Human sound localisation cues and their relation to morphology

Binaural soundfield reproduction has the potential to create realistic threedimensional sound scenes using only a pair of normal headphones. Possible
applications for binaural audio abound in, for example, the music, mobile
communications and games industries. A problem exists, however, in that
the head-related transfer functions (HRTFs) which inform our spatial perception of sound are affected by variations in human morphology, particularly in the shape of the external ear. It has been observed that HRTFs
simply based on some kind of average head shape generally result in poor
elevation perception, weak externalisation and spectrally distorted sound
images. Hence, HRTFs are needed which accommodate these individual
differences. Direct acoustic measurement and acoustic simulations based
on morphological measurements are obvious means of obtaining individualised HRTFs, but both methods suffer from high cost and practical difficulties. The lack of a viable measurement method is currently hindering
the widespread adoption of binaural technologies. There have been many attempts to estimate individualised HTRFs effectively and cheaply using easily
obtainable morphological descriptors, but due to an inadequate understanding of the complex acoustic effects created in particular by the external ear,
success has been limited. The work presented in this thesis strengthens current understanding in several ways and provides a promising route towards
improved HRTF estimation. The way HRTFs vary as a function of direction is compared with localisation acuity to help pinpoint spectral features
which contribute to spatial perception. 50 subjects have been scanned using
magnetic resonance imaging to capture their head and pinna morphologies,
and HRTFs for the same group have been measured acoustically. To make
analysis of this extensive data tractable, and so reveal the mapping between
the morphological and acoustic domains, a parametric method for efficiently
describing head morphology has been developed. Finally, a novel technique,
referred to as morphoacoustic perturbation analysis (MPA), is described.
We demonstrate how MPA allows the morphological origin of a variety of
HRTF spectral features to be identified.