Abstract:

The synthesis of high quality virtual acoustic sources over headphones has been shown to depend on individual differences between subjects: Sources synthesized using one's own head-related transfer functions provide more accurate localization cues than sources generated with HRTFs obtained from others. A method of perceptually optimizing a generic set of HRTFs to more closely approximate those of a given subject is discussed. In this paper, HRTFs are represented as a cascade of a common and a directional transfer function [Middlebrooks and Green (1990)]. The DTFs are modeled as low-order pole-zero systems, whose parameters are varied using an adaptive iterative method [Wakefield and Runkle (1992)] that incorporates subject preferences with a search algorithm. Unlike previous attempts where no assumptions were made about the subject's criteria for judging the quality of prospective solutions, psychophysically based constraints have been imposed on the parameter space to facilitate rapid convergence. To assess the feasibility of using subjective feedback in the optimization of DTFs, subjects were asked to ``move'' a binaural source from an initial position to a nearby desired position. Issues of convergence and repeatability will be discussed. [Work supported by the Office of Naval Research in conjunction with the Naval Submarine Medical Research Laboratory.]