On the use of partial least square regression for spatial fitting of head-related transfer functions

The Head-Related Transfer Functions (HRTFs) are spatially continuous while the traditional acoustic measurements are not available to meet this demand. In this paper, we present a spatial fitting method for HRTFs on the basis of a few discrete measurements. We first make the HRTF matrix and the corresponding direction matrix normalized. Then we build an algorithm model between the normalized direction and HRTF matrices with partial least square regression, and determine the number of principle component by cross validation. After that, the regression was implemented on each of their principal components. Finally, the regression equation was recovered according to the inverse process of normalization. We also verified the effectiveness of this method through calculating the similarity between original data and predicted data. The experimental results show that our method is available in spatial fitting of HRTFs.