auto-save 2026-05-17 13:56 (~4)

api/.env.example

@@ -12,6 +12,10 @@ WEB_AUTH_COOKIE_SECURE=false
 # 模型分工
 ASR_MODEL=whisper-1
 ASR_FALLBACK_MODEL=gemini-2.5-flash
+ASR_TIMEOUT_SECONDS=45
+LOCAL_ASR_BIN=/opt/homebrew/bin/mlx_whisper
+LOCAL_ASR_MODEL=mlx-community/whisper-tiny
+LOCAL_ASR_TIMEOUT_SECONDS=180
 TRANSLATE_MODEL=gemini-2.5-flash
 REWRITE_MODEL=gemini-2.5-pro
 IMAGE_MODEL=gemini-3-pro-image-preview

api/main.py

@@ -1793,6 +1793,8 @@ def _validate_asr_segments(segments: list[dict], duration: float, source: str) -
             raise TranscriptionUnavailable(f"{source} returned timestamps outside audio duration")
         if duration > 10 and last_end < duration * 0.45 and words < 20:
             raise TranscriptionUnavailable(f"{source} returned too little transcript coverage")
+    for item in clean:
+        item["_source"] = source
     return clean
 
 
@@ -2064,6 +2066,7 @@ def _audio_profile_model_sync(wav: Path, segments: list[TranscriptSegment], targ
                 response_format={"type": "json_object"},
                 temperature=0.1,
                 max_tokens=900,
+                timeout=ASR_TIMEOUT_SECONDS,
             )
             content = (resp.choices[0].message.content or "").strip()
             data = json.loads(content)
@@ -2322,6 +2325,7 @@ def pipeline_transcribe(job_id: str, manage_job_status: bool = True) -> None:
         segments = _transcribe_sync(wav)
         if not segments:
             raise TranscriptionUnavailable("ASR 未返回可用字幕段")
+        asr_source = str(segments[0].get("_source") or ASR_MODEL)
 
         # 先把英文段落落到 job 上（让 UI 提前看到，翻译再补 zh）
         en_only = [
@@ -2368,7 +2372,7 @@ def pipeline_transcribe(job_id: str, manage_job_status: bool = True) -> None:
         if manage_job_status:
             update(job, transcript=full, status="transcribed", progress=100,
                    audio_script=audio_script,
-                   message=f"音频解析完成 · {len(full)} 段（{ASR_MODEL} + {TRANSLATE_MODEL} + {ASR_FALLBACK_MODEL} 音频分析）")
+                   message=f"音频解析完成 · {len(full)} 段（{asr_source} + {TRANSLATE_MODEL} + {ASR_FALLBACK_MODEL} 音频分析）")
         else:
             update(job, transcript=full, audio_script=audio_script)