Multi-dimensional analysis of HRTF using tensor singular value decomposition

Abstract

Researches on virtual three-dimensional sound have been implemented actively to boost a sense of reality to the listener for various multimedia contents. Head-related transfer function (HRTF), an acoustic transfer function between pressure of sound source and listener's ear drum, plays an important role in virtual sound technology. HRTFs of a random subject should be measured with different angles and distances of sound sources to produce virtual sound at any desired position for the listener. However, it is hard to measure all points of three-dimensional sound sources, so many HRTF personalization or customization techniques based on HRTF database have been developed. Previous researches mainly focused on median or horizontal plane only which is two dimensional analysis. In this paper, we analyze HRTF in multi-dimensional point of view using tensor singular value decomposition method because this method can handle many dimensions. Time (or frequency), azimuth, elevation, and subject dimension were adopted as controlling axes. CIPIC database was used and analyzed results were compared with ordinary analyzing method such as principal component analysis.