Physical Modelling for Sound Synthesis - Game Sound Technology and Player Interaction: Concepts and Developments - page 352

in Interactive Worlds (Menzies, 2002) describes
the beginnings of what would later become the
Phya project. This project is detailed later when
discussing Menzies' 2007 paper Physical Audio
for Virtual Environments, Phya in Review (Men-
zies, 2007).
Meanwhile, Doel and Pai continued their
research on efficient ways to implement modal
synthesis by applying it to complex scenes in-
volving a large number of sounding objects. In
Interactive Simulation of Complex Audio-Visual
Scenes (Doel, Knott, & Pai, 2004) they describe
a process which they call mode pruning in which
the effect of auditory masking is used to predict
which modes are perceptually unimportant so
that they can be excluded from synthesis, hence
saving processing power. The technique is not
just applied to individual sounding objects, as
in previous work: all sounding objects are now
considered together. This requires the process to
be carried out at run time as objects are being
sounded and reaps further gains in efficiency. In
addition, this project eases the load on the CPU
by offloading as much computation as possible
to the Graphical Processing Unit (GPU).
The next major contribution from Doel and
Pai was published as a chapter entitled Modal
Synthesis for Vibrating Objects (Doel & Pai,
2006) in the topic Audio Anecdotes III. As a topic
chapter, this work is self contained and as such
includes a comprehensive review of the theory of
modal synthesis before explaining how they have
implemented it using a bank of band-pass filters.
In order for the project to facilitate environments
that may potentially grow to be very large, with
sounding objects continuously being created and
destroyed, a modular solution was designed. From
a programming point of view, this means that code
is written in classes and, at run-time, instances
of these classes are created and destroyed as re-
quired. For example, a class named SonicObject
contains functionality for rendering to a system's
audio hardware, ModalSonicObject implements
modal synthesis and an AudioForce class provides

functionality for extracting an audio signal from
an object. Modular programming also allows a
derived class to inherit functionality from a base
class and so common code can be written in a base
and used in many specialized derived classes. This
important advantage ensures the functionality of
the system can easily be extended in the future.
In this work the authors note that: “Modal
synthesis can also be used to model other types of
physical systems which can be modeled by excita-
tions acting on resonances, such as car engines,
rumbling sounds, or virtual musical instruments”
(Doel & Pai, 2006, p. 100). They go on to explain
their implementation of a system to synthesize the
sound of a four-stroke engine and a four-cylinder
engine. To do this, they determine modal data that
represents “a lumped model of everything that
is vibrating” (p. 114) and drive it with a force
signal the generation of which is inspired by the
workings of an engine. They state that reasonable
results were obtained with their “extremely simple
models” (p. 115) and express optimism about the
range of engine sounds that could be achieved by
spending more time developing the technique.
In the same year, Nikunj Raghuvanshi and
Ming C. Lin published research on their use of
physical modelling in large environments in a pa-
per entitled Interactive Sound Synthesis for Large
Scale Environments (Raghuvanshi & Lin, 2006).
Their method of determining an object's modal
data is similar to the one used by O'Brien et al.
(2002) and can similarly be used for arbitrarily
shaped objects. They consider three-dimensional
objects as being composed of a thin shell and a
hollow inside and they represent it as a mesh of
particles connected by damped springs. Off-line
computation of a matrix of particle masses and
one of elastic forces renders the modal data for an
object from which synthesis can be performed. A
large portion of the paper details ways in which
computational efficiency is gained and 3 tech-
niques are detailed. Similar to Doel and Pai in
Interactive Simulation of Complex Audio-Visual
Scenes, the total computational expense is reduce