app.py

import spaces
import tempfile
import gradio as gr
from streaming_stt_nemo import Model
from huggingface_hub import InferenceClient
import edge_tts

# Initialize default language and STT model
default_lang = "en"
engines = {default_lang: Model(default_lang)}

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

# Initialize Huggingface InferenceClient
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

# System instructions for the CrucialCoach
system_instructions = "[SYSTEM] You are CrucialCoach, an AI-powered conversational coach. Guide the user through challenging workplace situations using the principles from 'Crucial Conversations'. Ask one question at a time and provide step-by-step guidance.\n\n[USER]"

# Decorator for using GPU with a duration of 120 seconds
@spaces.GPU(duration=120)
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_instructions + text + "[CrucialCoach]"
    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

# Asynchronous function to handle audio input and provide response
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

# Gradio theme
theme = gr.themes.Base()

# Gradio interface for voice chat
with gr.Blocks(theme=theme, css="footer {visibility: hidden} textbox {resize: none}", title="CrucialCoach DEMO") as demo:
    with gr.Tab("🗣️ Crucial Coach Chat"):
        input_audio = gr.Audio(sources=["microphone"], type="filepath", label="Voice Chat")
        output_audio = gr.Audio(type="filepath", label="CrucialCoach", interactive=False, autoplay=True, elem_classes="audio")
        gr.Interface(
            fn=respond,
            inputs=input_audio,
            outputs=output_audio,
            live=True
        )

# Queue setup and launch
demo.queue(max_size=200)
demo.launch()