Analysis of Individual Differences in Head-Related Transfer Functions by Spectral Distortion

Abstract

The inter-subject variability of head-related transfer function (HRTF) needs to be good considered for elevation perception of the sound image. This work investigates the variability of HRTF based on spectral distortion. As the representative of human subjects, CIPIC HRTF database was used for calculating spectral distortions between subjects. In the frequency range from 0 Hz to 14 kHz, a small spectral distortion was obtained at high elevation angles. At front and rear source positions, spectral distortion is larger than the case of high elevation source positions. Some 1/3 octave bands, the similar results were obtained under 11.3 kHz band when the same analysis is performed on 1/3 octave band frequency scale. To give a high-fidelity virtual impression without HRTF customization, Multi-channel speakers are needed at front and rear source positions because of large variations of HRTF. Moreover, when dimension reduction technique (for example, principal component analysis) is applied for HRTF database, the analysis should be focused on the region of high spectral distortion.