Lakh MIDI Dataset

dataset·symbolic music·active

Large deduplicated MIDI collection with matched and aligned subsets linked to the Million Song Dataset.

Recorded facts

Official site	https://colinraffel.com/projects/lmd/ ↗
Geography	United States
size	matched files: 45129 · deduplicated midi files: 176581
scope	Internet-collected MIDI files for symbolic MIR and audio-MIDI alignment.
genres	mixed
license	CC BY 4.0 dataset release
steward	Colin Raffel
creators	Colin Raffel
versions	0.1; LMD-full; LMD-matched; LMD-aligned
languages	not_applicable
modalities	MIDI; JSON; HDF5 metadata; alignment scores
geographies	Global
access method	Direct mirrors for full, matched, aligned and metadata subsets.
intended uses	symbolic MIR; music generation; audio-MIDI matching
related papers	Learning-Based Methods for Comparing Sequences, with Applications to Audio-to-MIDI Alignment and Matching
annotation method	Dynamic-time-warping matching and alignment to Million Song Dataset previews; MIDI metadata retained.
collection method	Scraped from publicly accessible internet sources and deduplicated by MD5.
provenance claims	Individual authorship cannot be reliably attributed because MIDI copyright events are inconsistent.
current availability	Official index and mirrors available.
commercial use limits	Dataset page states CC BY 4.0, but underlying composition and transcription rights are not resolved.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	A few thousand files may be corrupt.; Harmonically similar songs may be matched incorrectly.; Underlying MIDI authorship and copyright are often unknown.