Behaviour, Structure and Causality in Procedural Audio - Game Sound Technology and Player Interaction: Concepts and Developments - page 332

of an object designer is to wrap up or “collapse”
(often many) parameters into a small set of useful
ones. Realistically for the rain example we would
like to pick up only one or two variables from
the game engine: how much it is raining (maybe
flux in drops per square meter per second) and
the material texture tag of an object which will
sound (and perhaps this would be discovered au-
tomatically from nearby objects within a hearing
radius). Parameter range should be conformal and
continuously differentiable in the sensible space,
that is to say, it should contain no poles or zeros
where some combination of parameters causes an
anomalous signal output. In other words, we want
a model with the fewest meaningful parameters,
which for all settings will smoothly produce the
correct sonic behaviour. Such a model would at
least satisfy the criteria of Beck's (2000) “acoustic
viability” which I shall paraphrase as a “simple
set of parameters that work with a consistent
underlying physical process”.
Concerning hybridisation, the subject that was
fashionably called morphing in the late 1990s;
does this mean that given two parameter points
along a presumably continuous behavioural line,
a listener would correctly place a new example?
The work of Wessel (1973) and Grey (1975) on
classical instrument timbre space seems to say
that if sounds are hybridised by simple mixing of
spectra or interpolation of envelope curves then
the perceptual interpolation is relatively smooth
and 'navigable'. That is to say, given a trumpet and
flute somewhere in the middle, even though there
are numerous candidate parameter combinations,
there will be an area of “flumpets” and “trutes”
(and also two clear flute and trumpet boundaries
with hysteresis such that the matching space is
trisected). Yet the more complex the synthesis
becomes the less likely perceptual/behavioural in-
terpolation can be achieved. Even if isolated points
of sensible accordance can be found, unless the
model is causally and structurally defined all other
points fail, even those close to the working ones.
Often the parametric space between superficially

similar sounds just gives noise. The FM method
can have parametric spaces where this happens.

DEVELOPMENt ADVANtAGEs
OF PrOcEDUrAL AUDIO

I will here present only a short overview of the
developmental advantages of procedural audio
since I have dealt with these in some detail else-
where. As mentioned above, the side effects of
obtaining behaviour for free, as a result of object
design, are wonderful. It does, however, have
some potentially annoying but not insurmount-
able, artistic problems. Coupling between existing
abstractions might be problematic. Changing one
part of a model may change others in unexpected
ways, so object classes might have to be overloaded
for special cases and one off events.
The potential solution to combinatorial asset
growth is a benefit that alone may be enough for
developers to embrace procedural audio. Where
space is the issue, something we have not talked
about yet is source compactness. As code, sound
can be stored and transmitted with orders of mag-
nitude better space efficiency (say 10kB instead
of 10MB). Its value in replicated network games
and mobile applications is high. This is topical at
the present time in the UK where the gold-rush
to unbounded 3G bandwidth growth has hit a
wall, leaving many business strategies without
a paddle. Suddenly data reduction and efficient
representation are on the agenda again, at least
for casual mobile gaming.
Overall cost also improves compared to sample
playback (which must grow with linear cost) above
about 300 concurrent sources. In computer graph-
ics, there is a long-established concept of visual
level of detail. Objects at a distance, fast moving
objects, or those of low relevance are drawn in a
cursory fashion using techniques of texture MIP-
mapping and partial rendering. Especially relevant
to computer game production is to introduce the
idea of LOAD, a concept I am hoping to develop