Time and Frequency Decomposition of Head-Related Impulse Responses for the Development of Customizable Spatial Audio Models

This paper introduces a new approach to the decomposition of measured Head-Related Impulse Responses (HRIRs) based on simultaneous analysis in the time and frequency domains. This approach is computationally less demanding and faster than previous systematic approaches proposed for this purpose. Currently, HRIRs are the most usual representation of Head-Related Transfer Functions (HRTFs), which are, in turn, the basis of many 3D sound spatialization systems used for PC gaming and virtual reality applications, among others. Many of these applications, however, utilize “generic” sets of HRTFs, which may provide only a sub-optimal spatialization experience. The improved HRIR decomposition method will facilitate our first step towards creating easily customizable HRTF representations, as the process of decomposition yields the sets of parameters that can instantiate a simple functional model to be equivalent to the HRTF represented by the corresponding HRIR. Furthermore, the decomposition breaks down the measured HRIR, into multiple delayed and scaled damped sinusoids, which have characteristics (frequency, decaying rate, initial amplitude and latency), that can be associated with anatomical characteristics of the outer ear of the listener. Key-Words: Customizable 3D Spatial Audio, Head-Related Impulse Response (HRIR), Head-Related Transfer Function (HRTF), Prony method, Steiglitz-McBride method.