Back Soundata: Reproducible use of audio datasets

Soundata: Reproducible use of audio datasets

A new paper published in the Journal of OpenSource Software (JOSS) and presentation at the World Ecoacoustic Congress

09.07.2024

Imatge inicial

The paper “Soundata: Reproducible use of audio datasets” by Magdalena Fuentes, Genís Plaja-Roglans , Guillem Cortès-Sebastià, Tanmay Khandelwal, Marius Miron, Xavier Serra, Juan Pablo Bello, and Justin Salamon, has been published in the Journal of OpenSource Software (JOSS). The article presents a new major release of Soundata, an open-source Python library, developed with significant contributions from the MTG, designed for working with audio datasets in a standardized way.
 

Soundata aims to simplify audio research workflows by eliminating the need for custom dataset loaders. At the same time, it enhances reproducibility of research works. It provides tools to validate datasets against canonical versions, ensuring completeness and correctness. Researchers can quickly download and validate the datasets and access audio and annotations easily.

Key features of Soundata include:

  • Unified tools for downloading and validating datasets, ensuring they are complete and uncorrupted.
  • Loading annotation files into a common format for ease of use.
  • Compatibility with a wide range of audio datasets, including bioacoustics, environmental, urban, and spatial sounds.

Soundata is designed to be user-friendly and easy to contribute to, promoting reproducibility and standardization in the usage of sound datasets.

The current release supports 25 datasets including FSD50K and FSDnoisy18k, compiled by the MTG, and datasets from DCASE conference, in which the MTG has had a notable presence.

We invite you to explore Soundata and contribute to its development, helping to advance the field of audio research with open-source, standardized, and reproducible tools.

All the supported datasets are listed here: https://soundata.readthedocs.io/en/stable/source/quick_reference.html

In addition, Genís Plaja will be presenting Soundata at the 5th World Ecoacoustic Congress, which will take place at the Universidad Autónoma de Madrid from July 8th to 12th. The congress is a global platform for researchers, scientists, technicians, engineers, and practitioners to review and debate advances in ecoacoustics.

Article:

Magdalena Fuentes, Genís Plaja-Roglans , Guillem Cortès-Sebastià, Tanmay Khandelwal, Marius Miron, Xavier Serra, Juan Pablo Bello, and Justin Salamon (2024). Soundata: Reproducible use of audio datasets. Journal of Open Source Software, 9(98), 6634, https://doi.org/10.21105/joss.06634

We want to thank all the contributors to Soundata and Mirdata for all the great work done on this amazing project.

 

Multimedia

Categories:

SDG - Sustainable Development Goals:

Els ODS a la UPF

Contact