This repository contains the source code and dataset for training a deep learning-based model to perform inpainting on musical scores, i.e., to connect two musical excerpts in a musically meaningful manner (see figures below for schematics).
The approach followed relies on training a RNN-based architecture to learn to traverse the latent space of a VAE-based deep generative model.
Install anaconda or miniconda by following the instruction here.
Create a new conda environment using the enviroment.yml file located in the root folder of this repository. The instructions for the same can be found here.
To install, either download / clone this repository. Open a new terminal, cd into the root folder of this repository and run the following command
pip install -e .
The contents of this repository are as follows:
DatasetManger: Module for handling data.AnticipationRNN: Module implementing model, trainer and tester classes for the AnticipationRNN model.MeasureVAE: Module implementing model, trainer and tester classes for the MeasureVAE model.LatentRNN: Module implementing model, trainer and tester classes for the LatentRNN model.utils: Module with model and training utility classes and methods- other scripts to train / test the models
This research work is published as a conference paper at ISMIR, 2019. Arxiv Preprint available here.
Ashis Pati, Alexander Lerch, Gaëtan Hadjeres. "Learning to Traverse Latent Spaces for Musical Score Inpaintning", Proc. of the 20th International Society for Music Information Retrieval Conference (ISMIR), Delft, The Netherlands, 2019.
@inproceedings{pati2019inpainting,
title={Learning to Traverse Latent Spaces for Musical Score Inpaintning},
author={Pati, Ashis and Lerch, Alexander and Hadjeres, Gaëtan},
booktitle={20th International Society for Music Information Retrieval Conference (ISMIR)},
year={2019},
address={Delft, The Netherlands}
}
Please cite the above publication if you are using the code/data in this repository in any manner.
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
