app.py

import gradio as gr
from transformers import pipeline, AutoProcessor, AutoModelForCausalLM
from datasets import load_dataset
import torch
import numpy as np

# Load BLIP model for image captioning
caption_model = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")

# Load SpeechT5 model for text-to-speech
synthesiser = pipeline("text-to-speech", model="microsoft/speecht5_tts")

# Load Florence-2 model for OCR
ocr_device = "cuda:0" if torch.cuda.is_available() else "cpu"
ocr_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
ocr_model = AutoModelForCausalLM.from_pretrained("microsoft/Florence-2-large", torch_dtype=ocr_dtype, trust_remote_code=True).to(ocr_device)
ocr_processor = AutoProcessor.from_pretrained("microsoft/Florence-2-large", trust_remote_code=True)

# Load speaker embedding
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)


def process_image(image):
    try:
        # Generate caption from the image
        caption = caption_model(image)[0]['generated_text']

        # Convert caption to speech
        speech = synthesiser(
            caption,
            forward_params={"speaker_embeddings": speaker_embedding}
        )

        # Extract text (OCR) using Florence-2
        inputs = ocr_processor(text="<OCR>", images=image, return_tensors="pt").to(ocr_device, ocr_dtype)
        generated_ids = ocr_model.generate(
            input_ids=inputs["input_ids"],
            pixel_values=inputs["pixel_values"],
            max_new_tokens=4096,
            num_beams=3,
            do_sample=False
        )
        extracted_text = ocr_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

        # Prepare audio data
        audio = np.array(speech["audio"])
        rate = speech["sampling_rate"]

        # Return audio, caption, and extracted text
        return (rate, audio), caption, extracted_text

    except Exception as e:
        return None, f"Error: {str(e)}", ""


# Gradio Interface
iface = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type='pil', label="Upload an Image"),
    outputs=[
        gr.Audio(label="Generated Audio"),
        gr.Textbox(label="Generated Caption"),
        gr.Textbox(label="Extracted Text (OCR)")
    ],
    title="SeeSay with SpeechT5 and Florence-2 OCR",
    description="Upload an image to generate a caption, hear it described with SpeechT5's speech synthesis, and extract text using Florence-2 OCR."
)

iface.launch()