auto-save 2026-05-14 10:59 (~6)

api/main.py

@@ -97,7 +97,8 @@ def llm() -> OpenAI:
     return _llm_client
 
 # Pipeline 状态：
-#   created → downloading → downloaded（停，等用户点解析）→ splitting → frames_extracted
+#   created → downloading → downloaded（停，等用户点解析/提取音频）
+#         → splitting → frames_extracted
 #         → transcribing → transcribed | failed
 JobStatus = Literal[
     "created", "downloading", "downloaded",
@@ -1563,7 +1564,17 @@ async def pipeline_transcribe(job_id: str) -> None:
     wav = d / "audio.wav"
     try:
         if not wav.exists():
-            raise RuntimeError("audio.wav 不存在")
+            mp4 = d / "source.mp4"
+            if not mp4.exists():
+                raise RuntimeError("source.mp4 不存在，视频导入完成后再提取音频")
+            update(job, status="transcribing", message="ffmpeg 提取音频轨…", progress=max(job.progress, 45), error="")
+            run([
+                "ffmpeg", "-y", "-i", str(mp4),
+                "-vn", "-ac", "1", "-ar", "16000", "-c:a", "pcm_s16le",
+                str(wav),
+            ])
+            if not wav.exists():
+                raise RuntimeError("音频提取完成但找不到 audio.wav")
 
         if not LLM_API_KEY:
             # 无 key 模式：mock 数据
@@ -2004,9 +2015,12 @@ async def trigger_transcribe(job_id: str, bg: BackgroundTasks) -> Job:
     job = JOBS.get(job_id)
     if not job:
         raise HTTPException(404, "job not found")
-    if job.status not in {"frames_extracted", "transcribed", "failed"}:
-        raise HTTPException(409, f"status must be frames_extracted/transcribed/failed, got {job.status}")
-    update(job, status="transcribing", progress=max(job.progress, 72), error="", message="准备音频转写…")
+    mp4 = job_dir(job_id) / "source.mp4"
+    if job.status in {"created", "downloading"} or not mp4.exists():
+        raise HTTPException(409, f"video not ready, got {job.status}")
+    if job.status in {"splitting", "transcribing"} or job.audio_script.status == "rewriting":
+        raise HTTPException(409, f"job is busy, got {job.status}")
+    update(job, status="transcribing", progress=max(job.progress, 45), error="", message="准备提取音频…")
     bg.add_task(pipeline_transcribe, job_id)
     return job