MusicNet

dataset·classical note annotation·active

Classical recordings aligned to score-derived note, instrument and metrical-position labels.

Recorded facts

Official site	https://homes.cs.washington.edu/~thickstn/musicnet.html ↗
Geography	United States
size	recordings: 330 · instruments: 11 · annotated labels: over 1000000
scope	Supervised note and instrument recognition on classical recordings.
genres	Western classical chamber music
license	Mixed public-domain and Creative Commons works
steward	University of Washington research authors
creators	John Thickstun; Zaid Harchaoui; Sham Kakade
languages	not_applicable
modalities	WAV; CSV note labels; metadata
geographies	Global
access method	Historical project distribution and mirrors.
intended uses	music transcription; instrument recognition; benchmarking
related papers	Learning Features of Music from Scratch
annotation method	Dynamic-time-warping alignment of scores to recordings, followed by musician verification.
collection method	Freely licensed classical recordings paired with machine-readable scores.
provenance claims	Project descriptions state music and labels are public domain or Creative Commons.
current availability	Primary historical host is unstable; community mirrors exist.
commercial use limits	Depends on each recording's license.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Estimated annotation error rate is about 4%.; Corpus is concentrated in Western classical music.
redistribution restrictions	Per-recording attribution and license conditions apply.