Music Data for AI Training: 2,819 Licensed Stems Packs, More Than 63,000 Stems, No Source Separation
Overview
Music Data for AI Training: 2,819 Licensed Stems Packs, More Than 63,000 Stems, No Source Separation is an audio dataset. Relevant sectors: AI & Machine Learning, Education and Media & Entertainment. Pricing is not published and is provided on request.
Stems is the largest part of the the provider catalogue: 2,819 stems packs from finished electronic tracks, more than 63,000 stems, about 4,200 hours of audio, counted on 14 October 2026. No stem in this set was made by a separation model. What it is Every WAV is one lane of the producer's own session, printed from the original project with its plugins loaded. The lanes keep the producer's names, normalised to one of 53 instrument labels in the manifest (kick, bass, lead, pad, vocal and so on). Where the session had MIDI, the MIDI comes with the stems. 2,250 of the tracks come with every lane of the session, 63,764 stems in all, about 4.3 TB, from Logic and Ableton sessions; 569 are packs the producers exported and sold themselves, in mixed DAWs and lane counts. 2,081 of the tracks are on the public store today; 169 are not listed yet, but they are in the catalogue for licensing. BPM, key and genre are set by the producer who made each track. Where it comes from The tracks were sold on WeMakeDanceMusic, a music production marketplace online since 2012. Every creator in this set accepted seller terms that name machine learning in writing, and every creator is paid on every licence. Nothing is scraped and nothing comes from the open internet. Many of these songs also have their DAW session and MIDI in the sets beside this one: 4,586 tracks in the catalogue have the session, the stems and the MIDI all lined up. Delivery Private S3 prefix or your own bucket, with a scoped key per de
Technical specifications
- Modalities
- Audio
- File formats
- WAV, JSONInferred
- Volume
Related datasets
More in Speech & Audio →Premium Studio-Grade, Multi-Speaker English Conversational Audio Dataset with Stems + Transcripts - 35K Hours, 2+ Speakers, 500+ New Hours/Month
This premium studio-grade conversational audio dataset delivers over 34.5K hours of multi-speaker English content captured from professional sports and talk studio…
- Language
- English
- Volume
- 35k h
- Domain
- Education +2
- Use case
- Speech recognition