Computational Bioacoustics of Deer Mice: Identifying and Comparing Vocal Syllables Across Species and Infant Development

Abstract

Vocal behaviors are diverse across mammals, but the mechanisms underlying this diversity are poorly understood. Unlike traditional models of vocal behavior such as the laboratory mouse (Mus musculus), deer mice (genus Peromyscus) exhibit variation in vocal behaviors that has evolved in natural environments. Furthermore, this rodent group is abundant across North American habitats and inter-fertile among distinct species, providing a unique opportunity to study the evolution of vocalization. Yet their vocal ontogeny lacks quantitative description, data necessary to investigate the biological causes of this variation. In pursuit of describing the deer mouse vocal repertoire and its changes across development, I recorded vocalizations from mice belonging to 9 deer mouse species and subspecies (vocal development in 3 of which has not been previously described) at 1, 3, 5, 7, 9, 11, and 13 days of age. I then built a computational bioacoustics pipeline to detect, quantify, and compare vocal syllables across development and between species. These data reveal that infant deer mouse vocalizations cluster into two types—which I refer to as cries and whistles—that are markedly different in their spectrographic and acoustic features. Over the course of development, pups switch from almost exclusively crying to almost exclusively whistling. Furthermore, these two types are differentially stereotyped in their spectrographic representations by species across development. These results lay the foundation for genetic mapping and tissue staining, among other tools, to be used to identify biological mechanisms underlying these cross-specific differences. This work will ultimately contribute to our understanding of the evolution of mammalian behavior.