Update app.py

app.py

@@ -175,14 +175,14 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic, voice_clea
             gpt_cond_latent,
             speaker_embedding,)            
         try:
-        
             for i, chunk in enumerate(chunks):
+                wav_chunks.append(chunk)
                 print(f"Received chunk {i} of audio length {chunk.shape[-1]}")
                 out_file = f'{i}.wav'
                 write(out_file, 24000, chunk.detach().cpu().numpy().squeeze())
                 audio = AudioSegment.from_file(out_file)
                 audio.export(out_file, format='wav')
-                yield (gr.make_waveform(audio=out_file),out_file, speaker_wav)
+                yield (None, out_file, None)
         except RuntimeError as e :
             if "device-side assert" in str(e):
                 # cannot do anything on cuda device side error, need tor estart
@@ -200,6 +200,17 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic, voice_clea
             else:
                 print("RuntimeError: non device-side assert error:", str(e))
                 raise e
+        
+        wav = torch.cat(wav_chunks, dim=0)
+        torchaudio.save("output.wav", wav.squeeze().unsqueeze(0).cpu(), 24000)
+        
+        return (
+            gr.make_waveform(
+                audio="output.wav",
+            ),
+            "output.wav",
+            speaker_wav,
+        )                
     else:
         gr.Warning("Please accept the Terms & Condition!")
         return (