MarmAudio is a database of common marmoset vocalisations, recorded from an animal facility that houses ~20 marmosets in three cages. The dataset comprises more than 800,000 files of a few seconds each, amounting to 253 hours of data. These recordings capture the marmosets’ social vocalisations, encompassing their entire known vocal repertoire. The vocalisations were projected into a 16-dimensional auto-encoder latent space then visualised in 2D using UMAP — Uniform Manifold Approximation and Projection. Vocalisation types (e.g., trill, twitter) are indicated by colour.
Charly Lamothe, Manon Obliger-Debouche, Paul Best, Régis Trapeau, Sabrina Ravel, Thierry Artières, Ricard Marxer, and Pascal Belin.
A Large Annotated Dataset of Vocalizations by Common Marmosets
2025. Scientific Data 12 (1): 782. — @HAL