Modeling head-related transfer functions with spherical wavelets

Abstract The head-related transfer function (HRTF) describes the sound transmission characteristics from a sound source to a listener’s ears. Recently, spherical harmonic decomposition has been extensively used for modeling the HRTF spatial patterns. Despite its advantage of approximating the coarse structure of HRTF spatial variations with modeling up to a low order, there are still some limitations since spherical harmonics take significant values in all directions. First, rapidly changing HRTF spatial variations in some local regions may require modeling up to a rather high order; this is not wise in terms of the modeling efficiency. Second, the expansion coefficients of the spherical harmonics describe the spatial frequency of the target dataset in all directions, and thus have difficulties in revealing the direction dependent HRTF characteristics. In this study, a method for locally modeling HRTF spatial patterns is proposed based on spherical wavelets, which take significant values only over a local region on the sphere. Results of numerical experiments show that our proposed method yields smaller approximation errors than the conventional method when representing HRTFs inside the local regions under evaluation. Furthermore, the expansion coefficients in the proposed method could well correspond to the HRTF local features on the sphere, which makes it a useful tool for the analysis and visualization of HRTF spatial patterns.