Contrastive Learning Enables Marine Soundscape Discovery

Acs et al. (published March 6, 2026) present a scalable, unsupervised contrastive learning framework for passive acoustic monitoring that organizes underwater recordings into acoustically coherent clusters across multiple Caribbean spawning aggregation sites. The method, using multi-positive contrastive learning with a teacher network and acoustically informed augmentations, outperforms cepstral features, VAEs, and supervised pipelines, enabling label-efficient cross-site comparisons and reduced manual annotation.