(Nature Research, 2025-10-03) van Toor, Astrid; Qazi, Nadeem; Paladini, Stefania
Accurate age estimation is essential for advancing interspecies communication but remains a challenge across non-human species. This study presents the first dataset of domestic feline vocalisations specifically designed for age prediction and introduces a novel deep learning pipeline for this purpose. By applying transfer learning with models like VGGish, YAMNet, and Perch, we demonstrate the potential for automated age classification, with VGGish achieving the best results. Our findings hold significant potential for applications in veterinary care and wildlife conservation, building on existing research and pushing forward the boundaries of automated age classification within digital bioacoustics. Future work could explore improving model generalisability and robustness, potentially expanding its application across species.
