MAESTRO

dataset·aligned piano audio midi·active

Paired piano audio and high-precision MIDI from International Piano-e-Competition performances.

Verified facts

Official site	https://magenta.tensorflow.org/datasets/maestro ↗
Geography	United States
size	version: 3.0.0 · performances: 1276 · duration hours: 198.7 · notes millions: 7.04 · uncompressed gb: 120.2
scope	Virtuosic classical piano performances captured on Yamaha Disklaviers.
genres	classical piano
license	CC BY-NC-SA 4.0
steward	Google LLC
creators	Curtis Hawthorne; Andriy Stasyuk; Adam Roberts; Ian Simon; Cheng-Zhi Anna Huang; Sander Dieleman; Erich Elsen; Jesse Engel; Douglas Eck
versions	1.0.0; 2.0.0; 3.0.0
languages	not_applicable
modalities	WAV; MIDI; CSV; JSON
geographies	International competition; recordings released by US steward
access method	Direct Google-hosted full and MIDI-only archives.
intended uses	piano transcription; music generation; performance modeling
related papers	Enabling Factorized Piano Music Modeling and Generation with the MAESTRO Dataset
annotation method	Audio and MIDI aligned to about 3 ms; composer, title and year metadata supplied.
collection method	Competition performances recorded on Yamaha Disklaviers over ten years.
provenance claims	Official page identifies International Piano-e-Competition as the raw-data partner.
current availability	Version 3.0.0 full, MIDI-only and metadata downloads are available.
commercial use limits	Commercial use prohibited by the dataset license.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Repertoire is mostly Western classical music from the 17th to early 20th centuries.