 0.0)),
"end": float(s.get("end", 0.0)),
"text": (s.get("text") or "").strip(),
}
for s in segments
]
return {"text": out.get("text", "").strip(), "language": lang, "segments": normalized}
raw_segments = out.get("segments") or []
normalized = []
for s in raw_segments:
if isinstance(s, dict):
normalized.append({
"start": float(s.get("start", 0.0)),
"end": float(s.get("end", 0.0)),
"text": (s.get("text") or "").strip(),
})
elif isinstance(s, (list, tuple)) and len(s) >= 3:
normalized.append({
"start": float(s[0] or 0.0),
"end": float(s[1] or 0.0),
"text": str(s[2] or "").strip(),
})
else:
normalized.append({"start": 0.0, "end": 0.0, "text": str(s).strip()})
text = (out.get("text") or "").strip()
if not text and normalized:
text = " ".join(seg["text"] for seg in normalized).strip()
return {"text": text, "language": lang, "segments": normalized}
if self.engine == "faster":
segments_iter, info = self._impl.transcribe(str(audio_path), language=lang, vad_filter=True)