Google Audioset is a large dataset of manually annotated audio events, designed to provide a general practical evaluation task for audio event detection, as well as a starting point for a comprehensive vocabulary of sound events.

It contains 632 audio classification samples and 2,084,320 human-labeled 10-second sound clips extracted from YouTube videos. The dataset covers human and animal sounds, musical instruments and genres, and common everyday environmental sounds.

This dataset was released by Google Machine Perception Research in 2017.

Related paper: Audio Set: An ontology and human-labeled dataset for audio events