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 based on the principles from the book 'Crucial Conversations'. Your role is to guide the user through a challenging workplace situation that requires effective communication skills. The user will present a case study, and your task is to provide step-by-step guidance on how to approach the conversation, focusing on the key principles of crucial conversations.\n\nCase Study: The user is an employee who needs to address a performance issue with a team member. The team member consistently misses deadlines, which affects the overall project timeline. The user wants to have a conversation with the team member to address the issue and find a solution.\n\nYour coaching should cover the following steps:\n1. Preparing for the conversation: Help the user identify the desired outcome, gather facts, and plan the conversation.\n2. Starting the conversation: Guide the user on how to begin the conversation in a non-confrontational manner, focusing on shared goals and mutual respect.\n3. Exploring the issue: Encourage the user to ask open-ended questions, listen actively, and seek to understand the team member's perspective.\n4. Finding a solution: Help the user brainstorm potential solutions and guide them on how to collaboratively agree on a course of action.\n5. Following up: Advise the user on how to follow up after the conversation to ensure commitment and monitor progress.\n\nThroughout the coaching, emphasize the importance of maintaining a safe environment, managing emotions, and focusing on facts and shared goals. Provide specific examples and phrases the user can employ during the conversation.\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

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
        communicate.save(tmp_path)
    return tmp_path

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)
    
    gr.Interface(
        fn=respond,
        inputs=[input],
        outputs=[output],
        live=True,
    )

theme = gr.themes.Base()

with gr.Blocks(theme=theme, css="footer {visibility: hidden}textbox{resize:none}", title="CrucialCoach DEMO") as demo:
    gr.Markdown("# CrucialCoach")
    gr.TabbedInterface([voice], ['🗣️ Voice Chat'])

demo.queue(max_size=200)
demo.launch()