WikiMuTe: A web-sourced dataset of semantic descriptions for music audio

This repository contains auxiliary code for the dataset described in the paper

WikiMuTe: A web-sourced dataset of semantic descriptions for music audio. Benno Weck, Holger Kirchhoff, Peter Grosche, Xavier Serra. to appear in MultiMedia Modeling, 2024.

Code

We provide a short Python script that can be used to download the audio files from the URLs provided in the dataset. Please update the user agent string in the script before running it. Moreover, the script shows how the dataset can be loaded to get the data in a convenient format.

Dataset

The dataset contains rich text description for music audio files collected from Wikipedia articles. The audio files are available for download through the URLs provided in the dataset.

We provide three variants of the dataset in the data folder. All are described in the paper.

1. all.csv contains all the data we collected, without any filtering.
2. filtered_sf.csv contains the data obtained using the self-filtering method.
3. filtered_mc.csv contains the data obtained using the MusicCaps dataset method.

Downloading the dataset

The dataset is available to download from Zenodo. A download script is provided in download.bash.

bash download.bash

The audio files have to be downloaded separately using the provided Python script.

python3 -m venv venv  # create a virtual environment
source venv/bin/activate  # activate the virtual environment
pip install -r requirements.txt  # install dependencies
python3 wikimute.py  # run the script

File structure

Each CSV file contains the following columns:

• file: the name of the audio file
• pageid: the ID of the Wikipedia article where the text was collected from
• aspects: the short-form (tag) description texts collected from the Wikipedia articles
• sentences: the long-form (caption) description texts collected from the Wikipedia articles
• audio_url: the URL of the audio file
• url: the URL of the Wikipedia article where the text was collected from

Citation

If you use this dataset in your research, please cite the following paper:

@inproceedings{wikimute,
    title = {WikiMuTe: A web-sourced dataset of semantic descriptions for music audio},
    author = {Weck, Benno and Kirchhoff, Holger and Grosche, Peter and Serra, Xavier},
    booktitle = {to appear in MultiMedia Modeling.},
    year = {2024},
    publisher="Springer International Publishing",
    address="Cham",
}

Note: The final paper is not yet published. The citation will be updated once the paper is published.

License

This repository is released under the MIT License. Please see the LICENSE file for more details.

The data is available under the Creative Commons Attribution-ShareAlike 3.0 Unported (CC BY-SA 3.0) license. Each entry in the dataset contains a URL linking to the article, where the text data was collected from.