app.py

import os
import io
import requests
import gradio as gr
from groq import Groq
from transformers import MarianMTModel, MarianTokenizer, AutoModelForCausalLM, AutoTokenizer
from deep_translator import GoogleTranslator
from PIL import Image, ImageDraw
import joblib
import time
import torch
import warnings
from huggingface_hub import InferenceApi
from diffusers import StableDiffusionPipeline
# Load text generation model and tokenizer
device = "cuda" if torch.cuda.is_available() else "cpu"
text_generation_model = AutoModelForCausalLM.from_pretrained("gpt2").to(device)  # Move model to the correct device
text_generation_tokenizer = AutoTokenizer.from_pretrained("gpt2")

# Set the padding token
text_generation_tokenizer.pad_token = text_generation_tokenizer.eos_token  # Use EOS token as padding token

# Function to transcribe, translate, analyze sentiment, and generate image
def process_audio(audio_path, image_option):
    if audio_path is None:
        return "Please upload an audio file.", None, None, None

    # Step 1: Transcribe audio
    try:
        with open(audio_path, "rb") as file:
            transcription = client.audio.transcriptions.create(
                file=(os.path.basename(audio_path), file.read()),
                model="whisper-large-v3",
                language="ta",
                response_format="verbose_json",
            )
        tamil_text = transcription.text
    except Exception as e:
        return f"An error occurred during transcription: {str(e)}", None, None, None

    # Step 2: Translate Tamil to English
    try:
        translator = GoogleTranslator(source='ta', target='en')
        translation = translator.translate(tamil_text)
    except Exception as e:
        return tamil_text, f"An error occurred during translation: {str(e)}", None, None

    # Step 3: Generate creative text
    def generate_creative_text(english_text):
        if not english_text:
            return "Please provide text to generate creative content."

        try:
            inputs = text_generation_tokenizer(english_text, return_tensors="pt", padding=True, truncation=True).to(device)  # Move inputs to the same device
            generated_tokens = text_generation_model.generate(
                **inputs,
                max_length=60,
                num_return_sequences=1,
                no_repeat_ngram_size=3,
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                early_stopping=True
            )
            creative_text = text_generation_tokenizer.decode(generated_tokens[0], skip_special_tokens=True).strip()
            return creative_text
        except Exception as e:
            return f"An error occurred during text generation: {str(e)}"

    creative_text = generate_creative_text(translation)

    # Step 4: Generate image (if selected)
    image = None
    if image_option == "Generate Image":
        try:
            image = pipe(translation).images[0]
        except Exception as e:
            return tamil_text, translation, f"An error occurred during image generation: {str(e)}", None

    return tamil_text, translation, image, creative_text

# Create Gradio interface
with gr.Blocks() as iface:
    gr.Markdown("# Audio Transcription, Translation, and Image Generation")
    with gr.Row():
        with gr.Column():
            audio_input = gr.Audio(type="filepath", label="Upload Audio File")
            image_option = gr.Dropdown(["Generate Image", "Skip Image"], label="Image Generation", value="Generate Image")
            submit_button = gr.Button("Process Audio")
        with gr.Column():
            tamil_text_output = gr.Textbox(label="Tamil Transcription", interactive=False)
            translation_output = gr.Textbox(label="English Translation", interactive=False)
            image_output = gr.Image(label="Generated Image")
            creative_text_output = gr.Textbox(label="Creative Text", interactive=False)

    submit_button.click(
        fn=process_audio,
        inputs=[audio_input, image_option],
        outputs=[tamil_text_output, translation_output, image_output, creative_text_output]
    )

# Launch the interface
iface.launch()