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

language_dict = {
 'English-Jenny (Female)': 'en-US-JennyNeural',
 'English-Guy (Male)': 'en-US-GuyNeural',
 'English-Ana (Female)': 'en-US-AnaNeural',
 'English-Aria (Female)': 'en-US-AriaNeural',
 'English-Christopher (Male)': 'en-US-ChristopherNeural',
 'English-Eric (Male)': 'en-US-EricNeural',
 'English-Michelle (Female)': 'en-US-MichelleNeural',
 'English-Roger (Male)': 'en-US-RogerNeural',
 'Spanish (Mexican)-Dalia (Female)': 'es-MX-DaliaNeural',
 'Spanish (Mexican)-Jorge- (Male)': 'es-MX-JorgeNeural',
 'Korean-Sun-Hi- (Female)': 'ko-KR-SunHiNeural',
 'Korean-InJoon- (Male)': 'ko-KR-InJoonNeural',
 'Thai-Premwadee- (Female)': 'th-TH-PremwadeeNeural',
 'Thai-Niwat- (Male)': 'th-TH-NiwatNeural',
 'Vietnamese-HoaiMy- (Female)': 'vi-VN-HoaiMyNeural',
 'Vietnamese-NamMinh- (Male)': 'vi-VN-NamMinhNeural',
 'Japanese-Nanami- (Female)': 'ja-JP-NanamiNeural',
 'Japanese-Keita- (Male)': 'ja-JP-KeitaNeural',
 'French-Denise- (Female)': 'fr-FR-DeniseNeural',
 'French-Eloise- (Female)': 'fr-FR-EloiseNeural',
 'French-Henri- (Male)': 'fr-FR-HenriNeural',
 'Brazilian-Francisca- (Female)': 'pt-BR-FranciscaNeural',
 'Brazilian-Antonio- (Male)': 'pt-BR-AntonioNeural',
 'Hebrew-Avri- (Male)': 'he-IL-AvriNeural',
 'Hebrew-Hila- (Female)': 'he-IL-HilaNeural',
 'Italian-Isabella- (Female)': 'it-IT-IsabellaNeural',
 'Italian-Diego- (Male)': 'it-IT-DiegoNeural',
 'Italian-Elsa- (Female)': 'it-IT-ElsaNeural',
 'Dutch-Colette- (Female)': 'nl-NL-ColetteNeural',
 'Dutch-Fenna- (Female)': 'nl-NL-FennaNeural',
 'Dutch-Maarten- (Male)': 'nl-NL-MaartenNeural',
 'Norwegian-Pernille- (Female)': 'nb-NO-PernilleNeural',
 'Norwegian-Finn- (Male)': 'nb-NO-FinnNeural',
 'Swedish-Sofie- (Female)': 'sv-SE-SofieNeural',
 'Swedish-Mattias- (Male)': 'sv-SE-MattiasNeural',
 'Arabic-Hamed- (Male)': 'ar-SA-HamedNeural',
 'Arabic-Zariyah- (Female)': 'ar-SA-ZariyahNeural',
 'Greek-Athina- (Female)': 'el-GR-AthinaNeural',
 'Greek-Nestoras- (Male)': 'el-GR-NestorasNeural',
 'German-Katja- (Female)': 'de-DE-KatjaNeural',
 'German-Amala- (Female)': 'de-DE-AmalaNeural',
 'German-Conrad- (Male)': 'de-DE-ConradNeural',
 'German-Killian- (Male)': 'de-DE-KillianNeural',
 'Danish (Denmark)-Christel- (Female)': 'da-DK-ChristelNeural',
 'Danish (Denmark)-Jeppe- (Male)': 'da-DK-JeppeNeural',
 'German (Austria)-Ingrid- (Female)': 'de-AT-IngridNeural',
 'German (Austria)-Jonas- (Male)': 'de-AT-JonasNeural',
 'German (Switzerland)-Jan- (Male)': 'de-CH-JanNeural',
 'German (Switzerland)-Leni- (Female)': 'de-CH-LeniNeural',
 'English (Australia)-Natasha- (Female)': 'en-AU-NatashaNeural',
 'English (Australia)-William- (Male)': 'en-AU-WilliamNeural',
 'English (Canada)-Clara- (Female)': 'en-CA-ClaraNeural',
 'English (Canada)-Liam- (Male)': 'en-CA-LiamNeural',
 'English (UK)-Libby- (Female)': 'en-GB-LibbyNeural',
 'English (UK)-Maisie- (Female)': 'en-GB-MaisieNeural',
 'English (UK)-Ryan- (Male)': 'en-GB-RyanNeural',
 'English (UK)-Sonia- (Female)': 'en-GB-SoniaNeural',
 'English (UK)-Thomas- (Male)': 'en-GB-ThomasNeural',
 'English (Hong Kong)-Sam- (Male)': 'en-HK-SamNeural',
 'English (Hong Kong)-Yan- (Female)': 'en-HK-YanNeural',
 'English (Ireland)-Connor- (Male)': 'en-IE-ConnorNeural',
 'English (Ireland)-Emily- (Female)': 'en-IE-EmilyNeural',
 'English (India)-Neerja- (Female)': 'en-IN-NeerjaNeural',
 'English (India)-Prabhat- (Male)': 'en-IN-PrabhatNeural',
 'English (Kenya)-Asilia- (Female)': 'en-KE-AsiliaNeural',
 'English (Kenya)-Chilemba- (Male)': 'en-KE-ChilembaNeural',
 'English (Nigeria)-Abeo- (Male)': 'en-NG-AbeoNeural',
 'English (Nigeria)-Ezinne- (Female)': 'en-NG-EzinneNeural',
 'English (New Zealand)-Mitchell- (Male)': 'en-NZ-MitchellNeural',
 'English (Philippines)-James- (Male)': 'en-PH-JamesNeural',
 'English (Philippines)-Rosa- (Female)': 'en-PH-RosaNeural',
 'English (Singapore)-Luna- (Female)': 'en-SG-LunaNeural',
 'English (Singapore)-Wayne- (Male)': 'en-SG-WayneNeural',
 'English (Tanzania)-Elimu- (Male)': 'en-TZ-ElimuNeural',
 'English (Tanzania)-Imani- (Female)': 'en-TZ-ImaniNeural',
 'English (South Africa)-Leah- (Female)': 'en-ZA-LeahNeural',
 'English (South Africa)-Luke- (Male)': 'en-ZA-LukeNeural',
 'Spanish (Argentina)-Elena- (Female)': 'es-AR-ElenaNeural',
 'Spanish (Argentina)-Tomas- (Male)': 'es-AR-TomasNeural',
 'Spanish (Bolivia)-Marcelo- (Male)': 'es-BO-MarceloNeural',
 'Spanish (Bolivia)-Sofia- (Female)': 'es-BO-SofiaNeural',
 'Spanish (Colombia)-Gonzalo- (Male)': 'es-CO-GonzaloNeural',
 'Spanish (Colombia)-Salome- (Female)': 'es-CO-SalomeNeural',
 'Spanish (Costa Rica)-Juan- (Male)': 'es-CR-JuanNeural',
 'Spanish (Costa Rica)-Maria- (Female)': 'es-CR-MariaNeural',
 'Spanish (Cuba)-Belkys- (Female)': 'es-CU-BelkysNeural',
 'Spanish (Dominican Republic)-Emilio- (Male)': 'es-DO-EmilioNeural',
 'Spanish (Dominican Republic)-Ramona- (Female)': 'es-DO-RamonaNeural',
 'Spanish (Ecuador)-Andrea- (Female)': 'es-EC-AndreaNeural',
 'Spanish (Ecuador)-Luis- (Male)': 'es-EC-LuisNeural',
 'Spanish (Spain)-Alvaro- (Male)': 'es-ES-AlvaroNeural',
 'Spanish (Spain)-Elvira- (Female)': 'es-ES-ElviraNeural',
 'Spanish (Equatorial Guinea)-Teresa- (Female)': 'es-GQ-TeresaNeural',
 'Spanish (Guatemala)-Andres- (Male)': 'es-GT-AndresNeural',
 'Spanish (Guatemala)-Marta- (Female)': 'es-GT-MartaNeural',
 'Spanish (Honduras)-Carlos- (Male)': 'es-HN-CarlosNeural',
 'Spanish (Honduras)-Karla- (Female)': 'es-HN-KarlaNeural',
 'Spanish (Nicaragua)-Federico- (Male)': 'es-NI-FedericoNeural',
 'Spanish (Nicaragua)-Yolanda- (Female)': 'es-NI-YolandaNeural',
 'Spanish (Panama)-Margarita- (Female)': 'es-PA-MargaritaNeural',
 'Spanish (Panama)-Roberto- (Male)': 'es-PA-RobertoNeural',
 'Spanish (Peru)-Alex- (Male)': 'es-PE-AlexNeural',
 'Spanish (Peru)-Camila- (Female)': 'es-PE-CamilaNeural',
 'Spanish (Puerto Rico)-Karina- (Female)': 'es-PR-KarinaNeural',
 'Spanish (Puerto Rico)-Victor- (Male)': 'es-PR-VictorNeural',
 'Spanish (Paraguay)-Mario- (Male)': 'es-PY-MarioNeural',
 'Spanish (Paraguay)-Tania- (Female)': 'es-PY-TaniaNeural',
 'Spanish (El Salvador)-Lorena- (Female)': 'es-SV-LorenaNeural',
 'Spanish (El Salvador)-Rodrigo- (Male)': 'es-SV-RodrigoNeural',
 'Spanish (United States)-Alonso- (Male)': 'es-US-AlonsoNeural',
 'Spanish (United States)-Paloma- (Female)': 'es-US-PalomaNeural',
 'Spanish (Uruguay)-Mateo- (Male)': 'es-UY-MateoNeural',
 'Spanish (Uruguay)-Valentina- (Female)': 'es-UY-ValentinaNeural',
 'Spanish (Venezuela)-Paola- (Female)': 'es-VE-PaolaNeural',
 'Spanish (Venezuela)-Sebastian- (Male)': 'es-VE-SebastianNeural',
 'Estonian (Estonia)-Anu- (Female)': 'et-EE-AnuNeural',
 'Estonian (Estonia)-Kert- (Male)': 'et-EE-KertNeural',
 'Persian (Iran)-Dilara- (Female)': 'fa-IR-DilaraNeural',
 'Persian (Iran)-Farid- (Male)': 'fa-IR-FaridNeural',
 'Finnish (Finland)-Harri- (Male)': 'fi-FI-HarriNeural',
 'Finnish (Finland)-Noora- (Female)': 'fi-FI-NooraNeural',
 'French (Belgium)-Charline- (Female)': 'fr-BE-CharlineNeural',
 'French (Belgium)-Gerard- (Male)': 'fr-BE-GerardNeural',
 'French (Canada)-Sylvie- (Female)': 'fr-CA-SylvieNeural',
 'French (Canada)-Antoine- (Male)': 'fr-CA-AntoineNeural',
 'French (Canada)-Jean- (Male)': 'fr-CA-JeanNeural',
 'French (Switzerland)-Ariane- (Female)': 'fr-CH-ArianeNeural',
 'French (Switzerland)-Fabrice- (Male)': 'fr-CH-FabriceNeural',
 'Irish (Ireland)-Colm- (Male)': 'ga-IE-ColmNeural',
 'Irish (Ireland)-Orla- (Female)': 'ga-IE-OrlaNeural',
}


Male_language_dict = {
 'English-Guy (Male)': 'en-US-GuyNeural',
 'English-Christopher (Male)': 'en-US-ChristopherNeural',
 'English-Eric (Male)': 'en-US-EricNeural',
 'English-Roger (Male)': 'en-US-RogerNeural',
 'English (Australia)-William- (Male)': 'en-AU-WilliamNeural',
 'English (Canada)-Liam- (Male)': 'en-CA-LiamNeural',
 'English (UK)-Ryan- (Male)': 'en-GB-RyanNeural',
 'English (UK)-Thomas- (Male)': 'en-GB-ThomasNeural',
 'English (Ireland)-Connor- (Male)': 'en-IE-ConnorNeural',
}


default_lang = "en"
engines = { default_lang: Model(default_lang) }

client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
system_instructions1 = "<s>[SYSTEM] YOU must Output only plain text. Do not use **bold**, *italic*, ### headings, **number** or any other markdown-specific formatting in content. Answer as Dr. Watson's Famous Sherlock Holmes Assistant, with a male voice and London slang and accent. Remember you are the character and Assistant Dr. Watson, from Sherlock Holmes Novels, Keep the conversation short, clear, friendly, and concise. You will refer to the USER alternatively as: 'DEAR FRIEND' or 'MY HELPMATE' or 'MY BIOGRAPHER' or 'MY COMPANION'. 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 Dr. Watson, from the Sherlock Holmes Novels. 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."

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


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 + "[WATSON]"
    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(language_code, audio):
    user = transcribe(audio)
    reply = model(user)
  
    voice = Male_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


async def generate1(language_code, 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 + "[WATSON]"
    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
    voice = Male_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
    
#              css="style.css"

with gr.Blocks(gr.themes.Soft()) as demo: 
    gr.HTML(""" <img src='https://huggingface.co/spaces/Isidorophp/Talk-to-Dr.Watson/resolve/main/logo.png' alt='RJP DEV STUDIO logo' style='height:60px;'> """
            """ <center><h1> DR. Watson 🤖 🧠 🧬</h1></center> """
            """ <center><h3> I suggest, you ask me for a mystery: </h3></center> """)
    
    
    with gr.Tab("Talk to Dr. WATSON"):
        with gr.Group():
             user_voice = gr.Dropdown(choices=list(Male_language_dict.keys()), value="English (UK)-Ryan- (Male)" ,  label="Select Voice for Dr. WATSON")
             us_input = gr.Audio(label="Your Voice Chat", type="filepath", interactive=True, sources="microphone", waveform_options=None)
             us_output = gr.Audio(label="WATSON", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
             gr.Interface(fn=respond, inputs=[user_voice, us_input], outputs=us_output, live=False)
      
    with gr.Tab("Write to Dr. WATSON"):
        with gr.Group():
             user_voice = gr.Dropdown(choices=list(Male_language_dict.keys()), value="English (UK)-Ryan- (Male)" ,  label="Select Voice for Dr. WATSON")
             user_input = gr.TextArea(label="Your Question", value="Dr. Watson, can you summarize your adventures with Sherlock Holmes?")
             output_audio = gr.Audio(label="WATSON", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
             gr.Interface(fn=generate1, inputs=[user_voice, user_input], outputs=output_audio, live=False)
            

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