app.py

import gradio as gr
from transformers import (
    pipeline,
    AutoProcessor,
    AutoModelForCausalLM,
    AutoTokenizer,
    GenerationConfig,
    TextStreamer
)
from datasets import load_dataset
import torch
import numpy as np

# Set device and dtype
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Image Captioning
caption_model = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")

# Text-to-Speech
synthesiser = pipeline("text-to-speech", model="microsoft/speecht5_tts")

# Florence-2-base for OCR
ocr_model = AutoModelForCausalLM.from_pretrained(
    "microsoft/Florence-2-base",
    torch_dtype=torch_dtype,
    trust_remote_code=True
).to(device)
ocr_processor = AutoProcessor.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True)

# Doge model for context generation
doge_tokenizer = AutoTokenizer.from_pretrained("SmallDoge/Doge-320M-Instruct")
doge_model = AutoModelForCausalLM.from_pretrained("SmallDoge/Doge-320M-Instruct", trust_remote_code=True).to(device)
doge_config = GenerationConfig(
    max_new_tokens=100,
    use_cache=True,
    do_sample=True,
    temperature=0.8,
    top_p=0.9,
    repetition_penalty=1.0
)

# Speaker embedding (600-dim)
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
embedding = None
for entry in embeddings_dataset:
    vector = torch.tensor(entry["xvector"]).unsqueeze(0)
    if vector.shape[1] >= 600:
        embedding = vector[:, :600]
        break
if embedding is None:
    raise ValueError("No suitable speaker embedding of at least 600 dimensions found.")

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

        # OCR
        ocr_inputs = ocr_processor(text="<OCR>", images=image, return_tensors="pt").to(device, torch_dtype)
        generated_ids = ocr_model.generate(
            input_ids=ocr_inputs["input_ids"],
            pixel_values=ocr_inputs["pixel_values"],
            max_new_tokens=1024,
            num_beams=3,
            do_sample=False
        )
        extracted_text = ocr_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

        # Doge context generation
        prompt = f"Determine the context of this image. Caption: {caption} Extracted text: {extracted_text}"
        conversation = [{"role": "user", "content": prompt}]
        doge_inputs = doge_tokenizer.apply_chat_template(
            conversation=conversation,
            tokenize=True,
            return_tensors="pt"
        ).to(device)

        outputs = doge_model.generate(
            doge_inputs,
            generation_config=doge_config
        )
        context = doge_tokenizer.decode(outputs[0], skip_special_tokens=True)

        # TTS
        speech = synthesiser(
            context,
            forward_params={"speaker_embeddings": embedding}
        )
        audio = np.array(speech["audio"])
        rate = speech["sampling_rate"]

        return (rate, audio), caption, extracted_text, context

    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)"),
        gr.Textbox(label="Generated Context")
    ],
    title="SeeSay Contextualizer with Doge & BLIP",
    description="Upload an image to generate a caption, extract text, determine context, and convert it to audio."
)

iface.launch()