Update app.py

app.py

@@ -3,55 +3,51 @@ import gradio as gr
 import librosa
 import torch
 import spaces
+import numpy as np
 
 @spaces.GPU(duration=120)
 def transcribe_and_respond(audio_file):
     try:
-        # Load the model pipeline
         pipe = transformers.pipeline(
             model='sarvamai/shuka_v1',
             trust_remote_code=True,
             device=0,
             torch_dtype=torch.bfloat16
         )
-        
+
         # Load the audio file
         audio, sr = librosa.load(audio_file, sr=16000)
-        
-        # Print the path of the audio file
-        print(f"Audio file path: {audio_file}")
-        
-        # Prepare turns with a placeholder for the audio
+
+        # 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|>'}
         ]
-        
-        # Print the constructed prompt
-        print(f"Constructed prompt: {turns}")
-        
-        # Run the pipeline with the audio and constructed prompt
+
+        # Debug: Print the initial turns
+        print(f"Initial turns: {turns}")
+
+        # Call the model with the audio and prompt
         output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
-        
-        # Print the output from the model
+
+        # Debug: Print the final output from the model
         print(f"Model output: {output}")
-        
-        # Return the output for the Gradio interface
+
         return output
 
     except Exception as e:
         return f"Error: {str(e)}"
 
-# Set up the Gradio interface
 iface = gr.Interface(
     fn=transcribe_and_respond,
-    inputs=gr.Audio(sources="microphone", type="filepath"),  # Accept audio input from microphone
-    outputs="text",  # Output as text
+    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  # Enable live processing
+    live=True
 )
 
-# Launch the interface
 if __name__ == "__main__":
     iface.launch()