Efficient algorithm and localization experiment on spherical microphone array recording and binaural rendering

Abstract

Spherical microphone array recording and binaural rendering (SMABR) is a novel spatial sound technique, which records spatial information of sound field, transfers to dynamic binaural signals, and then renders via headphone. To improve the computational efficiency of SMABR, the present work proposes a PCA-based (principal component analysis) algorithm for dynamic binaural synthesis from the beamforming outputs of spherical microphone array, in which only 33 pairs of sharable filters are required. Incorporated the proposed algorithm, a PC-based SMABR system consisting of a 64-channel spherical microphone array, an electromagnetic head-tracker, and a headphone is established. A virtual source localization experiment is carried out to evaluate the system and algorithm. Results indicate that the system yields reasonable localization performance within a target virtual source region near the horizontal plane. However, larger localization error is also observed for high elevation above 30 degrees or low elevation below -30 degrees. Localization error at high or low elevation is caused by the spatial aliasing error of spherical microphone array at high frequency, which spoils the spectral cue for elevation localization. Therefore, further improvement on spherical microphone array recording is needed. [Supported by the National Natural Science Foundation of China, Grant No.11174087.]