Correr Whisper en Perro Negro como benchmark

requirements.txt

git+https://github.com/openai/whisper.git
transformers
pytube
pandas

whisper-test.py

import whisper
from pytube import YouTube
from pathlib import Path
import logging
import time
import pandas as pd
from textwrap import TextWrapper

logging.basicConfig(level=logging.INFO)
model = whisper.load_model(name="base")


def get_text(url: str = None):
    if url:
        yt = YouTube(url)
        final_file_name = f"{yt.title}.mp3"
        if not Path(final_file_name).exists():
            video = yt.streams.filter(only_audio=True).first()
            out_file = video.download(output_path=".")

            file_ref = Path(out_file)
            file_stats = file_ref.stat()
            logging.info(f'Size of audio file in Bytes: {file_stats.st_size}')

            file_ref.rename(final_file_name)

        output = []
        start = time.time()
        for i in range(10):
            start_iter = time.time()
            result = model.transcribe(final_file_name)
            end_iter = time.time()
            output.append(end_iter - start_iter)
        end = time.time()

        # Variance and mean
        df = pd.DataFrame(output)
        print("")
        print(f"Var: {df.var()[0]}")
        print(f"Mean: {df.mean()[0]}")
        print(f"Time: {(end - start) / 10}")

        return result['text'].strip()
    return False


if __name__ == "__main__":
    text = get_text("https://www.youtube.com/watch?v=a6tgD_CsYTQ")
    wrapper = TextWrapper(width=120)
    print(wrapper.fill(text))