The Morphologically Informed Perceptual Enhancement of Spatial Audio

With the proliferation of immersive technologies in virtual reality (VR), broadcasting
and home entertainment, estimating an individual’s head-related transfer
functions (HRTFs) conveniently and with a satisfactory perceptual performance
is of great topical interest. To facilitate this, a deep understanding of how the
head and pinnae form the acoustic cues used to perceive spatial sound is required.
This thesis presents the refinement of a powerful research tool, morphoacoustic
perturbation analysis (MPA), for advancing knowledge in this field.
To simplify analysis a novel method is presented for smoothing HRTFs based
on an equivalent rectangular bandwidth (ERB) criterion. The approach is first
evaluated using an auditory localisation model and these results are validated
by means of listening tests. It is shown that ERB smoothing achieves perceptual
transparency using fewer parameters than a similar constant-bandwidth approach.
Furthermore, it simplifies the structure of the HRTFs, since additional
perceptually irrelevant features are discarded.
It has been well established that the boundary element method can satisfactorily
generate HRTFs based on a three-dimensional (3D) mesh of a listener’s head
and pinnae shape. A proof-of-principle for MPA, upon which this thesis builds,
has successfully inverted this process, making it possible to identify the morphological
regions of the head and pinnae responsible for creating an HRTF feature.
However, first-generation MPA suffered from significant weaknesses including low
mesh resolution, restricted frequency range, and topological issues created by the
mesh slicing approach used. In this work these issues are addressed through the
use of optimised spherical mesh mapping and spherical harmonic deformations.
The theory, implementation and validation of the new method is described to the
point where the creation of a full MPA database capable of probing in depth the
relationship between human morphology and HRTFs is now possible.