app.py

import argparse
import os
import soundfile as sf
import gradio as gr
import numpy as np
from groq import Groq
from models import build_model
from kokoro import generate
import torch

### NEW: We'll store the default voice name in a global variable so we can change it later.
DEFAULT_VOICE_NAME = 'bm_george'

def initialize_model():
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    model = build_model('/content/Kokoro-82M/fp16/kokoro-v0_19-half.pth', device)

    ### UPDATED: Use our DEFAULT_VOICE_NAME here, so we can swap it out later
    voice_name = DEFAULT_VOICE_NAME

    voicepack = torch.load(f'voices/{voice_name}.pt', weights_only=True).to(device)
    print(f'Loaded voice: {voice_name}')
    return model, voicepack, device

MODEL, VOICEPACK, DEVICE = initialize_model()
client = None

def initialize_groq(api_key):
    """
    Initialize the Groq client with the provided API key.
    """
    global client
    try:
        client = Groq(api_key=api_key)
        return "API key configured successfully"
    except Exception as e:
        return f"Error configuring API key: {str(e)}"

### NEW: Function to change voice pack after the program has started
def set_voice_name(voice):
    """
    Dynamically load the requested voicepack.
    """
    global VOICEPACK, DEFAULT_VOICE_NAME
    try:
        VOICEPACK = torch.load(f'voices/{voice}.pt', weights_only=True).to(DEVICE)
        DEFAULT_VOICE_NAME = voice  # store new default so we can keep track
        return f"Voice changed to {voice}"
    except FileNotFoundError:
        return f"Voice pack {voice} not found in voices/"

def answer(question):
    if not client:
        return "Please configure Groq API key first"
    try:
        chat_completion = client.chat.completions.create(
            messages=[
                {
                    "role": "system",
                    "content": "you are a helpful assistant. your answers must be short"
                },
                {
                    "role": "user",
                    "content": question
                }
            ],
            model="llama3-8b-8192",
            temperature=0.5,
            max_tokens=1024,
            top_p=1,
            stop=None,
            stream=False,
        )
        return chat_completion.choices[0].message.content
    except Exception as e:
        return f"Error: {str(e)}"

def conversation_pipeline(audio_path=None, text_input=None):
    if not client:
        return "Please configure Groq API key first", None, None
        
    if audio_path and os.path.exists(audio_path):
        with open(audio_path, "rb") as file:
            transcription = client.audio.transcriptions.create(
                file=(audio_path, file.read()),
                model="distil-whisper-large-v3-en",
                response_format="verbose_json",
            ).text
    elif text_input:
        transcription = text_input
    else:
        return None, None, None

    gemini_response = answer(transcription)
    generated_audio, _ = generate(MODEL, gemini_response, VOICEPACK, lang="a")

    return transcription, gemini_response, generated_audio

def process_audio(audio=None, text_input=None):
    if audio is not None:
        audio_path = "temp_audio.wav"
        sf.write(audio_path, audio[1], audio[0])
        transcription, response, audio_out = conversation_pipeline(audio_path=audio_path)
        os.remove(audio_path)
    else:
        transcription, response, audio_out = conversation_pipeline(text_input=text_input)

    if audio_out is not None:
        audio_out = np.array(audio_out).flatten().astype(np.float32)
        return (
            transcription if transcription else "",
            response if response else "",
            (24000, audio_out),
        )
    return "", "", None

def main():
    with gr.Blocks() as interface:
        gr.Markdown("# Voice Chat Interface")
        gr.Markdown("Configure API key and start chatting")

        # Row for API key
        with gr.Row():
            api_key = gr.Textbox(label="Groq API Key", type="password")
            api_status = gr.Textbox(label="API Status", interactive=False)
            configure_btn = gr.Button("Configure API")

        ### NEW: Row to change voice name on the fly
        with gr.Row():
            voice_name_input = gr.Textbox(label="Voice Name", value=DEFAULT_VOICE_NAME)
            voice_status = gr.Textbox(label="Voice Status", interactive=False)
            set_voice_btn = gr.Button("Set Voice")

        # Row for user input
        with gr.Row():
            audio_input = gr.Audio(sources=["microphone"], type="numpy", label="Speak")
            text_input = gr.Textbox(label="Or type your message here")

        # Row for outputs
        with gr.Row():
            transcription = gr.Textbox(label="Transcription", interactive=False)
            response = gr.Textbox(label="AI Response", interactive=False)

        with gr.Row():
            audio_output = gr.Audio(label="AI Voice Response", autoplay=True)

        # Configure API key
        configure_btn.click(
            initialize_groq,
            inputs=[api_key],
            outputs=[api_status]
        )

        ### NEW: Set voice dynamically
        set_voice_btn.click(
            set_voice_name,
            inputs=[voice_name_input],
            outputs=[voice_status]
        )

        # Process with either audio or text
        audio_input.change(
            process_audio,
            inputs=[audio_input, text_input],
            outputs=[transcription, response, audio_output]
        )

        text_input.submit(
            process_audio,
            inputs=[audio_input, text_input],
            outputs=[transcription, response, audio_output]
        )

    interface.launch(share=True)

if __name__ == "__main__":
    main()