app.py

import gradio as gr
import requests
import soundfile as sf
import tempfile
import os

# URL of your dedicated processing server.
# Adjust the SERVER_URL (and port/endpoint) as needed.
SERVER_URL = "http://204.12.245.139:5000/process_audio"

def process_audio(audio):
    """
    Receives audio from the browser (a tuple: (sample_rate, numpy_array)),
    writes it to a temporary WAV file, then sends it in a POST request
    to your GPU-accelerated server endpoint. Expects a JSON response with either
    a 'transcription' or 'response' key.
    """
    if audio is None:
        return "No audio provided. Please record something."
    
    sample_rate, audio_data = audio
    # Write audio to a temporary WAV file
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
        wav_path = tmp_file.name
        sf.write(wav_path, audio_data, sample_rate)
    
    try:
        with open(wav_path, "rb") as f:
            files = {"file": f}
            response = requests.post(SERVER_URL, files=files, timeout=30)
        if response.status_code == 200:
            json_data = response.json()
            # Try to retrieve 'transcription' then fallback to 'response'
            result = json_data.get("transcription") or json_data.get("response")
            if not result:
                result = "Server processed the audio, but did not return a result."
        else:
            result = f"Server error {response.status_code}: {response.text}"
    except Exception as e:
        result = f"Exception during processing: {e}"
    finally:
        os.remove(wav_path)
    return result

# Create a Gradio interface.
# Note: The "source" keyword is omitted because your installed Gradio version does not accept it.
iface = gr.Interface(
    fn=process_audio,
    inputs=gr.Audio(type="numpy", label="Record Your Voice"),
    outputs=gr.Textbox(label="Server Response"),
    title="Live AI Call Agent – Browser Mic Frontend",
    description=(
        "Record audio using your browser microphone. The audio will be sent to our dedicated "
        "server for processing with GPU acceleration. Your server should return a transcription or "
        "an AI-generated response."
    )
)

if __name__ == "__main__":
    # Launch the app so that it listens on all interfaces. Adjust the port if needed.
    iface.launch(server_name="0.0.0.0", server_port=7860)