Bioacoustics

Bioacoustics studies animal sounds and soundscapes to identify species, detect calls, track behavior, and investigate communication and ecosystem change. Common approaches include spectrogram-based CNNs and transformers, pretrained audio representations, self-supervised and weakly supervised learning, and methods for localizing calls or separating them from noise. Researchers use these tools with passive acoustic monitoring, annotated datasets, active learning, and compact models for field sensors, while also probing what learned representations reveal about vocal structure. A central challenge is reliable transfer across species, sites, sensors, and recording conditions, especially when labels are scarce and performance on curated benchmarks may not reflect field use.