app.py

import gradio as gr
import requests
import os
import tempfile
from groq import Groq

import torch
import soundfile as sf
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
from datasets import load_dataset


# ==============================
# API KEYS
# ==============================

HF_TOKEN = os.getenv("HF_TOKEN")
GROQ_API_KEY = os.getenv("GROQ_API_KEY")

groq_client = Groq(api_key=GROQ_API_KEY)


# ==============================
# LOAD TTS MODELS
# ==============================

processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")

tts_model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")

vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

embeddings_dataset = load_dataset(
    "Matthijs/cmu-arctic-xvectors",
    split="validation"
)

speaker_embeddings = torch.tensor(
    embeddings_dataset[7306]["xvector"]
).unsqueeze(0)


# ==============================
# TEXT → SPEECH
# ==============================

def text_to_speech(text):

    inputs = processor(
        text=text,
        return_tensors="pt",
        truncation=True,
        max_length=500
    )

    speech = tts_model.generate_speech(
        inputs["input_ids"],
        speaker_embeddings,
        vocoder=vocoder
    )

    temp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")

    sf.write(temp_audio.name, speech.numpy(), samplerate=16000)

    return temp_audio.name


# ==============================
# IMAGE GENERATION
# ==============================

def generate_image(prompt):

    API_URL = "https://router.huggingface.co/hf-inference/models/stabilityai/stable-diffusion-xl-base-1.0"

    headers = {
        "Authorization": f"Bearer {HF_TOKEN}"
    }

    payload = {
        "inputs": prompt
    }

    response = requests.post(API_URL, headers=headers, json=payload)

    print("HF STATUS:", response.status_code)

    if response.status_code != 200:
        print(response.text)
        return None

    image_bytes = response.content

    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".png")

    temp_file.write(image_bytes)

    temp_file.close()

    return temp_file.name


# ==============================
# GROQ CHATBOT
# ==============================

def ask_llm(question):

    response = groq_client.chat.completions.create(
        model="llama-3.1-8b-instant",
        max_tokens=200,
        messages=[
            {"role": "user", "content": question}
        ]
    )

    return response.choices[0].message.content


# ==============================
# MAIN ASSISTANT FUNCTION
# ==============================

def ai_assistant(user_input):

    reply = ask_llm(user_input)

    image = None

    if "image" in user_input.lower() or "generate" in user_input.lower():
        image = generate_image(user_input)

    audio = text_to_speech(reply)

    return reply, audio, image


# ==============================
# GRADIO UI
# ==============================

with gr.Blocks() as demo:

    gr.Markdown("# 🤖 AI Assistant (Chat + Voice + Image)")

    user_input = gr.Textbox(
        label="Ask something or request an image"
    )

    text_output = gr.Textbox(
        label="Assistant Response"
    )

    audio_output = gr.Audio(
        label="Voice Response"
    )

    image_output = gr.Image(
        label="Generated Image"
    )

    submit_btn = gr.Button("Submit")

    submit_btn.click(
        fn=ai_assistant,
        inputs=user_input,
        outputs=[text_output, audio_output, image_output]
    )


demo.launch()