Joint sound scene and event dataset

A dataset of synthentic sound scenes created in scaper [1]using real-world recordings. Synthesized scenes are split into train and test such that no original recordings are in both splits. 

Dataset created for the task of performing both acoustic scene classification jointly with sound event detection. Time stamped annotations and jams files included. 

10 acoustic scene classes, 32 sound event classes.

filename syntax is [sceneLabel_recordingID_polyphonylevel]

Full credits / citation details are below, please contact Yogi at h.bear@qmul.ac.uk with any queries. 

Citation:

Helen L Bear, Ines Nolasco, and Emmanouil Benetos, Towards joint sound scene and polyphonic sound event detection. Interspeech 2019. Graz. 

References

1. Salamon, Justin, et al. "Scaper: A library for soundscape synthesis and augmentation." Applications of Signal Processing to Audio and Acoustics (WASPAA), 2017 IEEE Workshop on. IEEE, 2017.