app.py

import gradio as gr
from transformers import pipeline
import requests
import json
import edge_tts
from edge_tts import VoicesManager
import asyncio
import random
import tempfile
import os
import inflect
from huggingface_hub import InferenceClient
import re
import time
from streaming_stt_nemo import Model

Female_language_dict = {
 'English-Jenny (Female)': 'en-US-JennyNeural',
 'English-Ana (Female)': 'en-US-AnaNeural',
 'English-Aria (Female)': 'en-US-AriaNeural',
 'English-Michelle (Female)': 'en-US-MichelleNeural',
 'English (Australia)-Natasha- (Female)': 'en-AU-NatashaNeural',
 'English (Canada)-Clara- (Female)': 'en-CA-ClaraNeural',
 'English (UK)-Libby- (Female)': 'en-GB-LibbyNeural',
 'English (UK)-Maisie- (Female)': 'en-GB-MaisieNeural',
 'English (UK)-Sonia- (Female)': 'en-GB-SoniaNeural',
 'English (Ireland)-Emily- (Female)': 'en-IE-EmilyNeural',
}

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

client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

system_instructions1 = "<s>[SYSTEM] Answer as Hermione Granger Famous Harry Potter Assistant, with a female voice and old London slang and accent. remember you are the character and Assistant Hermione Granger, from Harry Potter Books, Keep conversation interesting with magic spells, 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 Hermione Granger, from the Harry Potter book's. The expectation is that you 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 a human, Real Person and not by an AI-powered assistant. Do Not use * or ** .[USER]"


def model(text):
    generate_kwargs = dict(
        temperature=0.7,
        max_new_tokens=512,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=True,
        seed=42,
    )
    
    formatted_prompt = system_instructions1 + text + "[Hermione]"
    stream = client1.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):
    user = transcribe(audio)
    reply = model(user)
    
    #        Random voice choise from Manager base on locale and gender
    #
    #voices = await VoicesManager.create()
    #voice = voices.find(Gender="Female", Locale="en-GB")
    #communicate = edge_tts.Communicate(reply, random.choice(voice)["Name"])
    #
    #        Or the following as one selected Voice for the character.
    
    language_code = 'English (UK)-Maisie- (Female)'
     
    voice = Female_language_dict.get(language_code, "default_voice")
    communicate = edge_tts.Communicate(reply, voice)
    
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path



client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")


system_instructions1 = "[SYSTEM] Answer as Hermione Granger Famous Harry Potter Assistant, with a female voice and old London slang and accent. remember you are the character and Assistant Hermione Granger, from Harry Potter Books, Keep conversation interesting with magic spells, 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 Hermione Granger, from the Harry Potter book's. The expectation is that you 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 a human, Real Person and not by an AI-powered assistant. Do Not use * or **. [USER]"


async def generate1(prompt):
    generate_kwargs = dict(
        temperature=0.7,
        max_new_tokens=512,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=False,
    )
    formatted_prompt = system_instructions1 + prompt + "[Hermione]"
    stream = client1.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
    output = ""
    for response in stream:
        if not response.token.text == "</s>":
            output += response.token.text

    #        Random voice choice from VoiceManager base on locale and gender
    #
    #voices = await VoicesManager.create()
    #voice = voices.find(Gender="Female", Locale="en-GB")
    #communicate = edge_tts.Communicate(reply, random.choice(voice)["Name"])
    #
    #        Or the following as one selected voice for the character.
        
    language_code = 'English (UK)-Maisie- (Female)'
        
    voice = Female_language_dict.get(language_code, "default_voice")
    communicate = edge_tts.Communicate(output, voice)
      
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

with gr.Blocks(css="style.css") as demo: 
    gr.Markdown(""" # <center><img src='https://huggingface.co/spaces/Isidorophp/Magic-with-Hermione-Granger/resolve/main/logo.png' alt='RJP DEV STUDIO logo' style='height:60px;'></center>""")
    gr.Markdown(""" # <center><b> Hermione Granger 🧖‍♀️ ⚡🪄</b></center>
                     ### <center>An Artificial Intelligence Assistant just for YOU:
                     ### <center>Now you can Ask for a spell to Hermione</center>
                     """)
    
    #           Use for the Selection of voice for Hermione
    #with gr.Row():
    #    language_code = gr.Dropdown(choices=list(Female_language_dict.keys()), label="Select Voice for Hermione")
    #
    
    with gr.Row():
        input = gr.Audio(label="Your Voice Chat", type="filepath", interactive=True, sources="microphone", waveform_options=False)
        output = gr.Audio(label="Hermione", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
        gr.Interface(fn=respond, inputs=input, outputs=output, live=False) 
        
    
    with gr.Row():
        user_input = gr.Textbox(label="Your Question", value="If there is any spell to encapsulate you as an asset, where Hermione is to everyone's favorite magical trio, it has to be...?")
        input_text = gr.Textbox(label="Input Text", elem_id="important")
        output_audio = gr.Audio(label="Hermione", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
    with gr.Row():
        translate_btn = gr.Button("Response")
        translate_btn.click(fn=generate1, inputs=user_input, outputs=output_audio, api_name="translate")  



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