Decomposition of Head-Related Transfer Functions into Multiple Damped and Delayed Sinusoidals

There are currently two options to achieve binaural sound spatialization using Head-Related Impulse Responses (HRIRs): measure every intended listener's HRIR or use generic HRIRs. However, measuring HRIRs requires expensive and specialized equipment, which removes its availability to the general public. In contrast, use of generic HRIRs results in higher localization errors. Another possibility that researchers, including our group, are pursuing is the customization of HRIRs. Our group is pursuing this by developing a structural model in which the physical measurements of a new intended listener could be used to synthesize his/her custom-fitted HRIRs, to achieve spatialization equivalent to measured HRIRs. However, this approach requires that HRIRs from multiple subjects be initially broken down in order to reveal the parameters of the corresponding structural models. This paper presents a new method for decomposing HRIRs and tests its performance on simulated examples and actual HRIRs.