Update app.py

app.py

@@ -182,21 +182,7 @@ def model_inference(input_dict, history):
     audio_path = tts(buffer)
     return audio_path  # Return the audio file path
 
-# Alternative approach: Use regular Interface instead of ChatInterface
-def combined_inference(input_dict, history):
-    """Modified function that returns both text and audio"""
-    text_response = ""
-    
-    # Get the streaming response
-    for response in model_inference(input_dict, history):
-        text_response = response
-    
-    # Generate audio from final text
-    audio_path = tts(text_response)
-    
-    return text_response, audio_path
-
-# Option 1: Use regular Interface (recommended)
+# Option 1: Use regular Interface with streaming (recommended)
 with gr.Blocks() as demo:
     gr.Markdown("# oolel-vision-experimental `@video-infer for video understanding`")
     
@@ -210,18 +196,23 @@ with gr.Blocks() as demo:
     clear = gr.Button("Clear")
     
     def respond(message, chat_history):
-        # Get text response through streaming
-        text_response = ""
-        for response in model_inference(message, chat_history):
-            text_response = response
-        
-        # Add to chat history
-        chat_history.append([message["text"], text_response])
+        # Add user message to chat history
+        bot_message = ""
+        chat_history.append([message["text"], ""])
         
-        # Generate audio
-        audio_path = tts(text_response)
+        # Stream the response
+        for response in model_inference(message, chat_history):
+            bot_message = response
+            chat_history[-1][1] = bot_message
+            yield "", chat_history, None
         
-        return "", chat_history, audio_path
+        # Generate audio after streaming is complete
+        try:
+            audio_path = tts(bot_message)
+            yield "", chat_history, audio_path
+        except Exception as e:
+            print(f"TTS Error: {e}")
+            yield "", chat_history, None
     
     msg.submit(respond, [msg, chatbot], [msg, chatbot, audio_output])
     clear.click(lambda: ([], None), outputs=[chatbot, audio_output])