A cognitive approach for binaural models using a top-down feedback structure

Abstract

Current models to explain binaural hearing generally focus on bottom-up processes of the auditory periphery and subcortical brain functions to simulate sound localization and other binaural tasks. While these models have been very successful in explaining a number of psychoacoustic phenomena, their architecture is not suitable to simulate experiments that involve cognition. The project presented here seeks to close the gap between functional binaural models and research in applied robotics. A software architecture that was originally designed to simulate the process of music improvisation using a combination of Computational Auditory Scene Analysis, machine learning and logic-based reasoning, the Creative Artificially-Intuitive and Reasoning Agent CAIRA was extended to simulate a number of basic binaural phenomena including sound localization of multiple-sources, resolving front/back confusions trough strategic head movements, and adapting inhibitory parameters to the presented signals to evoke localization dominance. © 2013 Acoustical Society of America.