A Magnitude-Based Parametric Model Predicting the Audibility of HRTF Variation

Abstract

This work proposes a parametric model for just noticeable differences of unilateral differences in head-related transfer functions (HRTFs). For seven generic magnitude-based distance metrics, common trends in their response to inter-individual and intra-individual HRTF differences are analyzed, identifying metric subgroups with pseudo-orthogonal behavior. On the basis of three representative metrics, a three-alternative forced-choice experiment is conducted, and the acquired discrimination probabilities are set in relation with distance metrics via different modeling approaches. A linear model, with coefficients based on principal component analysis and three distance metrics as input, yields the best performance, compared to a simple multi-linear regression approach or to principal component analysis–based models of higher complexity.