Music Variational AutoEncoder (MusicVAE) in PyTorch

This repository contains a PyTorch implementation of the Music Variational AutoEncoder (MusicVAE) model, as described in the paper "A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music" by Roberts et al.

MusicVAE is a deep learning model that learns a hierarchical representation of music and can generate new musical sequences. It combines a variational autoencoder (VAE) with a hierarchical decoder to capture long-term structure in music.

Features

• Implements the MusicVAE model architecture in PyTorch
• Trains the model on MIDI data to learn a latent representation of music
• Generates new musical sequences by sampling from the learned latent space
• Provides utility functions for processing MIDI files and converting between MIDI and tensor representations

Dataset

The MIDI data used in this repository is sourced from arman-aminian/lofi-generator. It consists of a collection of MIDI files that can be used to train the MusicVAE model.

Usage

1. Clone the repository: git clone https://github.com/yourusername/Learning-Music-Variational-AutoEncoder.git

2. Install the required dependencies: pip install -r requirements.txt

3. Prepare the MIDI data:

• Make a midi_songs folder and put your MIDI files into it.

4. Run the Jupyter notebook main.ipynb to train the MusicVAE model and generate new musical sequences.

Code Structure

• midi_utils.py: Contains utility functions for processing MIDI files and converting between MIDI and tensor representations.
• model.py: Defines the MusicVAE model architecture using PyTorch.
• loss.py: Implements the loss function used for training the MusicVAE model.
• main.ipynb: Jupyter notebook that demonstrates loading MIDI data, training the MusicVAE model, and generating new musical sequences.

Credits

• The MusicVAE model implementation is based on the paper "A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music" by Roberts et al.
• The MIDI utility functions and MIDI data are sourced from arman-aminian/lofi-generator.

Citation

If you use this code or the MusicVAE model in your research, please cite the following paper:
@inproceedings{roberts2018hierarchical, title={A hierarchical latent vector model for learning long-term structure in music}, author={Roberts, Adam and Engel, Jesse and Raffel, Colin and Hawthorne, Curtis and Eck, Douglas}, booktitle={International conference on machine learning}, pages={4364--4373}, year={2018}, organization={PMLR} }

License

This project is licensed under the MIT License.