lpd-full contains 174,154 multitrack pianorolls derived from the Lakh MIDI Dataset (LMD).
lpd-matched contains 115,160 multitrack pianorolls derived from the matched version of LMD. These files are matched to entries in the Million Song Dataset (MSD). To make use of the metadata provided by MSD, we refer users to the demo page of LMD.
matched_ids.txt provides a list of all file IDs and the matched MSD IDs in the matched subset.
lpd-cleansed contains 21,425 multitrack pianorolls collected from lpd-matched with the following rules. Note that lpd-cleansed contains songs from ALL genres, which is different from the description on the paper.
cleansed_ids.txt provides a list of all file IDs and the matched MSD IDs in the cleansed subset.
* The matching confidence scores come with the LMD, which is the confidence of whether the MIDI file match any entry in the MSD.
midi_info.json contains useful information lost during the conversion from LMD to LPD. It was used to create lpd-cleansed.
first_beat_time: the actual timing of the first beatnum_time_signature_change: the number of time signature change eventsconstant_time_signature: the only time signature used (None if it
changes within a song)constant_tempo: the only tempo (in bpm) used (None if it changes within
a song)midi_info_v2.json
has the same values for first_beat_time, num_time_signature_change and
constant_time_signature. However, constant_tempo is now a boolean value that
indicates whether the tempo is constant throughout the song. There is an
additional key tempo that stores the initial tempo value (in bpm).
In LPD-5, the tracks are merged into five common categories: Drums, Piano, Guitar, Bass and Strings according to the program numbers provided in the MIDI files.
Note that instruments out of the five categories are considered as part of the strings except those in the Percussive, Sound effects and Synth Effects families (see here).
In LPD-17, the tracks are merged into drums and sixteen instrument families according to the program numbers provided in the MIDI files and the specification of General MIDI (see here). The seventeen tracks are Drums, Piano, Chromatic Percussion, Organ, Guitar, Bass, Strings, Ensemble, Brass, Reed, Pipe, Synth Lead, Synth Pad, Synth Effects, Ethnic, Percussive and Sound Effects.