FSD50K

dataset·sound event audio·active

Open human-labeled Freesound clips across 200 AudioSet-ontology classes, including many musical sounds.

Verified facts

Official site	https://zenodo.org/records/4060432 ↗
Geography	Spain
doi	10.5281/zenodo.4060432
size	clips: 51197 · classes: 200 · duration hours: over 100
scope	General sound-event recognition with music and instrument classes.
genres	mixed audio; includes music
license	Dataset curation CC BY; clips individually CC0, CC BY, CC BY-NC or Sampling+
steward	Music Technology Group, Universitat Pompeu Fabra
creators	Eduardo Fonseca; Xavier Favory; Jordi Pons; Frederic Font; Xavier Serra
versions	1.0
languages	multilingual descriptions/unspecified
modalities	audio clips; multi-label annotations; clip metadata; license mappings
geographies	Global
access method	Open Zenodo multipart deposit.
intended uses	sound event recognition; audio representation learning; music/instrument classification
related papers	FSD50K: An Open Dataset of Human-Labeled Sound Events
annotation method	Human labeling using the AudioSet ontology; label-noise analysis documented.
collection method	Curated from Freesound uploads.
provenance claims	Per-clip uploader, metadata and license are included.
current availability	Zenodo files available.
commercial use limits	Some clips prohibit commercial reuse.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Class distribution is unequal.; Global license choice does not replace individual clip licenses.