Information Technology Reference
In-Depth Information
Models
untypical of the class and peculiar to the target,
like an intermittent rattle due to a broken part.
All other environmental variables should
remain constant. But this ideal of independent
parameters is rare in the real world, so it's often
necessary to play with several data sets. The en-
vironment we work in is the physical world. This
is another way of saying that physics is an objec-
tive representation of reality with which we can
describe sound signals (and their effect) produced
by objects. We describe the perceived signal in
terms of the things that produce the signal, object
and stimuli. Clearly a knowledge of sound physics
is useful, particularly mechanics, solid vibrational
physics, fluid dynamics, and gas phase acoustics
given in standard textbooks (Elmore & Heald,
1969), (Subrahmanyan & Lal, 1974).
We could reduce running water to a model of
overlapping exponentially frequency-modulated
sinusoids corresponding to Helmholtz oscillators
caused by entrained air cavities. We can reduce
fire to a componentised model of crackles, hisses,
and low frequency noise corresponding to physical
features like fuel fragmentation, gaseous expan-
sion, and turbulence during combustion. At the
surface, it is only necessary to express the model in
clear words, or in simple formulae with correctly
ordered relationships between features.
Notice that the first case can be taken superfi-
cially, only as a surface (signal) analysis, without
asking why the water comprises patterns of expo-
nentially rising sine bursts. We could just make
note of their average pitch, duration, overlap, and
density arriving at a broad model involving spectral
centroid and flux. Or, we can delve deeper into
the causes of features. The latter approach brings
physical behaviour into the picture and can offer
meaningful
performance parameters
while the
former can only ever yield a brittle, phenomeno-
logical model. The term
performance
is borrowed
from computer music where we would probably be
dealing with a musical instrument model. For the
general case of synthesis, a model is “performed”
by actions from its environment or energy changes
within itself. For a babbling brook we might say
it's performance parameters are speed of flow,
and depth of the water.
What we desire from a procedural model
is that it presents a parametric (performance)
interface, with the
smallest
set of useful control
signals corresponding to forces or values in the
environment relevant to the narrative. We say the
model captures the sound source well when this
correspondence is extensive and efficient. These
might be fixed, like the height of a waterfall,
continuously variable values, such as speed of
fluid flow or temperature, or a discontinuous or
ordinal feature set such as a texture tag (wood,
metal, stone) taken from the object's properties
MODELs, MEtHODs AND
IMPLEMENtAtIONs
Having obtained an analysis, an understanding
of the target, we wish to produce a design which
simulates its salient features and behaviours.
We break the design of sonic objects into three
conceptual strata: a
model
, which is an abstract
representation of desired behaviours; one or more
methods
, which allow us to realise audio signals
from a behavioural description; and an
implemen-
tation
, which provides the vehicle for delivery.
These decouplings, which I observed during many
years of synthetic sound design, allow modular
software engineering practice in which each layer
may be replaced whilst leaving the others intact. It
becomes possible to construct a sounding object,
and then completely replace its sound synthesis
method, perhaps swapping a subtractive for FM
method, or re-implementing the same model and
methods on a different DSP platform.
Search WWH ::
Custom Search