Update app.py

app.py

@@ -1,72 +1,41 @@
 import transformers
 import gradio as gr
+import librosa
 import torch
-import numpy as np
-from typing import Dict, List, Tuple
 import spaces
-import librosa
-import soundfile as sf
 
-MODEL_NAME = 'sarvamai/shuka_v1'
-SAMPLE_RATE = 16000
-MAX_NEW_TOKENS = 256
+@spaces.GPU(duration=120)
+def transcribe_and_respond(audio_file):
+    try:
+        pipe = transformers.pipeline(
+     model='sarvamai/shuka_v1',
+     trust_remote_code=True,
+     device=0,
+     torch_dtype=torch.bfloat16
+     )
 
-def load_pipeline():
-    return transformers.pipeline(
-        model=MODEL_NAME,
-        trust_remote_code=True,
-        device=0,
-        torch_dtype=torch.bfloat16
-    )
+        audio, sr = librosa.load(audio_file, sr=16000)
 
-pipe = load_pipeline()
+        turns = [
+            {'role': 'system', 'content': 'Respond naturally and informatively.'},
+            {'role': 'user', 'content': ''}
+        ]
 
-def create_conversation_turns(prompt: str) -> List[Dict[str, str]]:
-    return [
-        {'role': 'system', 'content': 'Respond naturally and informatively.'},
-        {'role': 'user', 'content': prompt}
-    ]
+        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
+
+        return output
 
-@spaces.GPU(duration=120)
-def transcribe_and_respond(audio_input: Tuple[int, np.ndarray]) -> str:
-    try:
-        # Unpack the audio input
-        sample_rate, audio = audio_input
-        
-        # Ensure audio is float32
-        if audio.dtype != np.float32:
-            audio = audio.astype(np.float32)
-        
-        if sample_rate != SAMPLE_RATE:
-            audio = librosa.resample(audio, orig_sr=sample_rate, target_sr=SAMPLE_RATE)
-        
-        # Convert the audio to WAV format
-        wav_data = librosa.util.buf_to_float(audio, n_bytes=2)
-        sf.write('temp_audio.wav', wav_data, SAMPLE_RATE)
-        
-        # Prepare the inputs for the model
-        turns = create_conversation_turns("")
-        inputs = {
-            'audio': wav_data,
-            'turns': turns,
-            'sampling_rate': SAMPLE_RATE
-        }
-        
-        response = pipe(inputs, max_new_tokens=MAX_NEW_TOKENS)
-        
-        return response
     except Exception as e:
-        return f"Error processing audio: {str(e)}"
+        return f"Error: {str(e)}"
 
 iface = gr.Interface(
     fn=transcribe_and_respond,
-    inputs=gr.Audio(sources="microphone", type="numpy"),
-    outputs="text",
-    title="Live Voice Input for Transcription and Response",
+    inputs=gr.Audio(source="microphone", type="filepath"),  # Accept audio input from microphone
+    outputs="text",  # Output as text
+    title="Live Transcription and Response",
     description="Speak into your microphone, and the model will respond naturally and informatively.",
-    live=True
+    live=True  # Enable live processing
 )
 
-# Launch the app
 if __name__ == "__main__":
     iface.launch()