app.py

import streamlit as st
import os
import torch
import soundfile as sf
from groq import Groq
from diffusers import AutoPipelineForText2Image
from streamlit_webrtc import webrtc_streamer, AudioRecorder

# 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").to(device)

# Function to transcribe audio
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",
            response_format="verbose_json"
        )
    return transcription["text"]

# Function to translate Tamil to English
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
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")

# Choose input method
input_method = st.radio("Choose Input Method:", ("Record Audio", "Upload Audio"))

audio_path = None

if input_method == "Record Audio":
    st.subheader("Record your Tamil speech")
    recorder = webrtc_streamer(key="record_audio", audio=True)
    
    if recorder.audio_receiver:
        audio_data = recorder.audio_receiver.get_frames()  # Get recorded audio
        audio_path = "recorded_audio.wav"
        sf.write(audio_path, audio_data, 16000)  # Save recorded audio
elif input_method == "Upload Audio":
    uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3"])
    if uploaded_file:
        audio_path = "uploaded_audio.wav"
        with open(audio_path, "wb") as f:
            f.write(uploaded_file.getbuffer())

if st.button("Generate"):
    if not audio_path:
        st.error("Please provide an audio file.")
        st.stop()

    # Process audio
    tamil_text = transcribe(audio_path)
    english_text = translate_text(tamil_text)
    story = generate_text(english_text)
    image = generate_image(english_text)

    # Display results
    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, caption="Generated Image")