Cadenza Lyric Intelligibility Prediction Dataset

dataset·lyric intelligibility·active·also: CLIP dataset

Popular-music signals with lyrics and listener-derived intelligibility scores for the ICASSP 2026 Cadenza Challenge.

Verified facts

Official site	https://doi.org/10.1016/j.dib.2026.112466 ↗
Geography	United Kingdom
doi	10.1016/j.dib.2026.112466
size	music signals: 11072
scope	Predicting sung-lyric intelligibility, including hearing-loss conditions.
genres	popular Western music
license	Creative Commons/open access with source-track conditions
steward	Cadenza project
creators	Cadenza challenge dataset authors
languages	English singing
modalities	popular music audio; ground-truth lyrics; listener intelligibility scores
geographies	Global
access method	Cadenza challenge dataset and open-access paper.
intended uses	lyric intelligibility prediction; music accessibility research
related papers	The Cadenza lyric intelligibility prediction dataset
annotation method	Human annotators transcribed lyrics and listening tests produced intelligibility scores.
collection method	Independent-artist tracks selected from FMA; non-derivative licenses and non-English singing excluded.
provenance claims	Paper states source tracks allow derivative use and were selected from FMA.
current availability	Challenge is complete; dataset documentation remains online.
commercial use limits	Depends on source FMA licenses and challenge data terms.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Restricted to English singing and popular Western music.; Listener scores depend on listening population and hearing profiles.
redistribution restrictions	Per-track FMA and challenge terms apply.