94 Psychoacoustics the actual differences are ineffable. We only perceive some mysterious quality
difference without being able to explain it (Broadbent and Ladefoged 1959).
Scaling If we can discriminate a quality then it may have some scale or dimensions
against which to measure it. Scaling is the act of applying something quanti-
tative to distinguish sounds. The hardness of a bell hammer and the distance
of the observer from the bell are two physical parameters that lead to obvi-
ous scalings. They reliably map something measurable in the physical domain
to something humans can say about the sound, something in the perceptual
domain. Not all physical parameters affect a sound. An obvious case is the mass
of a pendulum, which doesn’t even figure in behaviour, or the independence of
volume in liquid sounds (splashing in a river sounds the same as splashing in
the ocean; it’s depth that has an effect). Some physical parameters affect a
sound in more than one way. Temperature has a multitude of influences on the
microstructural sound of a fire burning. In sound design we wish to reduce the
set of parameters to the minimum needed to perceptually capture the sound.
Similarity The similarity of two sounds from the same object tells us something about that
object. Imagine a recording of a tin can being hit softly then successively harder
with a beater. If we split up the recording and play back the hits randomly we
would still be able to rank them back to the correct order by listening to them
carefully. The change in amplitude from quiet to loud and the shift in spectra
as more energy is imparted form a recognisable ordering. You could do the
same for a bottle and rank all the hits in order when it is hit differently with
the same beater. You would instantly know that any two sounds pulled from
the two different recordings were of different objects, because of their frequency
patterns. You could compare examples of each hit with the same energy from
the same beater to reveal a third new parameter, the beater. To know the para-
metric dimensions of a sound we can use tests of similarity to discover them.
As sound designers we do this all the time: we move a fader and compare the
before and after versions according to some space or multidimensional grid of
features. If a group of people is presented all possible pairs of examples from a
set of sounds and asked to rate similarity, the salient features of the sound are
revealed statistically. This is called
multidimensional scaling . Algorithms such
as the Kohonen self-organising map can be used to cluster sounds and provide
a
similarity space . Both are approaches to analysing the
timbre space of the
sound object to try and discover parameters. We can use this concept the other
way around too. If we already know the timbre space of an instrument or sound-
producing object, because we make it from a model, then we can predict the
similarity of various settings from what we know about discrimination of sound
features. In creative terms this offers a powerful device. Juxtaposition of sim-
ilar sounds can provide a strong associative link, such as the falling raindrops
blended with machine-gun-fire in the war film
Saving private Ryan .