app.py

import pathlib
import time

import gradio as gr
import youtube_dl
import whisper

ydl_opts = {
    'format': 'bestaudio/best',
    'forcefilename': True
}
ydl = youtube_dl.YoutubeDL(ydl_opts)

models = {
    "tiny": whisper.load_model("tiny"),
    "base": whisper.load_model("base"),
    "small": whisper.load_model("small"),
    "medium": whisper.load_model("medium")
}

examples = [
    ["tiny", "https://www.youtube.com/watch?v=-tJYN-eG1zk", "en"],
    ["base", "https://www.youtube.com/watch?v=kMNPv_HXffQ", "es"]
]


def get_lyrics(model_v, video_url, language):
    t_init = time.time()
    info = ydl.extract_info(video_url, download=False)
    print(time.time() - t_init)
    file_path = ydl.prepare_filename(info)
    print(time.time() - t_init)
    # ydl.download([video_url])
    if not pathlib.Path(file_path).exists():
        ydl.process_info(info)
        print(time.time() - t_init)

    return info["title"], file_path, models[model_v].transcribe(file_path, language=language, fp16=False)["text"]


iface = gr.Interface(fn=get_lyrics,
                     inputs=[
                         gr.Dropdown(choices=list(models.keys()), value="base"),
                         "text",
                         gr.Dropdown(choices=["en", "es", "fr", "pt", "it", "ge", "ko", "ch", "ja", "de", "nr", "sw",
                                              "ar"], value="en")
                     ],
                     examples=examples,
                     outputs=[
                         gr.Text(label="Título"),
                         gr.Audio(label="Audio"),
                         gr.Textbox(label="Letra")
                     ],
                     cache_examples=True)
iface.launch()