A reduced order model of head-related impulse responses based on independent spatial feature extraction

A reduced order model for large numbers of head-related impulse responses (HRIRs) is proposed for real-time three-dimensional (3D) sound rendering. Independent spatial features are firstly extracted from measured HRIRs using independent component analysis (ICA). These spatial feature vectors are not only mutually statistical independent but independent from all the measured azimuths. Therefore filtering sound sources with numerous HRIRs is transformed into filtering them using the extracted lower-dimensional feature vectors. Furthermore balanced model truncation (BMT) method in a state space is adopted to reduce the order of each independent feature vector. Simulation results demonstrate that our proposed algorithm not only acquires better approximated accuracy but has significantly lower computational complexity.