Dissertations, Theses, and Capstone Projects

Date of Degree

6-2023

Document Type

Dissertation

Degree Name

Ph.D.

Program

Psychology

Advisor

Mark E. Hauber

Advisor

Sarah M. N. Woolley

Committee Members

Peter A. Serrano

David C. Lahti

Ofer Tchernichovski

Subject Categories

Animal Studies | Applied Statistics | Behavioral Neurobiology | Behavior and Ethology | Comparative Psychology | Data Science | Evolution | Multivariate Analysis | Ornithology

Keywords

Machine learning, statistical classification, social communication and recognition, phylogenetic signal, bioacoustics, embryonic neural activation

Abstract

Acoustic communication is a process that involves auditory perception and signal processing. Discrimination and recognition further require cognitive processes and supporting mechanisms in order to successfully identify and appropriately respond to signal senders. Although acoustic communication is common across birds, classical research has largely disregarded the perceptual abilities of perinatal altricial taxa. Chapter 1 reviews the literature of perinatal acoustic stimulation in birds, highlighting the disproportionate focus on precocial birds (e.g., chickens, ducks, quails). The long-held belief that altricial birds were incapable of acoustic perception in ovo was only recently overturned, as researchers began to find behavioral and physiological evidence for the effects of prenatal acoustic stimulation in songbirds.

Chapter 2 provides the first published evidence of the effect of acoustic stimulation on gene expression in the auditory forebrain of perinatal estrildid finches. Importantly, immediate early gene (ZENK) expression was significantly higher in embryos and nestlings exposed to conspecific zebra finch songs than those exposed to silence. Although no other significant differences were found, our results suggest a trend with ZENK expression being highest in response to conspecific songs, followed by more closely related heterospecific songs (Bengalese finch), more distantly related heterospecific songs (pin-tailed whydah), and finally silence. These results may have implications for early-life species recognition strategies in songbirds, specifically that innate template-matching of vocalizations may be used in perinatal birds before song learning can shape recognition processes, and this template-matching requires species-specific acoustic features for adequate activation.

Chapter 3 presents a continuation to the exploration of the effects of prenatal acoustic stimulation in estrildids. Here, we aimed to identify a possible epigenetic mechanism by which embryos may encode information from acoustic cues for long-term effects throughout ontogeny. Using the tissue from our previous study, we found that decreased ZENK expression predicts greater genome-wide methylation. Methylation levels were significantly lower in embryos exposed to conspecific songs than those exposed to more distantly related heterospecific songs. Although no other significant differences were found, our results suggest a trend with methylation increasing as we moved from groups exposed to conspecific songs, followed by more closely related heterospecific songs, silence, and finally more distantly related heterospecific songs. These results have implications for the importance of analyzing the effects of both acoustic and phylogenetic relationships when exploring processes and mechanisms of recognition.

Chapter 4 is a comparative analysis of the vocalizations of seven estrildid finch species. Syllables of male song from various family lines in each species are segmented and quantified in terms of 21 acoustic measures. In order to characterize the acoustic relationships between species, song syllables were projected onto multidimensional acoustic feature space and varimax rotation following principal component analysis suggests that we can represent a majority (72%) of the variance in species song in terms of three acoustic components: spectral, spectral shape, and spectrotemporal features. Support vector machine algorithms defined the species boundaries in the three-dimensional acoustic space and allowed us to quantify the volume and space occupied by each species, as well as the overlap between species. Volume overlap analyses suggest that species cluster along phylogenetic relationships in spectral features. Clustering in spectral shape features occurs along classical qualitative descriptions of song (i.e., tonal vs. broadband), and no discernible species clustering occurs in spectrotemporal features. Ensemble tree procedures were used to identify the acoustic features that best discriminate species song and optimize classification of syllables. We found that fundamental frequencies, mean frequency, duration, and spectral flatness can be used to classify species song with high accuracy (93%). Finally, we used the known phylogeny of our estrildid species in order to test for phylogenetic signal in their acoustic features. The results suggested that acoustic relationships in spectral features negatively correlated with the evolutionary distances of these species as predicted under Brownian motion. Our finding implied that spectral features might be constrained to evolutionary change through polygenic or pleiotropic trait changes that occur in tandem with phylogenetic changes in this family. Spectral shape and spectrotemporal features, on the other hand, did not express phylogenetic signal, suggesting that these traits might be more labile to change through alternative selection processes such as cultural transmission, song learning, and sexual selection.

This dissertation is presented as a synthesis of the work completed under the overarching topic of acoustic recognition. Chapters 1-3 focus on reviewing and advancing our understanding of acoustic recognition in perinatal birds through biological assays. Chapter 4 focuses on presenting a statistical procedure for recognition using the acoustic features of vocalizations. It is the aim of this thesis to advocate for more research into both the biological and statistical models that will help advance our understanding of early ontogenetic abilities, vocal processing and preference, and even artificial intelligence algorithms for human speech recognition.

Share

COinS