Physical Modelling for Sound Synthesis - Game Sound Technology and Player Interaction: Concepts and Developments - page 351

Figure 1. Virtual rock in a wok, from the Foleyautomatic project. (© 2010 Kees van den Doel. Used
with permission.)

interactive simulations. The following year, how-
ever, O'Brien, along with Chen Shen and Chris-
tine M. Gatchalian published a paper which ad-
dressed this.
Synthesizing Sounds from Rigid-Body Simula-
tions (O'Brien, Shen, & Gatchalian, 2002) notes
that when a body's sound producing deformations
are small enough, which they generally are, they
can be decoupled from its rigid-body behaviour.
This means that it is possible to use a rigid body
simulator to calculate how objects move on a
macroscopic scale while audio is generated by a
separate process. O'Brien et al. have implemented
this twice, with two different third-party physics
engines. Modal synthesis is the method of choice
here and the unique contribution of this work is
the way in which modal data is obtained. Unlike
FoleyAutomatic , they do not require any experi-
mental data and, unlike Sonic Explorer, arbitrarily
shaped objects can be included because an object's
modal data is determined using a finite element
scheme. A tetrahedral method previously used
by O'Brien et al. (O'Brien et al., 2001) is now
modified to produce modal data for an object of
a given size, shape, and material. While this pre-
computation phase may take a few hours it allows
sound synthesis to happen in real-time during
the simulation. The authors also note that some
parameters can be changed after the computation
without affecting the modal frequencies while
other changes affect all frequencies by the same
ratio and so can be quickly computed.
The same year saw a contribution from Doel
and Pai, this time along with researchers at the

Institute for Hearing Accessibility Research in the
University of British Columbia. Measurements
of Perceptual Quality of Contact Sound Models
(Doel, Pai, Adam, Kortchmar, & Pichora-Fuller,
2002) explored ways to improve how the modal
synthesis technique can be employed. The authors
analysed the recording of a metal vase being struck
and found 179 modes. However after “laborious
trial and error” (Doel et al., 2002, p. 2) they found
that only 10 to 15 of those modes were perceptu-
ally important. By conducting listening tests, they
then set about finding an algorithm for sorting an
object's modes by perceptual importance so that
only the most important modes need actually be
synthesized, thus saving processing power. For
example, a simple approach was to weight the
importance of a mode by its gain while more
sophisticated techniques considered the effect of
masking. After describing their techniques, ex-
perimental procedure, and results, they concluded
that while results varied substantially among
participants the “efficiency of the synthesis can
be improved by several orders of magnitude by
a careful selection of the modal model” (Doel et
al, 2002, p. 4).
The year 2002 also saw the first published
contribution to the field from Dylan Menzies.
Realising the promise of physical modelling for
sound synthesis in virtual environments and the
potential of these techniques to work closely with
a physics engine, Menzies set about creating a
modular framework to enable the sounding and
interaction of many objects in a virtual world.
Scene Management for Modelled Audio Objects