trying things with cpu

app.py

@@ -53,22 +53,23 @@ mistral_llm = Llama(model_path=mistral_model_path,n_gpu_layers=35,max_new_tokens
 
 # Load XTTS Model
 print("Loading XTTS model")
-#os.environ["COQUI_TOS_AGREED"] = "1"
-#tts_model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
-#ModelManager().download_model(tts_model_name)
-#tts_model_path = os.path.join(get_user_data_dir("tts"), tts_model_name.replace("/", "--"))
-#config = XttsConfig()
-#config.load_json(os.path.join(tts_model_path, "config.json"))
-#xtts_model = Xtts.init_from_config(config)
-#xtts_model.load_checkpoint(
-#    config,
-#    checkpoint_path=os.path.join(tts_model_path, "model.pth"),
-#    vocab_path=os.path.join(tts_model_path, "vocab.json"),
-#    eval=True,
-#    use_deepspeed=True,
-#)
+os.environ["COQUI_TOS_AGREED"] = "1"
+tts_model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
+ModelManager().download_model(tts_model_name)
+tts_model_path = os.path.join(get_user_data_dir("tts"), tts_model_name.replace("/", "--"))
+config = XttsConfig()
+config.load_json(os.path.join(tts_model_path, "config.json"))
+xtts_model = Xtts.init_from_config(config)
+xtts_model.to("cpu")
+xtts_model.load_checkpoint(
+    config,
+    checkpoint_path=os.path.join(tts_model_path, "model.pth"),
+    vocab_path=os.path.join(tts_model_path, "vocab.json"),
+    eval=True,
+    use_deepspeed=True,
+)
 #xtts_model.cuda()
-print("UN-Loading XTTS model")
+#print("UN-Loading XTTS model")
 
 ###### Set up Gradio Interface ######
 
@@ -142,11 +143,10 @@ with gr.Blocks(title="Voice chat with LLM") as demo:
         def handle_speech_generation(sentence, chatbot_history, chatbot_voice):
             if sentence != "":
                 print("Processing sentence")
-                yield (sentence, chatbot_history, None)
-               # generated_speech = generate_speech_for_sentence(chatbot_history, chatbot_voice, sentence, xtts_model, xtts_supported_languages=config.languages, return_as_byte=True)
-               # if generated_speech is not None:
-               #     _, audio_dict = generated_speech
-               #     yield (sentence, chatbot_history, audio_dict["value"])
+                generated_speech = generate_speech_for_sentence(chatbot_history, chatbot_voice, sentence, xtts_model, xtts_supported_languages=config.languages, return_as_byte=True)
+                if generated_speech is not None:
+                    _, audio_dict = generated_speech
+                    yield (sentence, chatbot_history, audio_dict["value"])
 
         if initial_greeting:
             # Process only the initial greeting if specified