Update app.py

app.py

@@ -25,10 +25,15 @@ speaker_embeddings = torch.randn(1, 512)
 
 def voice_assistant(audio):
 
-    # Speech → Text
-    speech_text = stt(audio)["text"]
+    if audio is None:
+        return "No audio detected", "Please record something first.", None
 
-    # Generate AI response
+    sample_rate, audio_data = audio
+
+    # Speech to text
+    speech_text = stt(audio_data)["text"]
+
+    # AI response
     response = llm(
         speech_text,
         max_new_tokens=60
@@ -42,8 +47,7 @@ def voice_assistant(audio):
         speaker_embeddings
     )
 
-    # Convert tensor → numpy audio
-    audio_output = speech.cpu().numpy().astype("float32")
+    audio_output = speech.cpu().numpy()
 
     return speech_text, response, (16000, audio_output)