williamTLmiller/youtube-transcripts-creative-commons

YouTube Video Transcripts Dataset: Creative Commons, Multilingual (CSV Sample)

Release name: YouTube-Commons Normalized v1: free CSV toy sample

A free, CSV-only toy sample of a normalized, de-duplicated and labeled build of YouTube-Commons: speech transcripts of YouTube videos that their creators published under the CC BY license.

Free toy sample

This public repo is a deliberately tiny, CSV-only toy sample: 500 videos plus the rows linked to them. It is meant for looking at the schema and running a pipeline once (a toy training run), not for training a real model. The full and training-ready versions (williamTLmiller/youtube-transcripts-creative-commons-full) are private and licensed separately; contact energea@energeatech.com.

What is in this toy sample

500 videos from 449 channels, with their 574 transcripts. For each video it includes the primary transcript (English when available) and the original-language transcript when that differs.

• Videos were picked by a fixed hash of video_id, spread over many channels (at most 2 videos per channel: 398 channels have 1 video here, 51 have 2).
• The pick makes sure every topic and format label appears, and that the validation and test splits each have at least 10 videos.
• Upload dates run from 2009-10-10 to 2024-04-23. The videos are in 18 original languages (en 340, hi 40, es 28, fr 21, it 10, id 9, vi 9, ar 7, de 7, pt 6, ko 6, nl 5, ru 5, tr 2, ja 2, iw 1, uk 1, hu 1).
• Transcripts are in 7 languages (en 496, es 29, fr 22, it 10, de 7, nl 5, ru 5): 410 are the original-language transcript and 164 are machine translations. Together they hold about 1.1 million words.

file	rows	one row per
csv/videos.csv	500	video
csv/channels.csv	449	channel that has a video in this sample
csv/transcripts.csv	574	video and transcript language

Every channel_id in videos.csv is in channels.csv, and every video_id in transcripts.csv is in videos.csv.

Full and trainable versions

These numbers describe the private releases, not this repo.

Package	Contents	Format
Free toy sample (this repo)	500 videos, 449 channels, 574 transcripts	CSV
Trainable (private)	3,112,197 videos with a non-empty transcript, one primary transcript each, train/validation/test 98/1/1	Parquet
Full (private)	3,156,666 videos, 585,789 channels, 21,736,981 transcripts in 128 languages	Parquet + CSV tables

The trainable and full versions are commercial products. Contact energea@energeatech.com.

Files

• csv/videos.csv, csv/channels.csv: tables described below
• csv/transcripts.csv: the transcripts, with channel and video_url repeated on each row for attribution

videos (one row per video)

video_id, video_url, title, channel, channel_id, channel_url, upload_date, upload_year, source_license, original_language, language_id_method, primary_transcript_language, primary_is_machine_translation, primary_word_count, topic, topic_confidence, format, quality_flags, transcript_languages, transcript_count, has_original_language_transcript, split

transcript_languages and transcript_count list every transcript the video has in the source; this sample only carries the primary and original-language ones, so most videos have fewer rows in transcripts.csv than transcript_count.

channels (one row per channel)

channel_id, channel, channel_url, video_count, first_upload_date, last_upload_date, main_original_language, main_topic, total_primary_words, split

The channel-level columns (video_count, first_upload_date, last_upload_date, main_original_language, main_topic, total_primary_words) describe the whole channel in the full release, not only the 1 or 2 videos it has in this sample.

transcripts (one row per video and language)

video_id, channel, video_url, language, is_original_language, is_machine_translation, word_count, character_count, quality_flags, split, text

Labels (silver)

Deterministic, rule-derived labels, reproducible from the public fields. They are not human-verified.

• topic (from the title plus the English transcript): education, science_technology, news_politics, government_public_meetings, religion_spirituality, health_medicine, business_finance, gaming, music, sports_fitness, food_cooking, travel_outdoors, autos_vehicles, how_to_diy, arts_culture_history, kids_family, other, unknown. unknown means no English evidence; other means weak evidence. topic_confidence is the winning topic's share of evidence. Videos per topic in this sample: other 80, unknown 75, education 48, religion_spirituality 33, business_finance 30, how_to_diy 29, gaming 28, science_technology 25, arts_culture_history 21, news_politics 19, autos_vehicles 18, health_medicine 18, music 16, kids_family 15, travel_outdoors 14, food_cooking 13, sports_fitness 11, government_public_meetings 7.
• format (from title and channel name): tts_narration, public_meeting, sermon_worship, lecture_talk, interview_podcast, news_report, tutorial_howto, review_unboxing, gameplay, music_performance, livestream, vlog, other. Videos per format in this sample: other 402, tutorial_howto 32, interview_podcast 16, lecture_talk 13, news_report 9, review_unboxing 6, music_performance 5, vlog 4, sermon_worship 3, livestream 3, public_meeting 3, gameplay 3, tts_narration 1.
• quality_flags (semicolon-separated, empty when clean): empty_transcript, very_short, mostly_music_or_noise, highly_repetitive, language_code_repaired. Videos flagged in this sample: very_short 95, mostly_music_or_noise 57, empty_transcript 10 (highly_repetitive and language_code_repaired do not occur here).
• split: assigned by channel, so no creator appears in more than one split. The full and trainable releases use train / validation / test = 98 / 1 / 1. This sample keeps the column as a toy split: 480 / 10 / 10 videos (433 / 7 / 9 channels, 552 / 11 / 11 transcripts).

License

• The MatrixNorm packaging layer (tables, labels, splits, documentation) is licensed CC BY-NC 4.0 (williamTLmiller), see LICENSE.
• The transcripts and video metadata themselves remain CC BY 4.0 from the original creators via PleIAs YouTube-Commons, see NOTICE and LICENSE-CC-BY-4.0-UNDERLYING-DATA.txt. Keep the creator attribution columns when you reuse them.

Source and limitations

• Source: PleIAs/YouTube-Commons (CC BY 4.0).
• Transcripts are automatic speech recognition and automatic translation output and contain errors.
• E-mail addresses and phone numbers in transcripts are masked as [EMAIL] / [PHONE].
• License status is as declared by each uploader on YouTube.
• In the full release a small number of source videos (under 0.1%) have no channel name or title; their video URL is kept as the attribution link. Every video in this sample has both.
• Labels are silver: use them as weak supervision, not ground truth.
• 500 videos is far too few to train a useful model; use this sample to inspect the schema and test code.

Contact: energea@energeatech.com