Rendering five-channel audio on headsets

Rendering multi-channel audio using a headset can be accomplished by convolving the incoming signals with the head-related impulse responses representing the sound directions of the audio channels. Although the idea is simple, the implementation of this idea is computationally intensive due to the high sampling rate of the audio samples and the long sequences of the head-related impulse responses (HRIRs). In this paper, we propose to use a dynamic programming algorithm to optimally truncate the original HRIRs to shorter ones, and thus reduce the computational burden of the convolution to about one-fifth of that of direct implementation. The proposed approach is compared with the approach based on the CAPZ (common-acoustical-pole and zero) model. The experimental results show that the proposed approach has a better error performance with almost the same computation.