AcousticBrainz Genre Dataset

dataset·genre benchmark·active

Four genre-annotation datasets joining AcousticBrainz features with AllMusic, Discogs, Last.fm and Tagtraum labels.

Recorded facts

Official site	https://mtg.github.io/acousticbrainz-genre-dataset/ ↗
Geography	Spain
size	recordings: over 4 million across task materials · source datasets: 4
scope	Hierarchical multi-label music genre recognition.
genres	AllMusic; Discogs; Last.fm; Tagtraum taxonomies
license	CC BY-NC-SA 4.0 except AllMusic extracts for noncommercial scientific research only
steward	Music Technology Group
creators	Dmitry Bogdanov; Alastair Porter; Philip Tovstogan; Robert Wilmering; Xavier Serra
languages	multilingual metadata
modalities	AcousticBrainz features; genre labels; subgenre taxonomies; hidden test labels
geographies	Global
access method	Training and validation on Zenodo; hidden-ground-truth test set by request.
intended uses	genre and subgenre recognition; MediaEval benchmarking
related papers	The AcousticBrainz Genre Dataset
annotation method	Expert/editorial taxonomies and collaborative tags propagated or inferred into multi-label annotations.
collection method	Release-level and recording-level genre metadata joined to AcousticBrainz features.
provenance claims	Official page identifies four upstream metadata sources.
current availability	Development/validation available; test labels controlled.
commercial use limits	Commercial use prohibited for the release and especially AllMusic extracts.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Genre taxonomies disagree across communities.; Release-level annotations are propagated to tracks.; Hidden tests limit independent audit.
redistribution restrictions	AllMusic results must cite the source; test ground truth is not public.