Enhanced modeling of head-related impulse responses towards the development of customizable sound spatialization

Audio spatialization is a rapidly growing field in acoustics and audio signal processing. Measurements and models of head-related impulse responses (HRIRs) of human subjects or KEMAR dummy heads are the primary source of information for research on spatial audio. The information contained in HRIRs measured at various azimuth and elevation angles is sufficient for synthesizing realistic three dimensional spatial audio for headphone or loudspeaker listening. Currently, however, designers must decide between the need for specialized equipment to measure the individual HRIRs of all potential users or the low fidelity achieved with the use of HRIRs obtained from a dummy head. To overcome this problem, research is underway to develop a method to create customized HRIRs. Using signal processing tools, such as Prony's signal modeling method, an appropriate set of time delays and a resonant frequency was used to approximate the measured HRIRs, with the goal of establishing a general HRIR model that could be instantiated from anatomical measurements of the prospective listener. During more recent experimentation, the Prony method was substituted by the Steiglitz-McBride iteration method and a noticeable improvement was achieved. This paper reports on our assessment of the statistically significant improvement in the approximation when the Steiglitz-McBride iteration method is used instead of the Prony method, for HRIR decomposition.