This research investigates the use of self-supervised learning (SSL) for bird sound classification, particularly in few-shot learning scenarios. The researchers demonstrate that SSL can acquire meaningful representations of bird sounds from audio recordings without the need for annotations. The experiments show that these learned representations can generalize to new bird species. The study also reveals that selecting windows with high bird activation for SSL using a pretrained audio neural network significantly enhances the quality of the learned representations. This work is particularly relevant in bioacoustics, where biologists collect extensive sound datasets from natural environments.
Publication date: 28 Dec 2023
Project Page: https://github.com/ilyassmoummad/ssl4birdsounds
Paper: https://arxiv.org/pdf/2312.15824
