Slakh2100

dataset·synthetic multitrack music·active

Synthetic multitrack audio and aligned MIDI rendered from Lakh MIDI files for separation and transcription.

Recorded facts

Official site	https://www.slakh.com/ ↗
Geography	United States
github repository	ethman/slakh-generation
size	tracks: 2100 · patches: 187 · duration hours: 145 · instrument classes: 34
scope	Automatically mixed multitrack songs synthesized from selected Lakh MIDI files.
genres	mixed MIDI-derived
license	CC BY 4.0
steward	Mitsubishi Electric Research Laboratories and Northwestern Interactive Audio Lab
creators	Ethan Manilow; Gordon Wichern; Prem Seetharaman; Jonathan Le Roux
versions	Slakh2100; Flakh2100
languages	not_applicable
modalities	audio mixtures; audio stems; aligned MIDI; YAML metadata
geographies	Global
access method	Zenodo dataset and GitHub generation code.
intended uses	music source separation; multi-instrument transcription
related papers	Cutting Music Source Separation Some Slakh
annotation method	MIDI track mapping supplies aligned instrument and note annotations.
collection method	Randomly selected Lakh MIDI subset synthesized with sample-based virtual instruments.
provenance claims	Derived from Lakh MIDI v0.1; official page discloses duplicated MIDI files caused by a bug.
current availability	Official download and support-code links active.
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	Synthetic renderings may not represent real recording conditions.; Some MIDI files were duplicated due to a documented bug.