Python library for downloading, loading & working with sound datasets. Check the API documentation and the contributing instructions.
For Music Information Retrieval (MIR) datasets please check mirdata.
This library provides tools for working with common sound datasets, including tools for:
- Downloading datasets to a common location and format
- Validating that the files for a dataset are all present
- Loading annotation files to a common format
- Parsing clip-level metadata for detailed evaluations
Here's soundata's list of currently supported datasets.
To install, simply run:
pip install soundata
import soundata
dataset = soundata.initialize('urbansound8k')
dataset.download() # download the dataset
dataset.validate() # validate that all the expected files are there
example_clip = dataset.choice_clip() # choose a random example clip
print(example_clip) # see the available data
See the documentation for more examples and the API reference.
We welcome and encourage contributions to this library, especially new dataset loaders. Please see contributing for guidelines. Feel free to open an issue if you have any doubt or your run into problems when working on the library.
The Soundata Zenodo repository is the preferred source for downloading the software releases.
If you use Soundata in your pipeline, please cite the version used with the corresponding DOI of the version release in Zenodo. For Soundata v1.0.1.:
If you refer to soundata's design principles, motivation etc., please cite the JOSS article:
@article{Fuentes2024,
title = {{Soundata: Reproducible use of audio datasets}},
author = {Fuentes, Magdalena and Plaja-Roglans, Genís and Cortès-Sebastià, Guillem and Khandelwal, Tanmay and Miron, Marius and Serra, Xavier and Bello, Juan Pablo and Salamon, Justin},
year = 2024,
month = jun,
journal = {Journal of Open Source Software},
volume = 9,
number = 98,
pages = 6634,
doi = {10.21105/joss.06634},
url = {https://joss.theoj.org/papers/10.21105/joss.06634}
}
When working with datasets, please include the reference of the dataset, which can be found in the respective dataset loader using cite()
.