change to cpu

app.py

@@ -176,7 +176,8 @@ with gr.Blocks(title="Voice chat with LLM") as demo:
                 generated_speech = generate_speech_for_sentence(chatbot_history, chatbot_voice, sentence, xtts_model, None, return_as_byte=True)
                 if generated_speech is not None:
                     _, audio_dict = generated_speech
-                    yield (sentence, chatbot_history, audio_dict["value"])
+                    yield (sentence, chatbot_history, audio_dict)
+                    #yield (sentence, chatbot_history, audio_dict["value"])
 
         if initial_greeting:
             # Process only the initial greeting if specified

utils.py

@@ -421,10 +421,13 @@ def generate_speech_for_sentence(history, chatbot_voice, sentence, xtts_model, x
                         f.setsampwidth(2)
                         f.setframerate(24000)
                         f.writeframes(wav_bytestream)
+
+                        return audio_unique_filename
                            
-                    return (history , gr.Audio(value=audio_unique_filename, autoplay=True))
+                    #return (history , gr.Audio.update(value=audio_unique_filename, autoplay=True))
                 else:
-                    return (history , gr.Audio(value=wav_bytestream, autoplay=True))
+                    return wav_bytestream
+                    #return (history , gr.Audio.update(value=wav_bytestream, autoplay=True))
     except RuntimeError as e:
         if "device-side assert" in str(e):
             # cannot do anything on cuda device side error, need tor estart