Analysis and morphing of monaural cues from individual head-related transfer functions
Master Thesis of Braren, Hark
The core of every 3D audio presentation is the head related transfer function (HRTF). It is quite specific to each individual, due to varying expressions of head shape and outer ear (pinna) geometries among humans. Even though some applications work with general HRTFs (i.e. from a dummy head), it is absolutely necessary to use individual transfer functions in order to achieve a truly immersive experience with correct positioning of virtual sources in the 3D space. Measuring the HRTFs takes specialized equipment and a lot of time. The goal of this thesis is to develop a method that can generate individualized head related transfer functions from a set of head and pinna measurements. The proposed method uses principal component analysis on a database of known HRTFs. Multiple regression analysis is used to find a direction specific relation between the principal components and the pinna and head dimensions. These relations can then be used to calculate an individualized set of HRTFs for an unknown subject from their personal pinna and head dimensions. In this thesis several methods to improve the accuracy of the individualized HRTFs are presented and evaluated. Additionally, the problem of inaccuracies in the measurement of pinna dimensions is discussed and the foundation of an automated measurement process using scans or photographs of the ear is presented.