Evaluating the Consistency Between Numerical Calculation and Acoustic Measurement of Head-related Transfer Functions

In the validation of numerically calculated Head-Related Transfer Functions (HRTFs), acoustic measurements are often considered the ground truth. This study demonstrates that discrepancies between the numerical and measured HRTFs do not necessarily indicate inaccuracies in the numerical calculations. Factors contributing to these discrepancies were systematically examined using KEMAR by calculating HRTFs with two mainstream numerical methods: Mesh2HRTF and the Finite Difference Time Domain (FDTD) approaches. For the FDTD method, the optimal virtual microphone placement was found to be within 3 mm of the ear canal mesh entrance. In numerical calculations, the spatial correlation of spectra computed with shoulder-inclusive meshes and full-torso meshes exceeded 0.9 in the low-frequency (<4 kHz) region of the median plane. This suggests that the shoulders within the torso are the primary contributors to elevation-direction HRTF variations. Furthermore, consistencies among different numerical methods, measurement datasets, and simulated versus measured HRTFs were evaluated. The results show that, for the dummy head, the similarity between the numerical and measured HRTFs was no less than the similarity observed among separate measurement datasets. Numerical calculations demonstrated greater stability than acoustic measurements, particularly in human-related scenarios where maintaining measurement consistency is challenging. These findings suggest that accurate head modeling combined with numerical simulation offers a promising approach for obtaining high-accuracy individualized HRTFs.