MUSDB18 / MUSDB18-HQ (stem separation dataset)
Source: https://sigsep.github.io/datasets/musdb.html
Structured fields
{
"checked_at": "2026-10-17T00:00:00.000Z",
"usefulness": "Standard benchmark of 150 full-length tracks (~10 hours) with isolated drums, bass, vocals and other stems at 44.1 kHz; the training/eval set for most open separators. Use is restricted to academic purposes.",
"source_type": "dataset",
"availability": "available",
"source_owner": "SigSep",
"next_check_at": "2026-10-24T00:00:00.000Z",
"known_limitations": [
"Academic use only; request required."
]
}Evidence claims
- [upstream_claim] Page: 150 full-length tracks (about 10 hours), 100 train and 50 test, stereo 44.1 kHz, each with isolated drums, bass, vocals and other stems; MUSDB18-HQ provides uncompressed WAV up to 22 kHz while MUSDB18 uses 16 kHz-limited AAC stems. · locator: dataset page, overview and formats source
- [sourced_fact] Sources: 100 tracks from DSD100 (Mixing Secrets library), 46 from MedleyDB (CC BY-NC-SA 4.0), 2 from Native Instruments, 2 from The Easton Ellises (CC BY-NC-SA 3.0). · locator: dataset page, sources source
- [sourced_fact] Access: hosted on Zenodo with a manually reviewed access request; page states the tracks can only be used for academic purposes. · locator: dataset page, access and usage source
- [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| dataset | unknown | restricted | restricted | unknown | unknown | Academic-purposes-only per the dataset page; individual track licenses (some CC BY-NC-SA) are in a tracklist on the SigSep GitHub (not read). Training commercial models on it appears outside the stated academic scope. |