ABSTRACT

(Email me if you want a full PDF copy - Look in bio section to get my email)

This thesis first presents a novel object-oriented scheme which
provides for extensive description of time-varying 3D audio scenes
using XML. The scheme, named XML3DAUDIO, provides a new format for
encoding and describing 3D audio scenes in an object oriented
manner. Its creation was motivated by the fact that other 3D audio
scene description formats are either too simplistic (VRML) and
lacking in realism, or are too complex (MPEG-4 Advanced AudioBIFS)
and, as a result, have not yet been fully implemented in available
decoders and scene authoring tools. This thesis shows that the
scene graph model, used by VRML and MPEG-4 AudioBIFS, leads to
complex and inefficient 3D audio scene descriptions. This
complexity is a result of the aggregation, in the scene graph
model, of the scene content data and the scene temporal data. The
resulting 3D audio scene descriptions, are in turn, difficult to
re-author and significantly increase the complexity of 3D audio
scene renderers. In contrast, XML3DAUDIO follows a new scene
orchestra and score approach which allows the separation of the
scene content data from the scene temporal data; this simplifies
3D audio scene descriptions and allows simpler 3D audio scene
renderer implementations. In addition, the separation of the
temporal and content data permits easier modification and
re-authoring of 3D audio scenes. It is shown that XML3DAUDIO can
be used as a new format for 3D audio scene rendering or can
alternatively be used as a meta-data scheme for annotating 3D audio content.

Rendering and perception of the apparent extent of sound sources
in 3D audio displays is then considered. Although perceptually
important, the extent of sound sources is one the least studied
auditory percepts and is often neglected in 3D audio displays.
This research aims to improve the realism of rendered 3D audio
scenes by reproducing the multidimensional extent exhibited by
some natural sound sources (eg a beach front, a swarm of insects,
wind blowing in trees etc). Usually, such broad sound sources are
treated as point sound sources in 3D audio displays, resulting in
unrealistic rendered 3D audio scenes. A technique is introduced
whereby, using several uncorrelated sound sources, the apparent
extent of a sound source can be controlled in arbitrary ways. A
new hypothesis is presented suggesting that, by placing
uncorrelated sound sources in particular patterns, sound sources
with apparent shapes can be obtained. This hypothesis and the
perception of vertical and horizontal sound source extent are then
evaluated in several psychoacoustic experiments. Results showed
that, using this technique, subjects could perceive the horizontal
extent of sound sources with high precision, differentiate
horizontally from vertically extended sound sources and could
identify the apparent shapes of sound sources above statistical
chance. In the latter case, however, the results show
identification less than 50 % of the time, and then only when
noise signals were used. Some of these psychoacoustic experiments
were carried out for the MPEG standardisation body with a view to
adding sound source extent description capabilities to the MPEG-4
AudioBIFS standard; the resulting modifications have become part
of the new capabilities in version 3 of AudioBIFS.

Lastly, this thesis presents the implementation of a novel
real-time 3D audio rendering system known as CHESS (Configurable
Hemispheric Environment for Spatialised Sound). Using a new signal
processing architecture and a novel 16-speaker array, CHESS
demonstrates the viability of rendering 3D audio scenes described
with the XML3DAUDIO scheme. CHESS implements all 3D audio signal
processing tasks required to render a 3D audio scene from its
textual description; the definition of these techniques and the
architecture of CHESS is extensible and can thus be used as a
basis model for the implementation of future object
oriented 3D audio rendering systems.

Thus, overall, this thesis presents contributions in three
interwoven domains of 3D audio: 3D audio scene description,
spatial psychoacoustics and 3D audio scene rendering.