NSynth Dataset

dataset·instrument note audio·active

Annotated four-second musical-note audio for timbre modeling, synthesis, and audio representation research.

Recorded facts

Official site	https://magenta.tensorflow.org/datasets/nsynth ↗
Geography	United States
size	notes: 305979 · instruments: 1006 · sample rate hz: 16000 · seconds per note: 4
scope	Single musical notes across instrument families, pitches, velocities and sonic qualities.
license	CC BY 4.0
steward	Google Magenta
creators	Jesse Engel; Cinjon Resnick; Adam Roberts; Sander Dieleman; Karen Simonyan; Mohammad Norouzi; Doug Eck
versions	full train/validation/test release
languages	not_applicable
modalities	16-bit PCM WAV; TFRecord; JSON metadata
geographies	Global
access method	Direct Google-hosted TFRecord and JSON/WAV archives.
intended uses	neural audio synthesis; timbre representation; audio classification
related papers	NSynth: Neural Audio Synthesis
annotation method	Instrument family, source, pitch, velocity and sonic-quality metadata supplied per note.
collection method	Notes generated from commercial sample-library instruments across pitches and velocities.
provenance claims	Official page identifies commercial sample libraries as the instrument source.
current availability	Download links available on official page.
commercial use limits	No dataset-level noncommercial restriction stated; attribution required.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Isolated four-second notes do not model full musical performance context.
redistribution restrictions	Redistribution permitted under CC BY 4.0 with attribution.