Personalization of head-related transfer function: An isogeometric solution

Understanding relations between the anatomical features of human and the frequency-dependent characteristics of head-related transfer function is an essential task towards the personalization of HRTF. Two major difficulties are encountered in parametric studies for establishing such relations. The first one is how to fast predict HRTF. Extensive simulations are involved in the parametric study while the acoustic boundary element method - the current workhorse - usually takes over ten hours to complete one single run. The second one concerns how to geometrically morph a baseline model to build a set of target models with parametric variations. Besides substantial re-meshing work, it is hard for the element-based method to capture tiny variations in the geometries of head and ear. Such variations, however, could cause huge deviations in the HRTF. To address these problems, an isogeometric solution is proposed in this paper for the personalization of HRTF. Instead of the element-based discretization, the numerical model is directly set up based on the Nonuniform Rational B-splines (NURBS) surfaces. The parametric nature of NURBS surface enables a fast, flexible and natural way for the desired geometric modification, eliminating any additional meshing or re-meshing effort. Moreover, the Kirchhoff-Helmholtz integral equation is solved directly using the isogeometric BEM. The fast multipole algorithm is steadily employed based on the multi-patch structure of NURBS surface, reducing the CPU time of one single HRTF simulation to an amazing level of less than one hour. The proposed isogeometric approach provides a unified and robust tool for the numerical analysis of HRTF, paving the way for the personalization of HRTF in the hearing-aid and earphone industries.