Update app.py

app.py

@@ -5,8 +5,11 @@ import torch
 import spaces
 import numpy as np
 
-@spaces.GPU(duration=60)
-def transcribe_and_respond(audio_file):
+# Maintain conversation history globally
+conversation_history = []
+
+@spaces.GPU(duration=15)
+def transcribe_and_respond(audio_file, chat_history):
     try:
         pipe = transformers.pipeline(
             model='sarvamai/shuka_v1',
@@ -21,32 +24,39 @@ def transcribe_and_respond(audio_file):
         # Print audio properties for debugging
         print(f"Audio dtype: {audio.dtype}, Audio shape: {audio.shape}, Sample rate: {sr}")
 
-        turns = [
-            {'role': 'system', 'content': 'Respond naturally and informatively.'},
-            {'role': 'user', 'content': '<|audio|>'}
-        ]
+        # Prepare conversation turns
+        turns = chat_history.copy()  # Take existing chat history
+
+        # Add the audio to the current turn
+        turns.append({'role': 'user', 'content': '<|audio|>'})
 
-        # Debug: Print the initial turns
-        print(f"Initial turns: {turns}")
+        # Debug: Print the updated turns
+        print(f"Updated turns: {turns}")
 
-        # Call the model with the audio and prompt
+        # Call the model with the updated conversation turns and audio
         output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
 
-        # Debug: Print the final output from the model
+        # Append the model's response to the chat history
+        turns.append({'role': 'system', 'content': output})
+
+        # Debug: Print the model's response
         print(f"Model output: {output}")
 
-        return output
+        return turns, turns  # Return updated history to display
 
     except Exception as e:
         return f"Error: {str(e)}"
 
+# Chat interface setup
 iface = gr.Interface(
     fn=transcribe_and_respond,
-    inputs=gr.Audio(sources="microphone", type="filepath"),
-    outputs="text",
-    title="Live Transcription and Response",
-    description="Speak into your microphone, and the model will respond naturally and informatively.",
-    live=True
+    inputs=[gr.Audio(sources="microphone", type="filepath", label="Your Audio (Microphone)"), gr.Chatbot(label="Conversation")],
+    outputs=gr.Chatbot(label="Conversation"),
+    title="🎙️ AI Chat with Live Transcription",
+    description="Talk to the AI through your microphone, and it will respond conversationally based on the ongoing chat. Keep the conversation going!",
+    live=True,
+    allow_flagging="auto",
+    enable_queue=True
 )
 
 if __name__ == "__main__":