Latent and Adversarial Data Augmentation for Sound Event Detection and Classification - Laboratoire Traitement et Communication de l'Information Access content directly
Conference Papers Year : 2022

Latent and Adversarial Data Augmentation for Sound Event Detection and Classification

Abstract

Invariance-based learning is a promising approach in deep learning. Among other benefits, it can mitigate the lack of diversity of available datasets and increase the interpretability of trained models. To this end, practitioners often use a consistency cost penalizing the sensitivity of a model to a set of carefully selected data augmentations. However, there is no consensus about how these augmentations should be selected. In this paper, we study the behavior of several augmentation strategies. We consider the task of sound event detection and classification for our experiments. In particular, we show that transformations operating on the internal layers of a deep neural network are beneficial for this task.
Fichier principal
Vignette du fichier
dcase.pdf (178.96 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03782827 , version 1 (21-09-2022)

Identifiers

  • HAL Id : hal-03782827 , version 1

Cite

David Perera, Slim Essid, Gaël Richard. Latent and Adversarial Data Augmentation for Sound Event Detection and Classification. International workshop on Detection and Classiffication of Acoustic Scenes and Events (DCASE), Nov 2022, Nancy, France. ⟨hal-03782827⟩
188 View
110 Download

Share

Gmail Facebook Twitter LinkedIn More