Efficient HRTF Computation using Adaptive Rectangular Decomposition

Accurate rendering of spatial audio over headphones requires the use of personalized head related transfer functions (HRTFs). These HRTFs are difficult to obtain due to the tedious and expensive measurement process requiring an anechoic chamber. An alternate approach uses accurate 3D meshes of human head and torso and numerical simulation techniques to compute personalized HRTFs. While these simulation techniques can compute accurate HRTFs, they require hours or days of computation on a desktop machine. We present an efficient technique to compute personalized HRTFs, combining a fast numerical solver, called adaptive rectangular decomposition, with the acoustic reciprocity principle and the Kirchhoff surface integral representation to reduce the overall computation. This technique requires only two numerical simulations and can compute the HRTF in 20 minutes on a desktop machine. We highlight the performance of our technique on the Fritz and KEMAR benchmarks and compare with measurements to test its accuracy.