Behaviour, Structure and Causality in Procedural Audio - Game Sound Technology and Player Interaction: Concepts and Developments - page 333

further in consultancy for the Thief title with audio
director Paul Weir. This is made possible in pro-
cedural audio because, while data playback incurs
a fixed cost per source, a computational source
may have variable cost. Also, the application of
a perceptual audio model to a game in which AI
enemies respond primarily to sound seems an
obvious direction in virtual (simulated) machine
listening research.

and getting around the limitations of what are
simplistic, inelegant tools.
Something I classed previously as a technical
development advantage, dynamic level of detail,
is also something to be seen as an aesthetic op-
portunity. Many remedial tools, which exist
because the philosophy of game audio is still in
essence brute force, might be dispensed with. In
a scene of sufficient complexity the procedural
approach allowing LOAD is to be preferred,
not only because it becomes more efficient,
but it also offers a better quality of sound. This
“sparse” quality accrues from the ability to se-
lect psychoacoustically relevant structures thus
dovetailing with existing sounds whereas, with
sample playback, there are ultimately limits to
unbridled superposition. Game audio is frequently
criticised as too dense. Designers are constantly
badgered to produce “Big” sounds which are then
clumsily overused. The loudness wars have run
out of earth to scorch and there's nowhere left to
go. The limits of superposition, grey goo mixes
(see also Hug, 2011), can be remedied only by
taking something away, which is only possible
if you have something to take away . Because
procedural mixes are constructed from atomic
contributions, it is possible to use psychoacous-
tic prioritisation (similar to Bark band masking
used in MP3 compression) to keep mixes sparse
and clean while reducing CPU cost at the same
time (for example see Moeck, Bonneel, Tsingos,
Drettakis, Viaud-Delmon, & Alloza, 2007). This
allows space for significant events to be “punched
through” under artistic control.
Remedial techniques employed in engines like
the Wwise and FMOD middleware can selectively
duck and filter sources to achieve focus. Beyond
about 300 sources, the real-time dynamic process-
ing required to tame and focus a mix becomes such
a burden that it is probably better to deconstruct
sources and to select behaviourally significant
structures rather than allowing them to compete
in a mix, hoping that the listener can properly at-
tend. This logic is natural to composers and music

Artistic Advantages of Procedural
and behaviouraltthinking

Though I am fond of advocating procedural game
audio from a technical position it seems inescap-
able that there are numerous artistic advantages
too. These have been difficult to formulate and
sometimes extremely hard to find a sympathetic
audience for, despite also being a sound artist
in my career and recognising the ways we have
become stuck as sound artists unable to move
forward artistically until the limitations of samples
have been surpassed; as I put it earlier; shaking
off the concrete and dreary realism. The goal of
sonic structuralism and formalism here is only to
open a doorway to effective computational models.
Once that door is open I think the results will be
spectacular, not just technically but artistically. An
accusation often levelled by sceptics of computa-
tional sound is that it's too complex for artists to
deal with. Oddly it's never actual artists or prac-
titioners who are saying this. My thought is that
the same was said of Pixar by pen and ink artists,
just as the same was said about word-processors.
And it is not just a question of tools. It is insult-
ing to artists to assume the requisite concepts for
advanced audio are beyond them. It is from here
that the most creative of the next generation of
sound designers will come to enjoy manipulating
and extending the exposed structure and func-
tions available at the new frontier of sonic arts.
In many ways the existing event-asset paradigm
is as arbitary as any other. Much creative energy
is currently spent learning proprietary interfaces,