Report SomethingUnreal built from the datasets alone
This page only knows about the videos the datasets list. To check every upload on the channel:
videos in AudioSet
A list of about a million ten-second YouTube clips, each labelled with the sounds in it, that Google published in 2017. Labs use it to train and test general audio and music models.
Including Music using ONLY sounds from Windows XP and 98!, Hatsune Miku singing in REALTIME! (リアルタイムで歌う初音ミク), [Download] BaWaMI (Basic Waveform MIDI Soft-synth) - Public Release!
video in MusicCaps
A set of 5,521 ten-second music clips from YouTube, each with a written description, that Google published in 2023. It is the standard test set for text-to-music models.
Including Music using ONLY sounds from Windows XP and 98!
Clip 30s to 40s · Open clip on YouTube · Labels: Music, Exciting music
youtube_video dsU3B0W3TMs · row eval:dsU3B0W3TMs:30:40
AudioSeteval
Clip 30s to 40s · Open clip on YouTube · Labels: Music, Speech
youtube_video 28OCzicDL_4 · row unbalanced_train:28OCzicDL_4:30:40
AudioSetunbalanced_train
Clip 210s to 220s · Open clip on YouTube · Labels: Music, Speech
youtube_video 2ebOKZRxG18 · row unbalanced_train:2ebOKZRxG18:210:220
AudioSetunbalanced_train
Clip 60s to 70s · Open clip on YouTube · Labels: Printer, Music
youtube_video 4EGSB-g2IfQ · row unbalanced_train:4EGSB-g2IfQ:60:70
AudioSetunbalanced_train
Clip 290s to 300s · Open clip on YouTube · Labels: Music, Speech, Inside, small room
youtube_video A1NByuDdEQM · row unbalanced_train:A1NByuDdEQM:290:300
AudioSetunbalanced_train
“The music features a glockenspiel playing a melody that starts off with a few notes and then develops into a fast-note arpeggio melody. Electronic sounds accompany the glockenspiel with a style that combines bass notes on beats one and three and chords on beats two and four. The whole atmosphere is playful.”
Clip 30s to 40s · Open clip on YouTube
youtube_video dsU3B0W3TMs · row dsU3B0W3TMs:30:40
MusicCaps
Last reviewed 2026-09-24.
Presence in a public research dataset shows that a clip of your recording was listed by academic or industry researchers as material for building or testing audio models. It does not by itself show which commercial products, if any, used the clip. The legal picture is moving. On 31 July 2026 the Regional Court of Munich I ruled in GEMA's case against Suno (case 42 O 763/25) that reproducing works for training, holding them in a model in retrievable form, and generating similar outputs can each infringe copyright under German and EU law; the judgment is not final and can be appealed. In the United States, Warner Music Group settled its claims against Suno on 25 November 2025 with a licensing deal, while the parallel claims by Universal and Sony were still pending at that point. Since then BMG (12 August 2026) and Believe/TuneCore (8 September 2026) have signed opt-in licensing deals with Suno.
3 routes. None is a guarantee.
JSON export SHA-256 9c5aa61fc07fe6c5edf301414c9c338c22182abce6f7bb5b941dd4ac554e9e80
| Dataset | File | Rows | Loaded | SHA-256 |
|---|---|---|---|---|
| audioset | balanced_train_segments.csv | 8,836 | 2026-09-24T10:30:36+00:00 | 3b83e5ff9ed72cfd… |
| audioset | eval_segments.csv | 7,848 | 2026-09-24T10:30:36+00:00 | f7055f53935f70e7… |
| audioset | unbalanced_train_segments.csv | 1,063,333 | 2026-09-24T10:30:43+00:00 | bb2dfd5e4f1ced39… |
| laion_disco_12m | train-00000-of-00005.parquet | 2,464,184 | 2026-09-24T12:18:19+00:00 | f8b1f0034f8f1e94… |
| laion_disco_12m | train-00001-of-00005.parquet | 2,464,183 | 2026-09-24T12:32:29+00:00 | 71d78ad14ee3f85d… |
| laion_disco_12m | train-00002-of-00005.parquet | 2,464,183 | 2026-09-24T12:49:06+00:00 | 0feba570a0e8e73c… |
| laion_disco_12m | train-00003-of-00005.parquet | 2,464,183 | 2026-09-24T13:06:37+00:00 | 21a77d99fd85ad2f… |
| laion_disco_12m | train-00004-of-00005.parquet | 2,464,183 | 2026-09-24T13:24:36+00:00 | 6ac5f59958b6b813… |
| musiccaps | musiccaps-public.csv | 5,521 | 2026-09-24T10:29:42+00:00 | 8cc109ac507b5e5a… |
These pages report where identifiers from public research datasets point. Presence in a dataset is a fact about the dataset, not a claim about the channel or artist, and says nothing about what any company did with the material. This report is informational and is not legal advice. It shows that identifiers matching your work are present in the named public research datasets; it does not establish how any particular company used them. If you are the subject of this page and want it removed, see removals.