Update app.py

app.py

@@ -2,36 +2,48 @@ import transformers
 import gradio as gr
 import librosa
 import torch
-import spaces
+import numpy as np
+
 
 @spaces.GPU(duration=120)
-def transcribe_and_respond(audio_file):
+def transcribe_and_respond(audio_input: Tuple[np.ndarray, int]) -> str:
     try:
         pipe = transformers.pipeline(
-     model='sarvamai/shuka_v1',
-     trust_remote_code=True,
-     device=0,
-     torch_dtype=torch.bfloat16
-     )
-
-        audio, sr = librosa.load(audio_file, sr=16000)
-
+    model='sarvamai/shuka_v1',
+    trust_remote_code=True,
+    device=0,
+    torch_dtype=torch.bfloat16
+)
+        # Unpack the audio input
+        audio, sr = audio_input
+        
+        # Ensure audio is float32
+        if audio.dtype != np.float32:
+            audio = audio.astype(np.float32)
+        
+        # Resample if necessary
+        if sr != 16000:
+            audio = librosa.resample(audio, orig_sr=sr, target_sr=16000)
+        
+        # Define conversation turns
         turns = [
             {'role': 'system', 'content': 'Respond naturally and informatively.'},
             {'role': 'user', 'content': ''}
         ]
 
-        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
+        # Run the pipeline with the audio and conversation turns
+        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': 16000}, max_new_tokens=512)
 
+        # Return the model's response
         return output
 
     except Exception as e:
-        return f"Error: {str(e)}"
+        return f"Error processing audio: {str(e)}"
 
 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(source="microphone", type="numpy"),  
+    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