Information Technology Reference
In-Depth Information
environmental sounds as de
ed a certain number of
perceptually relevant signal morphologies linked to these categories (Gaver
1993a
,
b
).
To date, we concluded on
ned by Gaver and we identi
five elementary sound morphologies based on impacts,
chirps and noise structures (Verron et al.
2009
). This
finding is based on a heuristic
approach that has been veri
ed on a large set of environmental sounds. Granular
synthesis processes based on these
then enabled the generation of
various environmental sounds (i.e. solid interactions, aerodynamic or liquid sounds).
Note that this atom dictionary may be completed or re
ve sound
“
atoms
”
ned in the future without
compromising the proposed methodology.
A
ned by a sum of
exponentially damped sinusoids. Such a grain is well adapted to simulate sounds
produced by solid interactions. Nevertheless, this type of grain cannot alone
account for any kind of solid impact sounds. Actually, impact sounds characterized
by a strong density of modes or by a heavy damping may rather be modelled as an
exponentially damped noise. This sound characterization stands for both perceptual
and signal points of view, since no obvious pitch can be extracted from such
sounds. Exponentially damped noise constitutes the second type of grain, the so-
called
first type of grain is the
“
tonal solid grain
”
that is de
. Such a grain is well adapted to simulate crackling
sounds. The third type of grain concerns liquid sounds. From an acoustic point of
view, cavitation phenomena (e.g. a bubble in a liquid) lead to local pressure vari-
ations that generate time-varying frequency components such as exponentially
damped linear chirps. Hence, the so-called
“
noisy impact grain
”
consists of an exponen-
tially damped chirp signal. Finally, aerodynamic sounds generally result from
complicated interactions between solids and gases and it is therefore dif
“
liquid grain
”
cult to
extract useful information from corresponding physical models. A heuristic
approach allowed us to de
ne two kinds of aerodynamic grains: the
“
whistling
grain
”
(slowly varying narrow band noise) and the
“
background aerodynamic
grain
”
(broadband
filtered noise). Such grains are well adapted to simulate wind and
waves.
By combining these
five grains using an accurate statistics of appearance, var-
ious environmental auditory scenes can be designed such as rainy ambiances, sea-
coast ambiances, windy environments,
fire noises, or solid interactions simulating
solid impacts or footstep noises. We currently aim at extracting the parameters
corresponding to these grains from the analysis of natural sounds, using matching
pursuit like methods. Perceptual evaluations of these grains will further allow us to
identify or validate signal morphologies conveying relevant information on the
perceived properties of the sound source.
4.4
Control Strategies for Synthesis Processes
The choice of synthesis model highly in
uences the control strategy. Physical
synthesis models have physically meaningful parameters, which might facilitate the
interpretation of the consequence of the control on the resulting sound. This is less
so for signal models obtained from mathematical representations of sounds.
