Computational and Perceptual Characterization of Auditory Attention

auditory;attention;computational modeling;psychoacoustics;eeg;Electrical Engineering

Kaya, Emine MerveElhilali, Mounya ;

Johns Hopkins University

关键词: auditory;    attention;    computational modeling;    psychoacoustics;    eeg;    Electrical Engineering;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/58626/KAYA-DISSERTATION-2017.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

Humans are remarkably capable at making sense of a busy acoustic environment in real-time, despite the constant cacophony of sounds reaching our ears. Attention is a key component of the system that parses sensory input, allocating limited neural resources to elements with highest informational value to drive cognition and behavior. The focus of this thesis is the perceptual, neural, and computational characterization of auditory attention.Pioneering studies exploring human attention to natural scenes came from the visual domain, spawning a number of hypotheses on how attention operates among the visual pathway, as well as a considerable amount of computational work that attempt to model human perception. Comparatively, our understanding of auditory attention is yet very elementary, particularly pertaining to attention automatically drawn to salient sounds in the environment, such as a loud explosion. In this work, we explore how human perception is affected by the saliency of sound, characterized across a variety of acoustic features, such as pitch, loudness, and timbre. Insight from psychoacoustical data is complemented with neural measures of attention recorded directly from the brain using electroencephalography (EEG). A computational model of attention is presented, tracking the statistical regularities of incoming sound among a high-dimensional feature space to build predictions of future feature values. The model determines salient time points that will attract attention by comparing its predictions to the observed sound features. The high degree of agreement between the model and human experimental data suggests predictive coding as a potential mechanism of attention in the auditory pathway.We investigate different modes of volitional attention to natural acoustic scenes with a ;;cocktail-party;; simulation. We argue that the auditory system can direct attention in at least three unique ways (globally, based on features, and based on objects) and that perception can be altered depending on how attention is deployed. Further, we illustrate how the saliency of sound affects the various modes of attention.The results of this work improve our understanding of auditory attention, highlighting the temporally evolving nature of sound as a significant distinction between audition and vision, with a focus on using natural scenes that engage the full capability of human attention.