Rijgersberg/YouTube-Commons-nl-audio

YouTube Commons NL Audio

This dataset has audio files for the Dutch-language videos in Rijgersberg/YouTube-Commons-nl-transcriptions, all under a CC BY 4.0 license.

It contains 11,669 files for a total runtime of 2493h 43m 5s, coming in at approximately 130 GB.

Source

The original source of the dataset (minus the titles, descriptions and audio files) is YouTube Commons:

YouTube-Commons is a collection of audio transcripts of 2,063,066 videos shared on YouTube under a CC BY 4.0 license.

Content

The collection comprises 22,709,724 original and automatically translated transcripts from 3,156,703 videos (721,136 individual channels).

Usage

The datasets into actually two datasets wrapped into one. Firstly it is a HuggingFace Dataset of Rijgersberg/Rijgersberg/YouTube-Commons-nl-transcriptions. Secondly all the audio files are present in the audiofiles subfolder of the dataset.

The use it, it is recommended to use Git LFS to clone the dataset locally as follows:

$ git lfs install
$ git clone https://huggingface.co/datasets/Rijgersberg/YouTube-Commons-nl-audio

You can then load the HuggingFace Dataset with the video info and transcriptions from disk just as you normally would, and access the audio files directly with a bit of path mapping:

from pathlib import Path
from pprint import pprint

import datasets


folder = Path('/path/to/where/you/cloned/YouTube-Commons-nl-audio')

# load the dataset containing everything but the audio files
dataset = datasets.load_dataset('parquet', data_dir=folder, split='train')

# build up an index of audio file paths
paths = {
    path.stem: path  # the stem of the file is the video_id
    for path in (folder / 'audiofiles').rglob('*')
    if path.suffix in {".m4a", ".webm", '.mp4'}
}

# example: get the path to the audio file of the first video in the data set
video = dataset[0]
pprint(video)
print(f"The file can be found at {paths[video['video_id']]}")

Acquisition

The audio files were acquired using Youtube-DLP and the following code:

import random
import time
from pathlib import Path

from datasets import load_dataset
from yt_dlp import YoutubeDL, DownloadError


def main():
    dataset = load_dataset('Rijgersberg/YouTube-Commons-descriptions', split='train')
    dataset = dataset.filter(lambda row: row['language'] == 'nl')

    output_dir = Path('/path/to/output/dir/youtube-audio')
    output_dir.mkdir(exist_ok=True, parents=True)

    # clean up old partial downloads
    parts = list(output_dir.glob('*.part'))
    for f in parts:
        f.unlink()

    # skip ids that have already been downloaded
    existing = set(p.stem for p in output_dir.glob('*.*'))
    ids = set(dataset['id'])
    to_get = ids - existing

    urls = [f'https://www.youtube.com/watch?v={video_id}' for video_id in to_get]
    random.shuffle(urls)

    t = 5
    while urls:
        print(len(urls))
        url = urls.pop()

        try:
            with YoutubeDL({'format': 'bestaudio/best',
                            'extractaudio': True,
                            'outtmpl': str(output_dir / '%(id)s.%(ext)s'),
                            'quiet': True}) as ydl:
                ydl.download([url])
            t = 5
        except DownloadError as e:
            print(e)
            if 'try again later' in str(e):
                urls.append(url)
                random.shuffle(urls)

                time.sleep(t)
                t = min(2*t, 10*60)
            else:
                continue


if __name__ == "__main__":
    main()