Skip to Main content Skip to Navigation
Journal articles

Polyphonic training set synthesis improves self-supervised urban sound classification

Abstract : Machine listening systems for environmental acoustic monitoring face a shortage of expert annotations to be used as training data. To circumvent this issue, the emerging paradigm of self-supervised learning proposes to pre-train audio classifiers on a task whose ground truth is trivially available. Alternatively, training set synthesis consists in annotating a small corpus of acoustic events of interest which are then automatically mixed at random to form a larger corpus of polyphonic scenes. Prior studies have considered these two paradigms in isolation, but rarely ever in conjunction. Furthermore, the impact of data curation in training set synthesis remains unclear. To fill this gap in research, this article proposes a two-stage approach. In the self-supervised stage, we formulate a pretext task (Audio2Vec skip-gram inpainting) on unlabeled spectrograms from an acoustic sensor network. Then, in the supervised stage, we formulate a downstream task of multilabel urban sound classification on synthetic scenes. We find that training set synthesis benefits more to overall performance than self-supervised learning. Interestingly, the geographical origin of the acoustic events in training set synthesis appears to have a decisive impact.
Complete list of metadata
Contributor : Mathieu Lagrange Connect in order to contact the contributor
Submitted on : Friday, July 23, 2021 - 12:32:29 PM
Last modification on : Friday, August 5, 2022 - 2:54:51 PM
Long-term archiving on: : Sunday, October 24, 2021 - 6:36:09 PM


gontier2021jasa (2).pdf
Publisher files allowed on an open archive


  • HAL Id : hal-03262863, version 1


Félix Gontier, Vincent Lostanlen, Nicolas Fortin, Mathieu Lagrange, Catherine Lavandier, et al.. Polyphonic training set synthesis improves self-supervised urban sound classification. Journal of the Acoustical Society of America, Acoustical Society of America, 2021. ⟨hal-03262863⟩



Record views


Files downloads