A new inverse processing approach to the modelling of head-related transfer functions for audio spatialization

Abstract

Currently, achieving high-fidelity sound spatialization requires the prospective user to undergo lengthy measurements in an anechoic chamber using highly specialized equipment. This, in turn, has increased the cost and reduced the availability of high-fidelity spatialization to the general public. Attempts to generalize 3D audio have been made using the measurement of a KEMAR dummy head or creating a database containing a sample of the public. Unfortunately, this leads to increased front/back reversals and localization errors in the median plane. Customizable head-related impulse responses (HRIRs) would reduce the errors caused by general HRIRs and remove the limitation of the measured HRIRs. This article reports an initial stage in the development of customizable HRIRs. The ultimate goal is to develop a compact functional model that is equivalent to empirically measured HRIRs but requires a smaller number of parameters that could be obtained from the anatomical characteristics of the intended listener. In order to arrive at such a model, the HRIRs must be decomposed into multiple-scaled and delayed-damped sinusoids, which would reveal the parameters that the compact model needs to have an impulse response similar to the measured HRIR. Previously this type of HRIR decomposition has been accomplished through an exhaustive search of the model parameters. A new method that approaches the decomposition simultaneously in the frequency (Z) and time domains is reported here.

Keywords:

• head-related transfer functions (HRTFs)
• head-related impulse responses (HRIRs)
• singal decomposition
• customizable spatial audio
• structural pinna model
• 3D audio

Acknowledgement

This work was sponsored by NSF grants IIS-0308155, CNS-0520811, HRD-0317692 and CNS-0426125.