app.py

import gradio as gr
import requests
from PIL import Image
from transformers import BlipProcessor, BlipForConditionalGeneration
from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
import IPython.display as ipd


processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large").to("cpu")


class Aspecto():
    pass

screen = Aspecto()
with gr.Blocks(theme=gr.themes.Ocean(primary_hue="pink", neutral_hue="indigo", font=[gr.themes.GoogleFont("Montserrat"), "Playwrite England SemiJoine", "Quicksand"])) as demo:
    textbox = gr.Textbox(label="Url")
    with gr.Row():
        button = gr.Button("Describir", variant="primary")
        clear = gr.Button("Borrar")
    output = gr.Textbox(label="Resumen")
    with gr.Row():
        button2 = gr.Button("Leer", variant="primary")
        clear = gr.Button("Borrar")
    output2 = gr.Audio(label="Audio")

    def describir(url):
      raw_image = Image.open(requests.get(url, stream=True).raw).convert('RGB')
      inputs = processor(raw_image, return_tensors="pt").to("cpu")
      out = model.generate(**inputs)
      return processor.decode(out[0], skip_special_tokens=True)

    def leer(texto):
        models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
        "facebook/fastspeech2-en-ljspeech",
        arg_overrides={"vocoder": "hifigan", "fp16": False}
        )
        model = models[0]
        TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
        generator = task.build_generator(model, cfg)
        
        text = texto
        
        sample = TTSHubInterface.get_model_input(task, text)
        wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)
        
        return ipd.Audio(wav, rate=rate)


    button.click(describir, [textbox], output)
    button2.click(leer, [output], output2)

demo.launch(debug=True)