Features extraction in spatial hearing based on nonlinear manifold learning algorithm

Abstract

In spatial hearing, head related transfer function (HRTF) play an important role. However, when modeling HRTF, the issue how to store mass data of HRTF or reduce computational complexity often be confronted, real-time is not accomplished effectively. In order to resolve the issue, we proposed a scheme. High dimensionality mapped into low dimensionality using nonlinear manifold learning and, data reduced dimensionality then are classified into several representative HRTF through unsupervised cluster algorithm. Features on sound directional information are extracted. Others HRTF can be reconstructed through modified interpolation using representative HRTF. In this paper, we provided simulation results. The results show that the scheme is effective to reducing data and degrading complexity and the performance of nonlinear manifold learning is better than PCA's.