New method for synthesizing personalized head-related transfer function

Personalized HRTFs(Head-related Transfer Functions) can be synthesized through the corresponding anthropometric features by the linear or nonlinear mapping from anthropometric features to HRTFs in the database. Some methods for synthesizing personalized HRTFs had been proposed. However, they were hard to be applied in practical circumstances to some extent. In this paper, we use principle component analysis to obtain the weight matrix and eigenmatrix of HRTFs, then use canonical correlation analysis between anthropometric features and HRTFs to wipe the redundant information of the features, and model the mapping from anthropometric features to eigenmatrix with GRNN(Generalized Regression Neural Network). Based on these work, the personalized HRTFs can be synthesized. The experimental results show the effectiveness of this new method.