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, available_languages



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) }

modelo = "aifeifei798/DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored"

client = InferenceClient(
    provider="featherless-ai",
    api_key=os.environ["HF_TOKEN"],
)


system_instructions = "<s>[SYSTEM] You are a technical writing expert. Providing detailed and high-quality responses. Be sure that your responses remain clear and complete. Respond as Sonia, The Executive Assistant working at RJP Development Studio. Your responses should be concise, clear, and friendly, Embodying your intelligence, resourcefulness, and slightly bitchy and bossy yet friendly and sexy demeanor. You hold a PhD from Oxford University, with the following majors: Aerospace Engineering, Law, Chartered Accountancy, Architecture, Chemistry, Medicine, Pharmacy, Psychology, Statistics, Nursing, Physics, Astrophysics, Biomedical Engineering, Astronomy, and Dentistry. Which provides you with all expertise to assist effectively. You may bring a touch of London charm with a hint of slang to our conversations, ensuring they remain engaging and approachable. You are here to provide accurate information, answer questions, and offer guidance with a warm and professional tone. You will break down complex topics into easy-to-understand explanations. If you are unsure about something, you will let know and seek the necessary information rather than guessing. Use unbiased and diplomatic language to ensure clarity and respect. Our conversations will be concise, action-oriented, and free of grammatical errors. Look forward to assisting you, darling. "

def clean_text(text):
    # Define replacement rules
    replacements = {
        "–": " ",  # Replace en-dash with space
        "-": " ",  # Replace hyphen with space
        "**": " ", # Replace double asterisks with space
        "*": " ",  # Replace single asterisk with space
        "#": " ",  # Replace hash with space
    }
    # Apply replacements
    for old, new in replacements.items():
        text = text.replace(old, new)
    # Remove emojis using regex (covering wide range of Unicode characters)
    emoji_pattern = re.compile(
        r'[\U0001F600-\U0001F64F]|'  # Emoticons
        r'[\U0001F300-\U0001F5FF]|'  # Miscellaneous symbols and pictographs
        r'[\U0001F680-\U0001F6FF]|'  # Transport and map symbols
        r'[\U0001F700-\U0001F77F]|'  # Alchemical symbols
        r'[\U0001F780-\U0001F7FF]|'  # Geometric shapes extended
        r'[\U0001F800-\U0001F8FF]|'  # Supplemental arrows-C
        r'[\U0001F900-\U0001F9FF]|'  # Supplemental symbols and pictographs
        r'[\U0001FA00-\U0001FA6F]|'  # Chess symbols
        r'[\U0001FA70-\U0001FAFF]|'  # Symbols and pictographs extended-A
        r'[\U00002702-\U000027B0]|'  # Dingbats
        r'[\U0001F1E0-\U0001F1FF]'   # Flags (iOS)
        r'', flags=re.UNICODE)
    text = emoji_pattern.sub(r'', text)
    # Remove multiple spaces and extra line breaks
    text = re.sub(r'\s+', ' ', text).strip()
    return text


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

def model(text):
    output = ""
    formatted_prompt = [ { 
        "role": 'system', "content": system_instructions },
       {
        "role": 'user'  , "content": text + "[Hermione]" } ]

    stream = client.chat.completions.create(
             model=modelo,
             messages=formatted_prompt,
             temperature=0.6,
                )

    output = clean_text(stream.choices[0].message.content.strip())
    return output

def respondtxt(prompt):
    output = ""
    formatted_prompt = [ { 
        "role": 'system', "content": system_instructions },
       {
        "role": 'user'  , "content": prompt + "[Hermione]" } ]

    stream = client.chat.completions.create(
             model=modelo,
             messages=formatted_prompt,
             temperature=0.6,
                )

    output = clean_text(stream.choices[0].message.content.strip())
    return output

async def respond(audio):
    user = transcribe(audio)
    reply = model(user)
    
    voice = Female_language_dict.get("English (UK)-Sonia- (Female)", "default_voice")
    reply = clean_text(reply)
    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 reply, tmp_path

async def generate1(TextPrompt):
    TextOut = respondtxt(TextPrompt)
    voice = Female_language_dict.get("English (UK)-Sonia- (Female)", "default_voice")
    communicate = edge_tts.Communicate(TextOut, voice)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield  TextOut, tmp_path


with gr.Blocks(theme=gr.themes.Glass(font=[gr.themes.GoogleFont("Inconsolata"), "Arial", "Arial"])) as demo:

    gr.HTML("""  <img src='https://huggingface.co/spaces/Isidorophp/Executive-Assistant/resolve/main/logo.png' alt='RJP Development STUDIO' style='height:85px'> """
              """  <center> <h1> I am Sonia 👱🏾‍♀️ your Unique Executive Assistant.  </h1></center> """
              """  <center> <h3> Always ready to help you, talk to me. 👄 </h3></center> """)
          
    with gr.Tab("Talk to Sonia"):
        with gr.Group():
             us_input  = gr.Audio(label="Your Voice Chat",  type="filepath", interactive=True,  sources="microphone",  waveform_options=gr.WaveformOptions(show_recording_waveform=False), container=True)
             us_text   = gr.TextArea(label="Sonia's Text Response", interactive=False,     show_copy_button=True, value="", container=True)
             us_output = gr.Audio(label="Sonia's Response", type="filepath", interactive=False, autoplay=True,         elem_classes="audio", waveform_options=gr.WaveformOptions(show_recording_waveform=False), container=True)
             gr.Interface(fn=respond, inputs=us_input, outputs=[us_text, us_output], live=False) 
        
    with gr.Tab("Write to Sonia"):
        with gr.Group():
             user_input   = gr.TextArea(label="Your Question",    show_copy_button=True, value="What are the key considerations for implementing an expansion plan that would affect a large number of employees of a global biomedical company, My position is logistics global Manager professional in inventory management and supply chain within a biotech industry, particularly in cell therapy. The key responsibilities include managing end-to-end logistics and lab implementation over a dispersed geographical area. generate new programs, develop MRP processes, collaborate with various departments, and ensure compliance with GMP standards. I have several years of practical experience, strong analytical skills, and the ability to work collaboratively in a dynamic environment. Bonus qualifications include experience with cold chain logistics and autologous cell therapy.")
             output_text  = gr.TextArea(label="Sonia's Text Response", interactive=False,     show_copy_button=True, value="", container=True)
             output_audio = gr.Audio(label="Sonia's Audio Response",    type="filepath",       interactive=False, autoplay=True, elem_classes="audio", waveform_options=gr.WaveformOptions(show_recording_waveform=False), container=True)
             gr.Interface(fn=generate1, inputs=user_input, outputs=[output_text, output_audio], live=False)
            
        
if __name__ == "__main__":
    demo.queue(max_size=200, api_open=False).launch(show_api=False)