Simulation of the head-related transfer functions using cloud computing

Abstract

Due to the complexity of measurements for obtaining individual head-related transfer functions (HRTFs), numerical simulations offer an attractive alternative for generating large HRTF data bases. In this study, HRTFs are simulated using a fast multipole boundary element method (BEM). The BEM is well suited for the HRTF simulations. Namely, only the surface of the model geometry is discretized which simplifies the pre-processing compared to other full-wave simulation methods (such as finite element and finite difference methods). The BEM is formulated in frequency domain and the model is solved separately for each frequency. Since a large number of frequencies is needed in wide-band HRTF simulations, the BEM simulation greatly benefits from distributed (or parallel) computing. That is, a single computing unit takes care of a single frequency. In this study, a distributed BEM using cloud computing is introduced. Simulations are computed in a public cloud (Amazon EC2) using a realistic head and torso geometry (3D laser scanned geometry of Bruel & Kjaer HATS 4128 mannequin). The frequency range of the simulations is from 20 to 20000 Hz. The feasibility of cloud computing for simulating HRTFs is examined and first analysis results for the simulated HRTFs are shown.

Pinna-related transfer functions and lossless wave equation using finite-difference methods: Verification and asymptotic solution

“…Rui et al [337] qualitatively compared BEM simulation of a KEMAR mesh with the CIPIC measurements [5]. Huttunen et al [299] and Kärkkäinen et al [308] employed Amazon's Elastic Compute cloud EC2 to obtain FMP BEM simulations of HRTFs. Tang et al [349] compared BEM simulations of a structured-light scanned dummy head (BHead210) with measurements at r = 1 m up to 12 kHz -some directions showed strong measurement-simulation inconsistencies.…”