update

app/core/asr_engine.py

@@ -138,25 +138,40 @@ def transcribe_long_audio(
                 combined_text_parts.append(part_text)
 
             raw_segs = out.get("chunks") or out.get("segments") or []
-            for s in raw_segs:
-                start = None
-                end = None
-                if isinstance(s.get("timestamp"), (list, tuple)) and len(s.get("timestamp")) >= 2:
-                    ts = s.get("timestamp")
-                    start, end = ts[0], ts[1]
-                elif s.get("start") is not None and s.get("end") is not None:
-                    start, end = s.get("start"), s.get("end")
-
-                text = (s.get("text") or "").strip()
-                if not text or start is None or end is None:
-                    continue
-
-                try:
-                    combined_chunks.append(
-                        {"start": float(start) + base_offset, "end": float(end) + base_offset, "text": text}
-                    )
-                except Exception:
-                    continue
+            if raw_segs:
+                for s in raw_segs:
+                    start = None
+                    end = None
+                    if isinstance(s.get("timestamp"), (list, tuple)) and len(s.get("timestamp")) >= 2:
+                        ts = s.get("timestamp")
+                        start, end = ts[0], ts[1]
+                    elif s.get("start") is not None and s.get("end") is not None:
+                        start, end = s.get("start"), s.get("end")
+
+                    text = (s.get("text") or "").strip()
+                    if not text or start is None or end is None:
+                        continue
+
+                    try:
+                        combined_chunks.append(
+                            {"start": float(start) + base_offset, "end": float(end) + base_offset, "text": text}
+                        )
+                    except Exception:
+                        continue
+            else:
+                # If model returned text but no timestamped segments for this chunk,
+                # create a fallback chunk spanning the chunk file duration.
+                if part_text:
+                    try:
+                        cinfo = get_audio_info(cp) or {}
+                        cdur = cinfo.get("duration") or chunk_length_s
+                        combined_chunks.append({
+                            "start": float(base_offset),
+                            "end": float(base_offset) + float(cdur),
+                            "text": part_text,
+                        })
+                    except Exception:
+                        logger.exception("failed to create fallback chunk for %s", cp)
 
     finally:
         for p in chunk_paths: