app.py

import transformers
import gradio as gr
import librosa
import torch
import spaces

@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
     )

        audio, sr = librosa.load(audio_file, sr=16000)

        turns = [
            {'role': 'system', 'content': 'Respond naturally and informatively.'},
            {'role': 'user', 'content': '<|audio|>'}
        ]

        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
        print(output)

        return output

    except Exception as e:
        return f"Error: {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
    title="Live Transcription and Response",
    description="Speak into your microphone, and the model will respond naturally and informatively.",
    live=True  # Enable live processing
)

if __name__ == "__main__":
    iface.launch()