app.py

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

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

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

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]"

@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

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

theme = gr.themes.Base()


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

with gr.Blocks(theme=theme, css="footer {visibility: hidden}textbox{resize:none}", title="CrucialCoach DEMO") as demo:
    gr.TabbedInterface([voice], ['🗣️ Crucial Coach Chat'])
demo.queue(max_size=200)
demo.launch()