Animal2vec and MeerKAT: A Self-Supervised Transformer for Bioacoustics

https://arxiv.org/abs/2406.01253

This study uses AI-driven classifiers to decode macaque vocalizations. By analyzing the modulation power spectrum of calls, the researchers identified key acoustic features, like fundamental frequency and FM sweep direction, that help classify individual callers. AI-based regularized classifiers showed that a combination of spectrotemporal components is essential for reliable identification, advancing the use of AI in understanding primate communication systems.

Arxiv: (Bioacoustics Research)