This is a beginner-level introduction to our two publications for disentangled music representations for monophonic and polyphonic music. This repository is originally made for the course Introduction to Computer Music at NYU Shanghai. It can also be served as a tutorial for other people who are interested to our projects and new to deep learning.
-
EC$^2$-VAE for monophonic pitch contour and rhythm disentanglement.
-
Poly-Dis for polyphonic chord and texture disentanglement.
We provide:
- The model architecture implemented in PyTorch. (The code is reformatted and all the code for training is removed.)
- (One-version of) the pre-trained model parameters. (Google drive links are provides inside
model_param
folders.) - Sample data and a tutorial jupyter notebook.
- Ruihan Yang et al., "Deep Music Analogy Via Latent Representation Disentanglement", ISMIR 2019
- https://arxiv.org/abs/1906.03626
- https://github.com/buggyyang/Deep-Music-Analogy-Demos
- Ziyu Wang et al., "Learning Interpretable Representation for Controllable Polyphonic Music Generation", ISMIR 2020.
- Ziyu Wang et al., "PIANOTREE VAE: Structured Representation Learning for Polyphonic Music", ISMIR 2020
- https://github.com/ZZWaang/polyphonic-chord-texture-disentanglement
- https://github.com/ZZWaang/PianoTree-VAE
- https://program.ismir2020.net/poster_5-05.html
- https://program.ismir2020.net/poster_3-06.html