Update LLMwithvoice.py

LLMwithvoice.py

@@ -1,6 +1,7 @@
 import requests
 import torch
-import soundfile as sf
+import numpy as np
+import sounddevice as sd
 from transformers import AutoTokenizer
 from parler_tts import ParlerTTSForConditionalGeneration
 
@@ -29,9 +30,16 @@ def query_roberta(api_token, prompt, context):
     headers = {"Authorization": f"Bearer {api_token}"}
     response = requests.post(API_URL_ROBERTA, headers=headers, json=payload)
     try:
-        return response.json()
-    except ValueError:
-        return {"error": "Invalid JSON response"}
+        response_json = response.json()
+        if 'error' in response_json:
+            raise ValueError(response_json['error'])
+        return response_json
+    except ValueError as e:
+        print(f"ValueError: {e}")
+        return {"error": str(e)}
+    except Exception as e:
+        print(f"Exception: {e}")
+        return {"error": "An unexpected error occurred"}
 
 def generate_speech(answer):
     input_ids = tokenizer(answer, return_tensors="pt").input_ids.to(device)
@@ -39,19 +47,17 @@ def generate_speech(answer):
     generation = model.generate(input_ids=input_ids).to(torch.float32)
     audio_arr = generation.cpu().numpy().squeeze()
 
-    # Construct the audio path dynamically
-    audio_filename = "parler_tts_out.wav"
-    audio_path = f"/mnt/data/{audio_filename}"
-
+    # Play the generated audio
     try:
-        sf.write(audio_path, audio_arr, model.config.sampling_rate)
+        sd.play(audio_arr, samplerate=model.config.sampling_rate)
+        sd.wait()  # Wait until the audio is done playing
     except Exception as e:
-        print(f"Error writing audio file: {e}")
+        print(f"Error playing audio: {e}")
         # Handle the error, raise or log it, or provide an alternative approach
 
-    return audio_path
-
 def gradio_interface(api_token, prompt, context):
     answer = query_roberta(api_token, prompt, context)
-    audio_path = generate_speech(answer)
-    return answer, audio_path
+    if 'error' in answer:
+        return answer['error'], None
+    generate_speech(answer.get('answer', ''))
+    return answer.get('answer', 'No answer found'), None