AudioSet

dataset·audio event labels·active

Human-labeled ten-second YouTube segments spanning music, speech and environmental sound classes.

Verified facts

Official site	https://research.google.com/audioset/dataset/index.html ↗
Geography	United States
size	segments: 2084320 · music hours: 2801.8 · music segments: 1011305 · annotated classes: 527 · seconds per segment: 10
scope	Large-scale weakly labeled audio event detection with a hierarchical ontology.
genres	many music genres and instruments
license	Metadata/code terms only; source-video rights and YouTube terms apply
steward	Google Sound Understanding group
creators	Jort Gemmeke; Daniel P. W. Ellis; Dylan Freedman; Aren Jansen; Wade Lawrence; R. Channing Moore; Manoj Plakal; Marvin Ritter
languages	multilingual/unspecified
modalities	YouTube video identifiers; time offsets; human labels; machine-extracted features
geographies	Global
access method	CSV metadata, ontology and feature downloads; audio must be resolved from YouTube.
intended uses	audio event detection; audio representation learning; music classification
related papers	Audio Set: An ontology and human-labeled dataset for audio events
annotation method	Human annotators verified sound-event presence using a subset of the ontology.
collection method	Candidate YouTube segments nominated from metadata and content search.
provenance claims	Official release distributes identifiers, timing and labels rather than a licensed audio archive.
current availability	Annotations and features available; source-video availability varies.
commercial use limits	No blanket commercial audio license is supplied.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Weak labels are clip-level.; Source URLs decay and YouTube availability changes.; Unbalanced training labels vary in quality.
redistribution restrictions	Do not treat linked YouTube audio as redistributable dataset content.