HRIR customization using common factor decomposition and joint support vector regression
Zhixin Wang, Cheung-Fat Chan · European Signal Processing Conference · 2013
A two-stage approach for the customization of head-related impulse response (HRIR) for individual subject is proposed. In the first stage, a two-dimension common factor decomposition (2D-CFD) algorithm is applied to extract a subject-dependent impulse response (SDIR) from full HRIR dataset of a subject. The SDIR is then represented as the weighted sum of some principal components using independent component analysis to further reduce the dimensionality of HRIR dataset. In the second stage, joint support vector regression is applied to construct a nonlinear model for mapping the weightings of a target subject from its anthropometric parameters where correlations between different weightings are also exploited. The proposed approach achieves a more accurate and consistent result as compared to the original support vector regression algorithm.