app.py

import os
import streamlit as st
import torch
import tempfile
from groq import Groq
from diffusers import AutoPipelineForText2Image
from io import BytesIO

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

# Initialize Groq client
client = Groq(api_key=GROQ_API_KEY)

# Load image generation model
device = "cuda" if torch.cuda.is_available() else "cpu"
image_gen = AutoPipelineForText2Image.from_pretrained("stabilityai/sdxl-turbo", use_auth_token=HF_API_KEY).to(device)

# Function to transcribe Tamil audio using Groq's Whisper
def transcribe(audio_bytes):
    if not audio_bytes:
        return "No audio provided."

    # Save the audio file temporarily
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio:
        temp_audio.write(audio_bytes)
        temp_audio_path = temp_audio.name

    # Call Whisper API
    with open(temp_audio_path, "rb") as file:
        transcription = client.audio.transcriptions.create(
            file=file,
            model="whisper-large-v3",
            language="ta",
            response_format="verbose_json"
        )

    # Cleanup temp file
    os.remove(temp_audio_path)
    
    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")

# Upload audio file
audio_file = st.file_uploader("Upload a Tamil audio file", type=["wav", "mp3"])

if st.button("Generate"):
    if audio_file is not None:
        # Read audio bytes
        audio_bytes = audio_file.read()

        # Process Steps
        tamil_text = transcribe(audio_bytes)
        english_text = translate_text(tamil_text)
        story = generate_text(english_text)
        image = generate_image(english_text)

        # Display Outputs
        st.subheader("📝 Transcribed Tamil Text")
        st.write(tamil_text)

        st.subheader("🔠 Translated English Text")
        st.write(english_text)

        st.subheader("📖 Generated Story")
        st.write(story)

        st.subheader("🖼️ Generated Image")
        st.image(image, caption="Generated Image from Story")

    else:
        st.warning("⚠️ Please upload an audio file before generating.")