Few-Shot Bioacoustic Event Detection: A Technical Report

The task of few-shot classification involves adapting a classifier to recognize new classes not encountered during training, with only a limited number of examples for each of these classes [1]. While humans are capable of accurately performing one-shot classification (i.e. recognizing new objects with just one example), few-shot classification remains a difficult task for even the most advanced machine learning algorithms [1]. Our technical report details our approach to addressing Task 5 of the DCASE2023 challenge, which involves detecting bioacoustic events using few-shot learning techniques.