• Setup
• Experiments
• Utility Scripts
• Acknowledgement

Create the environment.

conda env create -f environment.yml

Activate the environment.

conda activate mtmt

Run setup script.

python scripts/setup.py

• Make sure to place the X, Y and XY MIDI files in data/X/X/, data/Y/Y/ and data/XY/XY/ respectively. The X, Y, XY files for a sample should have the same filename.

• Run scripts/compute_info_flow.py

python compute_info_flow.py OUTPUT_FILE_PATH

OUTPUT_FILE_PATH is the file path where the results will be stored as a .pkl file. The path should include the name of the file. You can save them in results/.

Example:

python compute_info_flow.py results/example.pkl

• Open notebooks/visualize.ipynb to visualize results. The notebook has code to visualize the experiments used in the paper. The results/ directory has sample results for these experiments, so you don't have to run the compute_info_flow.py script with data.

Three utility scripts have been provided in the scripts/ directory:

1. change_instrument.py: Script to change instrument of a track in a midi file.
2. combine_midi.py: Script to combine 2 midi tracks into a single midi file.
3. split_midi.py: Script to split a midi file into 2 separate midi files, each containing a single track.

Usage instructions for these scripts are documented inside the files.

This repository uses the official implementation of "Multitrack Music Transformer" (ICASSP 2023).

Multitrack Music Transformer
Hao-Wen Dong, Ke Chen, Shlomo Dubnov, Julian McAuley and Taylor Berg-Kirkpatrick
Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023
[homepage] [paper] [code] [reviews]