revert format=wav (broke preprocess); convert mic webm via ffmpeg in handler; show_error

app.py

@@ -20,10 +20,28 @@ def get_model():
     return MODEL
 
 
+def _ensure_wav(path):
+    """Mic recordings arrive as webm/mp4; libsndfile can't read those — convert."""
+    import soundfile as sf
+    try:
+        sf.info(path)
+        return path
+    except Exception:
+        import subprocess
+        import tempfile
+        out = tempfile.NamedTemporaryFile(suffix=".wav", delete=False).name
+        subprocess.run(
+            ["ffmpeg", "-y", "-loglevel", "error", "-i", path, "-ar", "16000", "-ac", "1", out],
+            check=True,
+        )
+        return out
+
+
 def transcribe(audio_path):
     if not audio_path:
         return "(no audio)"
     import soundfile as sf
+    audio_path = _ensure_wav(audio_path)
     info = sf.info(audio_path)
     if info.duration > 60:
         return "Please keep clips under 60 seconds for this CPU demo."
@@ -38,7 +56,7 @@ examples = [[f"examples/{f}"] for f in sorted(os.listdir("examples"))] \
 
 demo = gr.Interface(
     fn=transcribe,
-    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", format="wav",
+    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath",
                     label="Nepali speech (mic or file, ≤60 s)"),
     outputs=gr.Textbox(label="Transcript (Devanagari)", rtl=False),
     title="NepaliConformer — Nepali ASR for real telephone calls",
@@ -57,4 +75,4 @@ demo = gr.Interface(
 )
 
 if __name__ == "__main__":
-    demo.launch(ssr_mode=False)
+    demo.launch(ssr_mode=False, show_error=True)