Assessing Spherical Harmonics Interpolation of Time-Aligned Head-Related Transfer Functions

High-quality spatial audio reproduction over headphones requires head-related transfer functions (HRTFs) with high spatial resolution. However, acquiring datasets with a large number of (individual) HRTFs is not always possible, and using large datasets can be problematic for real-time applications with limited resources. Consequently, interpolation methods for sparsely sampled HRTFs are of great interest, with spherical harmonics (SH) interpolation becoming increasingly popular. However, the SH representation of sparse HRTFs suffers from spatial aliasing and order truncation errors. To mitigate this, preprocessing methods have been introduced that time-align the sparse HRTFs before SH interpolation. This reduces the effective SH order and thus the number of HRTFs required for SH interpolation. In this paper, we present a physical evaluation of four state-of-the-art preprocessing methods, which showed very similar performance of the methods with notable differences only at low SH orders and contralateral HRTFs. We also performed a listening experiment with one selected method to determine the minimum required SH order required for perceptually transparent interpolation. For the selected method, a sparse HRTF set of order N ≈ 7 is sufficient for interpolating a frontal source presenting speech or percussion. Higher orders are, however, required for a lateral source and noise.