Detection and Classification of Acoustic Scenes and Events
Published March 26, 2021 | Version v1
Dataset Open

DESED_real

Creators

  • 1. Universite de Lorraine, CNRS, Inria, Loria, F-54000 Nancy, France
  • 2. Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, United States
  • 3. Adobe Research, San Francisco, CA, United States

Description

If you are in this page, you have to know that the content of this page is also in github associated.

Link to the associated github repository: https://github.com/turpaultn/Desed

Link to the paper: https://hal.inria.fr/hal-02160855

Domestic Environment Sound Event Detection (DESED) dataset.

Description

This dataset is the real part of the DESED dataset. It is a subpart of Audioset.

There is the material to:

  • Download the metadata of the subset of Audioset used in DCASE 2019 task 4.

Files:

  1. DESED_real_metadata.tar.gz : Annotations of subset of Audioset.
  2. audioset_strong.tsv: Metadata of an Audioset subpart strongly annotated (Shawn et al.)1. This TSV file is a modified version of "Audioset strong" curated for our needs:
    1. Labels converted from the ontology
    2. Filename changed (Y added and timestamp going from 30000 to 30.000_40.000)
    3. Only keeping the files available for us (some Audioset files are unavailable)

To download the associated soundfile, please visit: https://github.com/turpaultn/DESED .

 

1Shawn Hershey, Daniel P W Ellis, Eduardo Fonseca, Aren Jansen, Caroline Liu, R Channing Moore, and Manoj Plakal. The benefit of temporally-strong labels in audio event classification. In 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2021.

Files

Files (1.1 MB)

Name Size Download all
md5:88445011163455a0b49224793b677610
266.4 kB Download
md5:c4bd8827b91dce20ee438b14f0eeb5cb
812.0 kB Download

Additional details

Related works

Is supplement to
Conference paper: https://hal.inria.fr/hal-02160855v2 (URL)