DALI

dataset·aligned lyrics notes·active

Full-song lyric, vocal-note and audio alignments at note, word, line and paragraph granularity.

Recorded facts

Official site	https://transactions.ismir.net/articles/10.5334/tismir.30 ↗
Geography	France
size	v1 songs: 5358 · v2 songs: 7756 · v2 music hours: 488.1 · v2 vocal hours: 247.2
scope	Singing voice, lyrics alignment, transcription and melody research.
genres	primarily polyphonic popular music
license	unknown/mixed
steward	DALI authors
creators	Gabriel Meseguer-Brocal; Alice Cohen-Hadria; Geoffroy Peeters
versions	v1; v2
languages	many languages
modalities	audio links; lyrics; phonemes; vocal melody notes; hierarchical time alignments; metadata
geographies	Global
access method	Dataset tools and metadata; audio resolution depends on external sources.
intended uses	lyrics alignment; lyrics transcription; melody extraction; structural analysis
related papers	Creating DALI, a Large Dataset of Synchronized Audio, Lyrics, and Notes
annotation method	Machine learning refines and validates hierarchical timing alignments.
collection method	Karaoke-user annotations matched to web audio candidates through teacher-student iteration.
provenance claims	Audio is linked from the web rather than distributed as a uniformly licensed archive.
current availability	Metadata/tooling available; audio-link survival varies.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Annotations originate with non-expert karaoke users and are machine-refined.; Audio links can disappear.; Language distribution is uneven.
related models tools benchmarks	MIREX Lyrics Alignment