Prediction Model for Perceived Elevation of Ecologically Valid Sound Sources Intended for a Virtual Auditory Display

Abstract

Virtual auditory display (VAD) systems rely upon binaural technology to render sound sources at controlled directions in virtual acoustic spaces. The accuracy and precision with which human listeners can localise those sound sources, particularly in terms of perceived source elevation, depends upon spectral variation in the incident sound that is due to its interaction with head related transfer functions (HRTFs). The spectral processing developed in the current thesis has been optimised to reduce listener uncertainty regarding the perceived elevation of virtual sound sources, and to improve the overall spatial perception. A subset of individually measured HRTFs that supported the highest localisation accuracy was identified via preliminary listening sessions and used in the formation of a single ‘Collective’ HRTFs dataset that could be deployed for the entire group of listeners in a customised fashion. The customisation employed individually determined frequency scaling that was applied to the selected HRTFs before deploying the Collective HRTFs dataset, which could be readjusted for each individual through a calibration procedure that was based upon the individual’s localisation judgments. An evaluation of this customised HRTF dataset in a spatial auditory display of ecologically valid sound sources demonstrated improvement in localisation performance, in comparison with both the accuracy and precision of results obtained using individually measured HRTFs. Furthermore, the results informed the development of an adaptive processing of the proximal sound sources at runtime that showed good potential for improving localisation performance in a manner that adapts to listener responses. Based upon a runtime analysis of input sound source spectral variation, the adaptive processing was designed to improve accuracy and reduce uncertainty in apparent source elevation angle for the listener, and thus improve overall localisation performance.