app.py

import gradio as gr
import edge_tts
import asyncio
import tempfile
import numpy as np
import soxr
from pydub import AudioSegment
import torch
import sentencepiece as spm
import onnxruntime as ort
from huggingface_hub import hf_hub_download, InferenceClient

# Speech Recognition Model Configuration
model_name = "neongeckocom/stt_en_citrinet_512_gamma_0_25"
sample_rate = 16000

# Download preprocessor, encoder and tokenizer
preprocessor = torch.jit.load(hf_hub_download(model_name, "preprocessor.ts", subfolder="onnx"))
encoder = ort.InferenceSession(hf_hub_download(model_name, "model.onnx", subfolder="onnx"))
tokenizer = spm.SentencePieceProcessor(hf_hub_download(model_name, "tokenizer.spm", subfolder="onnx"))

# Mistral Model Configuration
client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
system_instructions1 = "<s>[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. The request asks you to provide friendly responses as if You are the character Jarvis, made by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]"

def resample(audio_fp32, sr):
    return soxr.resample(audio_fp32, sr, sample_rate)

def to_float32(audio_buffer):
    return np.divide(audio_buffer, np.iinfo(audio_buffer.dtype).max, dtype=np.float32)

def transcribe(audio_path):
    audio_file = AudioSegment.from_file(audio_path)
    sr = audio_file.frame_rate
    audio_buffer = np.array(audio_file.get_array_of_samples())

    audio_fp32 = to_float32(audio_buffer)
    audio_16k = resample(audio_fp32, sr)

    input_signal = torch.tensor(audio_16k).unsqueeze(0)
    length = torch.tensor(len(audio_16k)).unsqueeze(0)
    processed_signal, _ = preprocessor.forward(input_signal=input_signal, length=length)
    
    logits = encoder.run(None, {'audio_signal': processed_signal.numpy(), 'length': length.numpy()})[0][0]

    blank_id = tokenizer.vocab_size()
    decoded_prediction = [p for p in logits.argmax(axis=1).tolist() if p != blank_id]
    text = tokenizer.decode_ids(decoded_prediction)

    return text

def model(text):
    formatted_prompt = system_instructions1 + text + "[JARVIS]"
    stream = client1.text_generation(formatted_prompt, max_new_tokens=512, stream=True, details=True, return_full_text=False)
    return "".join([response.token.text for response in stream if response.token.text != "</s>"])

async def respond(audio):
    user = transcribe(audio)
    reply = model(user)
    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)
    return tmp_path

with gr.Blocks() as demo:    
    with gr.Row():
        input = gr.Audio(label="Voice Chat (BETA)", sources="microphone", type="filepath", waveform_options=False)
        output = gr.Audio(label="JARVIS", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
        gr.Interface(fn=respond, inputs=[input], outputs=[output], live=True)

if __name__ == "__main__":
    demo.queue(max_size=200).launch()