Song Describer is an open-source data annotation tool for collecting and evaluating textual descriptions of music recordings.
It was developed by a team of researchers from Queen Mary University of London and Universitat Pompeu Fabra Barcelona with the aim of crowdsourcing the first public research dataset of music captions.
The resulting dataset is available at Zenodo.
The online app can be found at song-describer.streamlit.app.
The Song Describer annotation process is structured into three main stages: user onboarding, annotation, and evaluation. The tool also provides additional features such as a leaderboard to display the users' progress and the ability to identify returning users via a unique user ID.
Instructions on how to run the app locally can be found in the annotation tool README.
If you use this tool in your research, please cite the following paper:
@inproceedings{manco2022song,
title={Song Describer: a Platform for Collecting Textual Descriptions of Music Recordings},
author={Manco, Ilaria and Weck, Benno and Tovstogan, Philip and Won, Minz and Bogdanov, Dmitry},
booktitle={Extended Abstracts for the Late-Breaking Demo Session of the 23rd Int. Society for Music Information Retrieval Conf},
year={2022},
address={Bengaluru, India}
}This software is licensed under the MIT License.