update

app/jobs/transcribe_job.py

@@ -9,6 +9,7 @@ import time
 from app.core.asr_engine import load_model, transcribe_file, transcribe_file_chunks
 from app.services.note_client import NoteServiceClient
 from app.core.audio_utils import get_audio_info
+from app.core.audio_utils import ensure_wav_16k_mono, make_temp_path
 
 def run_async(coro):
     try:
@@ -41,21 +42,43 @@ def transcribe_job(audio_url: str, note_id: str, user_id: str | None = None):
         # 1️⃣ Download audio
         wav_path = download_audio(audio_url)
 
+        # Ensure WAV is 16k mono for consistent chunking and ASR behavior
+        try:
+            info = get_audio_info(wav_path) or {}
+            if info.get("samplerate") != 16000 or info.get("channels") != 1:
+                tmp_wav = make_temp_path(suffix=".wav")
+                ensure_wav_16k_mono(wav_path, tmp_wav)
+                # replace wav_path with converted file and remove original
+                try:
+                    os.remove(wav_path)
+                except Exception:
+                    pass
+                wav_path = tmp_wav
+        except Exception:
+            logging.exception("Failed to ensure wav format for %s", wav_path)
+
         # 2️⃣ ASR
         text = transcribe_file(model, wav_path, 30.0, 5.0)
         chunks = transcribe_file_chunks(model, wav_path, 30.0, 5.0)
 
+        # normalize chunks list
         chunks = [
-            {
-                "text": c["text"],
-                "start": c.get("start"),
-                "end": c.get("end"),
-            }
-            for c in chunks
+            {"text": c["text"], "start": c.get("start"), "end": c.get("end")}
+            for c in (chunks or [])
             if c.get("text", "").strip()
         ]
 
-        note_status = "transcribed" if chunks else "error"
+        # If no timestamped chunks but raw text exists, create per-file fallback chunk
+        if not chunks and text and text.strip():
+            try:
+                info = get_audio_info(wav_path) or {}
+                duration = info.get("duration") or 0.0
+                chunks = [{"text": text.strip(), "start": 0.0, "end": float(duration)}]
+            except Exception:
+                logging.exception("failed to create fallback chunk for note %s", note_id)
+
+        # Consider transcribed if we have either timestamped chunks or non-empty text
+        note_status = "transcribed" if (chunks or (text and text.strip())) else "error"
         info = get_audio_info(wav_path) or {}
 
         payload = {