One of the most fascinating topics in acoustical
research of recent decades is certainly simulation and reproduction of
synthetic sound fields. The huge efforts made to get a plausible
(which doesn’t mean just audible) solution of this problem, are
justified by the numerous fields of application of such a know-how.
From room acoustics going to adaptive filtering and virtual reality,
the need of mastering in great detail every perceptional cue
(sometimes even in real time), is steeply growing. For this reason the
complexity of models and algorithms is growing at the same rate, and
demands greater and greater calculation resources. Also our research
is concerned with these goals, though in recent times it led us to
investigate new directions in the fields of sound processing and
composition. We will try to explain what this path really is, and how
it was conceived.
On one side we focused on the problem of audio
signal processing implemented through a parallel architecture; thus
the "Fly 40" system was born, a natural successor of the
"Fly 30". Unlike the latter, which featured a single
TMS320C30 chip, "Fly 40" is based (although the HW
architecture is open) on one Ariel card, bringing four TMS320C40 DSP’s
connected in a ring pipe fashion. Like the "Fly 30" system
and several applications built around it, also "Fly 40" is
going to be used by Centro Ricerche Fiat as a powerful environment for
synthesis and simulation.
On another side we are dealing with the topic of
spatialization. The whole process of simulation of an acoustic virtual
space is a very complex one. First of all we need a satisfactory model
to describe the propagation of the sound field over a well-defined
space with the wanted degree of precision. We think of this model as a
hybrid one: indeed, the wider the range of application of the model,
the more we have to deal with different aspects of propagation such as
diffraction, diffusion and normal modes. For this purpose we have
developed a program which calculates the point-to-point impulse
response of a space once the user has input all the features
(geometry, source, receiver, medium, planes) of the scenery he intends
to simulate. Great care has been taken in the modeling of reflections
according to a spatial oversampling criterion and to the frequency
dependence of air absorption. We have also provided some tools for
off-line analysis of the results. Once the impulse response has been
carried out, the subsequent step is its convolution with the input
audio signal; this work may be quite efficiently accomplished by the
above-mentioned DSP system, provided that the latter is brought to fit
this task.
The experience we earned carrying out this research and the needs
we reckoned with, led us to focus on a new architecture, which should
be actually dedicated to the processing of audio signals. We think of
it as composed of a general purpose host supported by a powerful audio
acceleration unit via a proper communication system. This unit may be
called "the transformer" by virtue of its skillness in
executing its main task: to provide a fast and useful representation
of a time-domain object, in any other frequency or time-frequency
domain, and vice versa. This approach may not only provide a powerful
tool for analysis and synthesis, but also cast a new light on the
topic of composition, thus allowing the composer to build by himself
different "sights" and transformations of the same musical
material.
