app.py

import gradio as gr
import whisper
from pytube import YouTube

loaded_model = whisper.load_model("medium")
current_size = 'medium'
def inference(link):
  yt = YouTube(link)
  global audio_stream
  audio_stream = yt.streams.filter(only_audio=True, file_extension='mp4').first()
  path = audio_stream.download()
  #path = yt.streams.get_audio_only().download(mp3=True)
  options = whisper.DecodingOptions(language= 'Spanish', without_timestamps=True)
  results = loaded_model.transcribe(path)
  return results['text']

def change_model(size):
  if size == current_size:
    return
  loaded_model = whisper.load_model(size)
  current_size = size

def populate_metadata(link):
  yt = YouTube(link)
  return yt.thumbnail_url, yt.title

title=""
description=""
block = gr.Blocks()
with block:
    gr.HTML(
        """
            <div style="text-align: center; max-width: 500px; margin: 0 auto;">
              <div>
              </div>
            </div>
        """
    )
    with gr.Group():
        with gr.Group():
          sz = gr.Dropdown(label="Model Size", choices=['tiny', 'base','small', 'medium', 'large'], value='medium')
          
          link = gr.Textbox(label="YouTube Link")
          
          gr.Markdown("Ejemplo:  https://www.youtube.com/watch?v=bnvgcQB01mQ")
        
          
          with gr.Row():
            title = gr.Label(label="Video Title")
            img = gr.Image(label="Thumbnail")
          text = gr.Textbox(
              label="Transcription", 
              placeholder="Transcription Output",
              lines=5)
          with gr.Row(): 
              btn = gr.Button("Transcribe")   
               
          
          # Events
          btn.click(inference, inputs=[link], outputs=[text])
          link.change(populate_metadata, inputs=[link], outputs=[img, title])
          sz.change(change_model, inputs=[sz], outputs=[])

block.launch(debug=True)

demo = gr.Interface(css="footer {visibility: hidden}", examples='https://www.youtube.com/watch?v=bnvgcQB01mQ')