Update app.py

app.py

@@ -151,8 +151,11 @@ def generate_speaker_clips(annotations, waveform, sample_rate, file_index):
 
         clip_waveform = waveform[:, start_sample:end_sample]
 
+        # soundfile has no FFmpeg/torchcodec dependency
+        import soundfile as sf
+        clip_np = clip_waveform.numpy().T  # (channels, samples) -> (samples, channels)
         buffer = io.BytesIO()
-        torchaudio.save(buffer, clip_waveform, sample_rate, format="wav")
+        sf.write(buffer, clip_np, sample_rate, format="WAV", subtype="PCM_16")
         buffer.seek(0)
         clips[speaker] = buffer.read()
 
@@ -676,7 +679,6 @@ try:
 
         current_renames = st.session_state.speakerRenames[currFileIndex]
         for sp in speakerNames:
-            # Read the live widget value so label updates immediately on rename
             widget_key = f"rename_{currFileIndex}_{sp}"
             live_name = st.session_state.get(widget_key, "").strip()
             display_label = live_name if live_name else current_renames.get(sp, sp)