This repository contains information about the Broad Sound Taxonomy, a simple general-purpose taxonomy for the classification of a broad range of sounds. This taxonomy organizes sounds into a two-level hierarchical structure comprising 29 classes:
- 5 broad top-level categories
- 24 semantically finer second-level categories
A description and examples of each category can be found in the taxonomy_description.
The taxonomy was developed based on principles and design criteria described in our [upcoming journal paper].
Supplementary material related to the taxonmy design is available in the taxonomy_creation/ folder.
Materials include information such as the design criteria scores, the tag_mappings, and the query_mappings.
To support and refine the taxonomy, we conducted a user experiment. The sound files used in the experiment are hosted on Zenodo. (Link to be inserted)
For more details, please refer to the full paper.
The Broad Sound Taxonomy is well-suited for diverse tasks such as organizing and retrieving heterogeneous sound collections.
One of its main real-world deployments is on Freesound, a large collaborative database of audio clips contributed by users around the world. On the Freesound search page sidebar, you can browse or filter sounds by BST categories by clicking a category (second-level categories will appear after you click a top-level category).
To be continued.
- BSD10k (Broad Sound Dataset 10k). It contains over 10,000 audio clips sourced from Freesound, each labeled with BST categories. BSD10k serves as a benchmark for heterogeneous sound classification and provides a semantically diverse testbed for evaluating machine learning models.
To be continued.
If you use this material, please cite our paper (citation coming soon).