app.py

import gradio as gr
import asyncio
import edge_tts
import os
from huggingface_hub import InferenceClient
import whisper
import torch
import tempfile


# Get the Hugging Face token from environment variable
hf_token = os.getenv("HF_TOKEN")
if not hf_token:
    raise ValueError("HF_TOKEN environment variable is not set")

# Initialize the Hugging Face Inference Client
client = InferenceClient("mistralai/Mistral-Nemo-Instruct-2407", token=hf_token)

# Load the Whisper model
whisper_model = whisper.load_model("tiny.en", device='cuda' if torch.cuda.is_available() else 'cpu')

# Initialize an empty chat history
chat_history = []

async def text_to_speech_stream(text):
    """Convert text to speech using edge_tts and return the audio file path."""
    communicate = edge_tts.Communicate(text, "en-US-AvaMultilingualNeural")
    audio_data = b""

    async for chunk in communicate.stream():
        if chunk["type"] == "audio":
            audio_data += chunk["data"]

    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_file:
        temp_file.write(audio_data)
        return temp_file.name

def whisper_speech_to_text(audio):
    """Convert speech to text using Whisper model."""
    try:
        result = whisper_model.transcribe(audio)
        return result['text']
    except Exception as e:
        print(f"Whisper Error: {e}")
        return None
    finally:
        if torch.cuda.is_available():
            torch.cuda.empty_cache()

async def chat_with_ai(message):
    global chat_history
    
    chat_history.append({"role": "user", "content": message})
    
    try:
        response = client.chat_completion(
            messages=[{"role": "system", "content": "You are a helpful voice assistant. Provide concise and clear responses to user queries."}] + chat_history,
            max_tokens=800,
            temperature=0.7
        )
        
        response_text = response.choices[0].message['content']
        chat_history.append({"role": "assistant", "content": response_text})
        
        audio_path = await text_to_speech_stream(response_text)
        
        return response_text, audio_path
    except Exception as e:
        print(f"Error: {e}")
        return str(e), None

def transcribe_and_chat(audio):
    text = whisper_speech_to_text(audio)
    if text is None:
        return "Sorry, I couldn't understand the audio.", None
    
    response, audio_path = asyncio.run(chat_with_ai(text))
    return response, audio_path

def create_demo():
    with gr.Blocks() as demo:
        gr.Markdown("# AI Voice Assistant")

        with gr.Row():
            with gr.Column(scale=1):
                audio_input = gr.Audio(type="filepath", label="Press 'Record' to Speak")

            with gr.Column(scale=1):
                chat_output = gr.Textbox(label="AI Response")
                audio_output = gr.Audio(label="AI Voice Response", autoplay=True)

        def process_audio(audio):
            response, audio_path = transcribe_and_chat(audio)
            return response, audio_path, None  # Return None to clear the audio input

        demo.load(None, js="""
            function() {
                document.querySelector("audio").addEventListener("stop", function() {
                    setTimeout(function() {
                        document.querySelector('button[title="Submit"]').click();
                    }, 500);
                });
                
                function playAssistantAudio() {
                    var audioElements = document.querySelectorAll('audio');
                    if (audioElements.length > 1) {
                        var assistantAudio = audioElements[1];
                        if (assistantAudio) {
                            assistantAudio.play();
                        }
                    }
                }

                document.addEventListener('gradioAudioLoaded', function(event) {
                    playAssistantAudio();
                });

                document.addEventListener('gradioUpdated', function(event) {
                    setTimeout(playAssistantAudio, 100);
                });
            }
        """)

        audio_input.change(process_audio, inputs=[audio_input], outputs=[chat_output, audio_output, audio_input])

    return demo

# Launch the Gradio app
if __name__ == "__main__":
    demo = create_demo()
    demo.launch(server_name="0.0.0.0", server_port=7860)