Contribution
Geometric Analysis of Automatic Registration of a Parametric Pinna Model for the calculation of Head-Related Transfer Functions
* Presenting author
Abstract:
Parametric pinna models (PPMs) can be used to describe the geometry of a listener's ear. For this study we will focus on a PPM developed at the Acoustics Research Institute. The mesh underlying our PPM is uniformly sampled and free of artifacts that are usually present in meshes obtained via optical scanning methods, making the mesh viable for the numerical calculation of individualized HRTFs. Yet, one of the still unsolved challenges is the automated registration of the PPM parameters to optimally match an individual pinna geometry. Here, we introduce an approach based on a neural network estimating the PPM parameters. We evaluate our approach by comparing meshes synthesized from the estimated PPM parameters to the corresponding meshes obtained from optical scans. The results were analyzed in the geometric domain focusing on pinna regions relevant for HRTF individualization. We show that an estimation of the PPM parameters via a neural network results in meshes with small geometric errors, which can be used for the calculation of individualized HRTFs.