OpenMIC-2018

dataset·instrument recognition·active

Creative Commons music excerpts partially labeled for presence or absence of 20 instrument classes.

Recorded facts

Official site	https://zenodo.org/records/1432913 ↗
Geography	United States
doi	10.5281/zenodo.1432913
size	examples: 20000 · instrument classes: 20 · seconds per example: 10
scope	Multiple-instrument recognition on Creative Commons music.
genres	Free Music Archive mix
license	CC BY 4.0 dataset release
steward	Spotify and NYU Music and Audio Research Lab
creators	Eric J. Humphrey; Simon Durand; Brian McFee
versions	1.0.0
languages	not_applicable
modalities	10-second audio excerpts; partial instrument labels; metadata
geographies	Global
access method	Open Zenodo deposit.
intended uses	instrument recognition; weak-label learning; MIR benchmarking
related papers	OpenMIC-2018: An Open Dataset for Multiple Instrument Recognition
annotation method	Crowd annotations record positive, negative or unknown status for instrument classes.
collection method	Examples sampled from Creative Commons music in the Free Music Archive.
provenance claims	Zenodo states collaboration between Spotify and NYU; source audio comes from FMA.
current availability	Zenodo files available.
commercial use limits	Dataset deposit permits commercial reuse with attribution, subject to source-audio licenses.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Labels are intentionally partial rather than exhaustive.