app.py

import gradio as gr
from transformers import AutoProcessor, BlipForConditionalGeneration, AutoTokenizer,SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
import librosa
import numpy as np
import torch
import image_text_model
import audio_model
import open_clip

#CONSTANTS

# Carga el modelo de clasificación de imagen a texto
blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")

device = "cuda" if torch.cuda.is_available() else "cpu"
blip_model_large.to(device)


##### IMAGE MODEL TO TEXT, MODEL 1
def generate_caption(processor, model, image, tokenizer=None, use_float_16=False):
    inputs = processor(images=image, return_tensors="pt").to(device)

    if use_float_16:
        inputs = inputs.to(torch.float16)
    
    generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)

    if tokenizer is not None:
        generated_caption = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
    else:
        generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
   
    return generated_caption


def generate_caption_coca(model, transform, image):
    im = transform(image).unsqueeze(0).to(device)
    with torch.no_grad(), torch.cuda.amp.autocast():
        generated = model.generate(im, seq_len=20)
    return open_clip.decode(generated[0].detach()).split("<end_of_text>")[0].replace("<start_of_text>", "")


def generate_captions_speech(image):

    caption_blip_large = generate_caption(blip_processor_large, blip_model_large, image)
    print('generate_captions>>>'+caption_blip_large)
    return caption_blip_large,text_to_speech(caption_blip_large,"Surprise Me!")
    
#####END IMAGE MODEL TO TEXT
    
# Define la interfaz de usuario utilizando Gradio entradas y salidas
inputsImg = [
    gr.Image(type="pil", label="Imagen"),
]

#Salidas es lo que genera de tetxo y el audio
outputs = [ gr.Textbox(label="Caption generated by BLIP-large"),gr.Audio(type="numpy",label='Transcripcion')] 
title = "Clasificación de imagen a texto y conversión de texto a voz"
description = "Carga una imagen y obtén una descripción de texto de lo que contiene la imagen, así como un archivo de audio de la trasncripcion de la imagen en audio descrito."
examples = []

interface = gr.Interface(fn=generate_captions_speech, 
                         inputs=inputsImg,
                         outputs=outputs,
                         examples=examples, 
                         title=title,
                         description=description)
interface.launch()