This repository contains a PyTorch implementation of the paper Learning Multi-Level Representations for Hierarchical Music Structure Analysis presented at ISMIR 2022.
The code is based on the implementation of Conditional Similarity Networks and the overall format used in the MSAF package.
The detault setting for this repo is a CSN with fixed masks, an embedding dimension 128 and four notions of temporal distance (from the coarsest to the most refined). The baseline denoted as Flat embeddings can be obtained by setting the n_conditions parameter to 1.
The network can be trained with:
python exp.py --feat_id {feature type} --ds_path {path to the dataset}
The dataset format should follow:
dataset/
├── audio # audio files (.mp3, .wav, .aiff)
├── features # feature files (.npy)
└── references # references files (.jams)
conda env create -f environment.yml
@inproceedings{buisson2022learning,
title={Learning Multi-Level Representations for Hierarchical Music Structure Analysis},
author={Buisson, Morgan and Mcfee, Brian and Essid, Slim and Crayencour, H{\'e}l{\`e}ne C},
booktitle={International Society for Music Information Retrieval (ISMIR)},
year={2022}
}