Fine-Tuning for Bird Sound Classification: An Empirical Study
摘要
In biodiversity research scientists now routinely acquire audio recordings of vocalizing bird species and are then faced with the task of identifying the species audible in these recordings. Here, we analyze the accuracy (precision, recall and \(F_1\) score) of several deep networks, in conjunction with pre-training and data augmentation techniques, for classifying audio recordings of twelve bird species under multiple data scarcity settings.