animal2vec and MeerKAT

A deep learning based framework
and a large-scale reference dataset for bioacoustics

Max Planck Institute of Animal Behavior

Department for the Ecology of Animal Societies

Communication and Collective Movement (CoCoMo) Group

Find the vocalizations

MeerKAT dataset

[1] Schäfer-Zimmermann, J. C., et al. (2024). Preprint at arXiv:2406.01253

Overall concept

Overall concept

Results

[1] Schäfer-Zimmermann, J. C., et al. (2024). Preprint at arXiv:2406.01253

Summary & Outlook

• We released the MeerKAT dataset and the animal2vec deep learning framework
• MeerKAT is the largest labeled dataset on non-human terrestrial mammals available
• animal2vec is a large transformer model, and a self-supervised training scheme tailored for sparse and unbalanced bioacoustic data that obtains best-in-class results


• We are currently building the next version of animal2vec
• We plan to build the largest possible model with all publicly available bioacoustic data
• We are currently rewriting the codebase to make the code more accessible

animal2vec and MeerKAT

A deep learning based framework
and a large-scale reference dataset for bioacoustics

Max Planck Institute of Animal Behavior

Department for the Ecology of Animal Societies

Communication and Collective Movement (CoCoMo) Group