Dataset Card for MUSDB-ALT

This dataset contains long-form lyric transcripts following the Jam-ALT guidelines for the test set of the dataset MUSDB18, with line-level timings.

Dataset Details

The dataset was constructed manually, based on the MUSDB18 lyrics extension as a starting point. The lyrics extension contains transcripts of the 45 English language songs out of the 50 in the MUSDB18 test set. We annotated 39 of those 45 songs, excluding 6 for the following reasons:

• Signe Jakobsen - What Have You Done To Me : Three overlapping vocal lines that could not be separated into lead and backing vocals
• PR - Happy Daze : Vocal content primarily from highly processed vocal samples
• PR - Oh No : Vocal content primarily from highly processed vocal samples
• Skelpolu - Resurrection : Vocal content primarily from highly processed vocal samples
• Timboz - Pony : Lyrics unintelligble due to screamed enunciation style
• Triviul feat The Fiend - Widows : Three overlapping vocal lines that could not be separated into lead and backing vocals

Dataset Description

Paper: The dataset was introduced in the paper Exploiting Music Source Separation for Automatic Lyrics Transcription with Whisper" published at the Workshop Artificial Intelligence For Music at ICME 2025

• Funding: This work was supported by InnovateUK [Grant Number 10102804]
• License: https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en

Citation

BibTeX:

@inproceedings{syed-2025-mss-alt,
  author       = {Jaza Syed and
                  Ivan Meresman-Higgs and
                  Ond{\v{r}}ej C{\'{\i}}fka and
                  Mark Sandler},
  title        = {Exploiting Music Source Separation for Automatic Lyrics Transcription with {Whisper}},
  booktitle    = {2025 {IEEE} International Conference on Multimedia and Expo Workshops (ICMEW)},
  publisher    = {IEEE},
  year         = {2025},
  note         = {In press}
}