Add comprehensive error handling with traceback

app.py

@@ -240,56 +240,66 @@ def transcribe(audio_upload: Any, lang_disp: str, return_ts: bool, progress=gr.P
     """
     Main transcription function with ZeroGPU support.
     """
-    if audio_upload is None:
-        return "", "", None, "<div style='color:#666'>Please upload an audio file first.</div>"
-
-    # Debug: log the type and content of audio_upload
-    print(f"[DEBUG] audio_upload type: {type(audio_upload)}")
-    print(f"[DEBUG] audio_upload: {audio_upload}")
+    import traceback
     
     try:
-        audio_obj = _parse_audio_any(audio_upload)
-    except ValueError as e:
-        return "", "", None, f"<div style='color:red'>Error: {str(e)}<br>Debug: type={type(audio_upload)}, value={audio_upload}</div>"
-
-    language = None
-    if lang_disp and lang_disp != "Auto":
-        language = lang_map.get(lang_disp, lang_disp)
-
-    # Get model (lazy-loaded on first use when GPU is available)
-    asr = get_asr_model()
-    
-    # Perform transcription
-    results = asr.transcribe(
-        audio=audio_obj,
-        language=language,
-        return_time_stamps=return_ts,
-    )
-
-    if not isinstance(results, list) or len(results) != 1:
-        return "", "", None, "<div style='color:red'>Unexpected result format.</div>"
-
-    r = results[0]
-
-    # Extract timestamps
-    ts_payload = None
-    if return_ts and hasattr(r, "time_stamps") and r.time_stamps:
-        ts_payload = [
-            dict(
-                text=getattr(t, "text", ""),
-                start_time=getattr(t, "start_time", 0),
-                end_time=getattr(t, "end_time", 0),
-            )
-            for t in r.time_stamps
-        ]
-
-    # Note: Visualization is generated separately when user clicks "Visualize Timestamps"
-    return (
-        getattr(r, "language", "") or "",
-        getattr(r, "text", "") or "",
-        ts_payload,
-        "",  # Empty HTML - visualization is triggered by separate button
-    )
+        if audio_upload is None:
+            return "", "", None, "<div style='color:#666'>Please upload an audio file first.</div>"
+
+        # Debug: log the type and content of audio_upload
+        print(f"[DEBUG] audio_upload type: {type(audio_upload)}")
+        print(f"[DEBUG] audio_upload: {audio_upload}")
+        
+        try:
+            audio_obj = _parse_audio_any(audio_upload)
+        except ValueError as e:
+            return "", "", None, f"<div style='color:red'>Error: {str(e)}<br>Debug: type={type(audio_upload)}, value={audio_upload}</div>"
+
+        language = None
+        if lang_disp and lang_disp != "Auto":
+            language = lang_map.get(lang_disp, lang_disp)
+
+        # Get model (lazy-loaded on first use when GPU is available)
+        print("[DEBUG] Loading ASR model...")
+        asr = get_asr_model()
+        print("[DEBUG] Model loaded, transcribing...")
+        
+        # Perform transcription
+        results = asr.transcribe(
+            audio=audio_obj,
+            language=language,
+            return_time_stamps=return_ts,
+        )
+        print(f"[DEBUG] Transcription complete, results: {results}")
+
+        if not isinstance(results, list) or len(results) != 1:
+            return "", "", None, "<div style='color:red'>Unexpected result format.</div>"
+
+        r = results[0]
+
+        # Extract timestamps
+        ts_payload = None
+        if return_ts and hasattr(r, "time_stamps") and r.time_stamps:
+            ts_payload = [
+                dict(
+                    text=getattr(t, "text", ""),
+                    start_time=getattr(t, "start_time", 0),
+                    end_time=getattr(t, "end_time", 0),
+                )
+                for t in r.time_stamps
+            ]
+
+        # Note: Visualization is generated separately when user clicks "Visualize Timestamps"
+        return (
+            getattr(r, "language", "") or "",
+            getattr(r, "text", "") or "",
+            ts_payload,
+            "",  # Empty HTML - visualization is triggered by separate button
+        )
+    except Exception as e:
+        error_trace = traceback.format_exc()
+        print(f"[ERROR] Exception in transcribe: {error_trace}")
+        return "", "", None, f"<div style='color:red'>Error: {str(e)}<br><pre>{error_trace}</pre></div>"
 
 
 def visualize_timestamps(audio_upload: Any, timestamps_json: Any):