HRIR Factorisation: A Regularised Approach

Abstract

HRIR factorisation allows the extraction of a common direction in- dependent component from a set of HRIRs leaving a set of rela- tively short direction dependent filters and provides both memory and computational savings in real time spatial audio applications. In this paper this technique is improved by introducing weighted regularisation to the iterative least squares process. This allows for a more robust algorithm, that is initial condition independent. Two forms of regularisation are introduced, one which allows the ITD to be maintained in the process and one which is suitable for minimum phase HRIRs. Results are shown which demonstrate the effective- ness of both variants of this technique.