Passive Acoustic Monitoring

Passive acoustic monitoring uses long-running recordings to study wildlife, ecosystems, and underwater sound without needing observers at every site. Researchers use spectrogram-based CNNs and transformers, event localization and tracking, and self-supervised or weakly supervised learning to find calls in large archives with limited annotations. Benchmarks increasingly test cross-site and cross-sensor transfer, while active learning and curated datasets aim to make labeling more efficient. A central challenge is ensuring models generalize to noisy, unfamiliar environments while meeting the power, storage, and reliability limits of field sensors.