The ACME dataset - Altered and Corrupted Midi Excerpts

Version 1.0 details

Generated using code from MDTK version 1.0: https://github.com/JamesOwers/midi_degradation_toolkit/releases/tag/v1.0

Count of degradations:

• none: 2501
• pitch_shift: 2502
• time_shift: 2502
• onset_shift: 2502
• offset_shift: 2501
• remove_note: 2502
• add_note: 2501
• split_note: 2502
• join_notes: 2501

You will find the generated data within the zipped directory acme with subdirectories:

• clean - contains the extracted clean excerpts
• altered - contains the excerpts altered by the degradations described in metadata.csv

metadata.csv describes:

• (the id number for) the type of degradation used for the alteration
• the path for the altered and clean files
• which split (train, valid, test) the file should be used in
• in which corpus and on what line the file is located

degradation_ids.csv is a mapping of degradation name to the id number used in metadata.csv

The {train,valid,test}_cmd_corpus.csv are command-based (note_on, note_off, shift) versions of the acme data more convenient for our provided pytorch Dataset classes avaialble in the Midi Degradation ToolKit package.

Similarly, {train,valid,test}_pr_corpus.csv are piano-roll-based versions of the acme data more convenient for our provided pytorch Dataset classes avaialble in the Midi Degradation ToolKit package.

Reproducibility

To reproduce this dataset again, install the Midi Degradation ToolKit package version 1.1 and run ./make_dataset --seed 208724510