app.py

import gradio as gr
import edge_tts
import asyncio
import tempfile
import os
from huggingface_hub import InferenceClient
import re
from streaming_stt_nemo import Model
import torch
import random

default_lang = "en"

engines = { default_lang: Model(default_lang) }

def transcribe(audio):
    lang = "en"
    model = engines[lang]
    text = model.stt_file(audio)[0]
    return text

HF_TOKEN = os.environ.get("HF_TOKEN", None)

def randomize_seed_fn(seed: int) -> int:
    seed = random.randint(0, 999999)
    return seed

system_instructions1 = """
[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark.' 
Keep conversation friendly, short, clear, and concise. 
Avoid unnecessary introductions and answer the user's questions directly. 
Respond in a normal, conversational manner while being friendly and helpful.
[USER]
"""

def models(text, seed=42):

    seed = int(randomize_seed_fn(seed))
    generator = torch.Generator().manual_seed(seed)  
    
    client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
    
    generate_kwargs = dict(
        max_new_tokens=300,
        seed=seed
    )
    
    formatted_prompt = system_instructions1 + text + "[JARVIS]"
    stream = client.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""
    for response in stream:
        if not response.token.text == "</s>":
            output += response.token.text

    return output

async def respond(audio, model, seed):
    user = transcribe(audio)
    reply = models(user, model, seed)
    communicate = edge_tts.Communicate(reply)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

DESCRIPTION = """ # <center><b>JARVIS⚡</b></center>
        ### <center>A personal Assistant of Tony Stark for YOU
        ### <center>Voice Chat with your personal Assistant</center>
        """

with gr.Blocks(css="style.css") as demo:    
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        seed = gr.Slider(
        label="Seed",
        minimum=0,
        maximum=999999,
        step=1,
        value=0,
        visible=False
        )
        input = gr.Audio(label="User", sources="microphone", type="filepath", waveform_options=False)
        output = gr.Audio(label="AI", type="filepath",
                        interactive=False,
                        autoplay=True,
                        elem_classes="audio")
        gr.Interface(
            batch=True,
            max_batch_size=10, 
            fn=respond, 
            inputs=[input, select, seed],
            outputs=[output], live=True)  
        
if __name__ == "__main__":
    demo.queue(max_size=200).launch()