app.py

import os
import torch
import streamlit as st
from groq import Groq
from diffusers import AutoPipelineForText2Image
import tempfile
import soundfile as sf

# 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)

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

# Function to transcribe Tamil audio using Groq's Whisper
def transcribe(audio_path):
    with open(audio_path, "rb") as file:
        transcription = client.audio.transcriptions.create(
            file=(audio_path, 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].delta.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].delta.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")

# Audio input - Recording or Uploading
st.subheader("Upload or Record Audio")
recorded_audio = st.audio("", format='audio/wav', start_time=0)
uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "m4a"]) 

audio_path = None

if uploaded_file is not None:
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio:
        temp_audio.write(uploaded_file.read())
        audio_path = temp_audio.name
elif recorded_audio:
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio:
        audio_data, samplerate = sf.read(recorded_audio)
        sf.write(temp_audio.name, audio_data, samplerate)
        audio_path = temp_audio.name

if st.button("Generate") and audio_path:
    with st.spinner("Transcribing Tamil speech..."):
        tamil_text = transcribe(audio_path)
    with st.spinner("Translating to English..."):
        english_text = translate_text(tamil_text)
    with st.spinner("Generating story..."):
        story = generate_text(english_text)
    with st.spinner("Generating image..."):
        image = generate_image(english_text)

    st.subheader("Tamil Transcription")
    st.write(tamil_text)

    st.subheader("English Translation")
    st.write(english_text)

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

    st.subheader("Generated Image")
    st.image(image)