app.py

import gradio as gr
import edge_tts
import asyncio
import tempfile
import os
from huggingface_hub import InferenceClient
import random
import torch

default_lang = "en"

# Sample STT transcription logic
def transcribe(audio):
    # This is a placeholder; replace with your STT logic
    return "Transcribed text from audio."

# Sample text generation logic using Hugging Face models
def client_fn(model):
    # Placeholder to return appropriate model based on the user's selection
    return InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

def randomize_seed_fn(seed: int) -> int:
    return random.randint(0, 999999)

system_instructions1 = """
[SYSTEM] Answer as a Swahili AI, made by 'Laocta Tech labs.'
Keep conversation friendly, short, clear, and concise.
[USER]
"""

def models(text, model="Mixtral 8x7B", seed=42):
    seed = int(randomize_seed_fn(seed))
    generator = torch.Generator().manual_seed(seed)
    client = client_fn(model)
    generate_kwargs = dict(
        max_new_tokens=300,
        seed=seed
    )
    formatted_prompt = system_instructions1 + text + "[Sema]"
    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, model, seed):
    user = transcribe(audio)
    reply = models(user, model, seed)
    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)
    yield tmp_path

# UI Design Customization using CSS
custom_css = """
    body {
        background-color: #1a2b40;
        color: white;
        text-align: center;
    }
    .container {
        display: flex;
        justify-content: center;
        align-items: center;
        height: 90vh;
        flex-direction: column;
    }
    .microphone-button {
        width: 120px;
        height: 120px;
        background-color: #00d084;
        border-radius: 50%;
        display: flex;
        justify-content: center;
        align-items: center;
        margin-bottom: 20px;
        cursor: pointer;
        transition: background-color 0.3s ease;
    }
    .microphone-button:hover {
        background-color: #009f65;
    }
    .microphone-icon {
        font-size: 48px;
        color: white;
    }
    .try-button {
        background-color: #00d084;
        color: white;
        padding: 10px 20px;
        border-radius: 20px;
        font-size: 18px;
        cursor: pointer;
        border: none;
        transition: background-color 0.3s ease;
    }
    .try-button:hover {
        background-color: #009f65;
    }
"""

DESCRIPTION = """# <center><b>Sema-Ai⚡</b></center>
    ### <center>Build, test, and deploy voice agents quickly and easily</center>
    """

with gr.Blocks(css=custom_css) as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Row(elem_id="container"):
        with gr.Column():
            # Microphone Button UI
            mic_btn = gr.Audio(label="Click the microphone", 
                               sources="microphone", 
                               type="filepath", 
                               elem_classes="microphone-button",
                               show_label=False)

            mic_icon = gr.Markdown("🎤", elem_classes="microphone-icon")
            
            # Output Audio
            output = gr.Audio(label="AI", type="filepath", interactive=False, autoplay=True)

            # Model selection and Seed hidden UI elements
            select = gr.Dropdown(
                ['Mixtral 8x7B', 'Llama 3 8B', 'Mistral 7B v0.3', 'Phi 3 mini'],
                value="Mistral 7B v0.3",
                label="Model"
            )

            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=999999,
                step=1,
                value=0,
                visible=False
            )

            # Try Button (action trigger)
            try_btn = gr.Button("Give it a try!", elem_classes="try-button")

    try_btn.click(fn=respond, inputs=[mic_btn, select, seed], outputs=[output], live=True)

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