Perceptual Musical Features for Interpretable Audio Tagging

Overview

This GitHub repository contains the code necessary to reproduce the results presented in the paper titled Perceptual Musical Features for Interpretable Audio Tagging.

Contents

• classification.py: This script is designed for running the classification experiments.
• interpretations.py: Use this script to obtain explanations for the classifications.

Data Repositories

• gtzan: Contains the data used for experiments related to the GTZAN dataset.
• jamendo: Includes data and explanations specific to the Jamendo dataset.

Additional Files

• features_descriptions.pdf: This file provides a comprehensive description of the features used as input for our models.
• environment.yml: Refer to this file for the exact environment specifications used in conducting the experiments.

Usage

To reproduce the experiments and results from the paper, follow the instructions in the provided scripts and utilize the relevant data from the gtzan and jamendo repositories.

Citation

If you find the paper's results useful, please consider citing the paper.

@misc{lyberatos2023perceptual,
      title={Perceptual Musical Features for Interpretable Audio Tagging}, 
      author={Vassilis Lyberatos and Spyridon Kantarelis and Edmund Dervakos and Giorgos Stamou},
      year={2023},
      eprint={2312.11234},
      archivePrefix={arXiv},
      primaryClass={cs.SD}
}