Reduced order modeling of head related impulse responses for virtual acoustic displays

This study investigated the use of reduced order head related impulse response (HRIR) models to improve the computational efficiency in acoustic virtual displays. State space models of varying order were generated from zero-elevation HRIRs using a singular value decomposition technique. A source identification experiment was conducted under anechoic conditions in which three subjects were required to localize sounds in the front horizontal plane. The sounds were either (1) real sources (emitted by individual loudspeakers in a semi-circular array), (2) virtual sources generated from the original HRIRs, or (3) virtual sources generated using reduced order state space models. All virtual sources were created by simultaneous activation of two loudspeakers at +/- 30 degrees using a virtual source imaging technique based on either the measured or modeled HRIRs. The errors in the perceived direction of the virtual sources generated from the reduced order models were compared to errors in localization using the original HRIRs. The results demonstrate that a very significant reduction in model size can be achieved without significantly affecting the fidelity of the virtual display of horizontally placed sources.