Behaviour, Structure and Causality in Procedural Audio - Game Sound Technology and Player Interaction: Concepts and Developments - page 318

or the surrounding world (such as footsteps over
changing ground materials). Making this mapping
well behaved can be a difficult programming task
that requires an artistic, human contribution.
A model must also explain/manifest the rela-
tionship between its own parts or subsystems. For
example, a helicopter is a complex machine with
many parametrically coupled sources, an engine,
gear box, exhaust system, and two propellers. Their
individual performance parameters are linked
by a higher set of control equations that accord
with proper flight manoeuvres. The ideal set of
parametric controls, from the viewpoint of game
audio design, would be the parameters used to
actually fly the vehicle (plus an observation point
vector). Consequently, procedural audio relies on
much tighter coupling with physics computation,
at least if efficiency gains are to made by avoiding
duplicate calculations.

and versatile control. Pragmatic implementations
often harness both classes.
Synthetic methods are not limited to the produc-
tion of signals at the audio rate, they apply also
to control systems above the audio DSP level, or
any other model feature that can be computed.
Examples are the rolling drinks can in Designing
Sound which has an inertial control model distinct
from the sound production, or fragmentation
models of Zheng and James which concentrate on
the control level of particle debris and leave the
actual sound generation to precomputed proxies.
In some racing games it is the sonic behaviour
of the engine and vehicle as an overall system
that could be described as procedural, while the
synthesis itself is at best a naïve granular, and at
worst carefully hand blended sample loops played
back under the control of the vehicle performance
model.
As a systematic example let's return again to
the helicopter, whose engine is modelled with a
parabolic pulse source and waveguide network,
gearbox by a closed form additive expression,
and the blades by a subtractive (noise-based)
method. But their inter-relation for given flight
manoeuvres may be calculated by differential
equations at the control rate. These controls are of
course independent of the DSP used to compute
the audio signatures, which could be replaced by
cheaper methods for reduced level of audio detail
(LOAD) as the vehicle recedes into the distance.
Yet overall, the model still retains its behaviour
as the detail of the sound (and cost of production)
diminishes. Dynamic adjustment or replacement
of model components or methods to obtain chan-
ing level of detail is one very powerful aspect of
properly stratified procedural audio.

Methods

Methods are the way we concretise models. They
connect the abstract model to the solid imple-
mentation as DSP. Methods are drawn from an
extensive set of known techniques that provide
particular spectral or time domain signal behav-
iour from an assumed input parameter set, such
as additive Fourier, subtractive, non-linear wave-
shaping, FEM (lumped mass), waveguide, MSD
(elastic), FM, AM, granular, wavelet, wavetable,
fractal, Walsh and others. These are essentially
mathematical formulations, functions of time,
systems of linear and non-linear equations from
which we obtain audio signals. They fall loosely
into two classes: parametric (signal) and source
(physical). In the former, the relationship between
model and implementation must be described by
the method. In the latter, the mapping of model to
output signal is the method. The latter promises
unparalleled realism at the expense of computa-
tion cost and lack of abstract control, while the
former trades detailed realism for cost savings

Implementations

Actual implementation of audio signals requires
knowledge of practical computer engineering and
is not a subject to ponder deeply here. We shall
just glance over it, because it changes from case