Labels

We derived the labels for the Lakh Pianoroll Dataset (LPD) from three different sources: the Last.fm Dataset, the Million Song Dataset (MSD) Benchmarks and the Tagtraum genre annotations. Note that these labels are derived based on the mapping between the Lakh MIDI Dataset (LMD) and the MSD, which may contain incorrect pairs (see here).

Please refer to the original sources for the license information.

Last.fm Dataset

• id_lists_lastfm.tar.gz: we derived the ID lists for most common labels in the Last.fm Dataset (see here).

Million Song Dataset Benchmarks

• id_lists_amg.tar.gz: we derived the ID lists for all the labels in the MSD AllMusic Top Genre Dataset (TopMAGD) provided in the Million Song Dataset Benchmarks (see here).
• masd_labels.txt: we also derived the labels from the MSD AllMusic Style Dataset (MASD) provided in the Million Song Dataset Benchmarks. Each entry has at most one label.
• masd_labels_cleansed.txt: this is a cleansed version of masd_labels.txt, where less common labels are discarded.

Tagtraum genre annotations

• id_lists_tagtraum.tar.gz: we derived the ID lists for all the labels in the Tagtraum genre annotations (see here).