A Wide Dataset of Ear Shapes and Pinna-Related Transfer Functions Generated by Random Ear Drawings
This work addresses the problem of HRTF individualization for binaural synthesis researchers by providing a freely available, larger dataset, though it is incremental as it builds on existing methods and data.
The researchers tackled the limited size of head-related transfer function (HRTF) databases by generating a synthetic dataset of 1000 ear shapes and matching pinna-related transfer functions (PRTFs), using random ear drawings and fast-multipole boundary element method (FM-BEM) computations.
Head-related transfer functions (HRTFs) individualization is a key matter in binaural synthesis. However, currently available databases are limited in size compared to the high dimensionality of the data. Hereby, we present the process of generating a synthetic dataset of 1000 ear shapes and matching sets of pinna-related transfer functions (PRTFs), named WiDESPREaD (wide dataset of ear shapes and pinna-related transfer functions obtained by random ear drawings) and made freely available to other researchers. Contributions in this article are three-fold. First, from a proprietary dataset of 119 three-dimensional left-ear scans, we build a matching dataset of PRTFs by performing fast-multipole boundary element method (FM-BEM) calculations. Second, we investigate the underlying geometry of each type of high-dimensional data using principal component analysis (PCA). We find that this linear machine learning technique performs better at modeling and reducing data dimensionality on ear shapes than on matching PRTF sets. Third, based on these findings, we devise a method to generate an arbitrarily large synthetic database of PRTF sets that relies on the random drawing of ear shapes and subsequent FM-BEM computations.