HEAR 2021

dataset·general audio representation·active

A NeurIPS challenge and evaluation suite for general-purpose audio representations across speech, sound and music tasks.

Recorded facts

Official site	https://hearbenchmark.com/hear-2021-neurips-challenge.html ↗
Geography	United States
size	tasks: 19
scope	Universal audio representation evaluation with fixed downstream tasks and metrics.
genres	music tasks plus speech and environmental sound
license	code and source-dataset-specific
steward	HEAR benchmark organizers
creators	HEAR Benchmark authors
languages	task-dependent
modalities	audio; labels; embeddings; evaluation code
geographies	Global
access method	Open challenge code; source datasets obtained separately.
intended uses	audio representation benchmarking
consent claims	Inherited from source datasets.
related papers	HEAR: Holistic Evaluation of Audio Representations
annotation method	Reuses source labels under standardized embedding evaluation.
collection method	Curated nineteen downstream tasks from public research datasets.
provenance claims	Official challenge documents task datasets and evaluation procedure.
current availability	Historical challenge materials and code remain available.
commercial use limits	Varies by task dataset.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Task selection can favor representations trained on overlapping data.