app.py

import os
import torch
import streamlit as st
from groq import Groq
from diffusers import AutoPipelineForText2Image

# Load API keys
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
HF_API_KEY = os.getenv("HF_API_KEY")

# Initialize Groq client with API key
client = Groq(api_key=GROQ_API_KEY)

# Select device (GPU if available, else CPU)
device = "cuda" if torch.cuda.is_available() else "cpu"
st.write(f"Using device: {device}")  # Display device info

# Load lightweight Hugging Face image generation model
image_gen = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/sdxl-turbo", use_auth_token=HF_API_KEY
)
image_gen.to(device)

# Function to transcribe Tamil audio using Groq's Whisper
def transcribe(audio_file):
    with open(audio_file, "rb") as file:
        transcription = client.audio.transcriptions.create(
            file=(audio_file, file.read()),
            model="whisper-large-v3",
            language="ta",  # Tamil
            response_format="verbose_json"
        )
    return transcription["text"]

# Function to translate Tamil to English using Groq's Gemma
def translate_text(tamil_text):
    response = client.chat.completions.create(
        model="gemma-7b-it",
        messages=[{"role": "user", "content": f"Translate this Tamil text to English: {tamil_text}"}]
    )
    return response.choices[0].message.content

# Function to generate text using Groq's DeepSeek R1
def generate_text(prompt):
    response = client.chat.completions.create(
        model="deepseek-coder-r1-7b",
        messages=[{"role": "user", "content": f"Write a short story about: {prompt}"}]
    )
    return response.choices[0].message.content

# Function to generate an image
def generate_image(prompt):
    img = image_gen(prompt=prompt).images[0]
    return img

# Streamlit UI
st.title("Tamil Speech to Image & Story Generator")

# File uploader for audio
uploaded_audio = st.file_uploader("Upload your Tamil speech", type=["wav", "mp3", "m4a"])

if uploaded_audio is not None:
    st.audio(uploaded_audio, format="audio/wav")
    
    if st.button("Generate"):
        with st.spinner("Transcribing..."):
            tamil_text = transcribe(uploaded_audio)
            st.success("Transcription complete!")
            st.text_area("Tamil Text Output", tamil_text)
        
        with st.spinner("Translating to English..."):
            english_text = translate_text(tamil_text)
            st.success("Translation complete!")
            st.text_area("Translated English Text", english_text)
        
        with st.spinner("Generating story..."):
            story = generate_text(english_text)
            st.success("Story generation complete!")
            st.text_area("Generated Story", story)
        
        with st.spinner("Generating image..."):
            image = generate_image(english_text)
            st.success("Image generation complete!")
            st.image(image, caption="Generated Image")