Information Technology Reference
In-Depth Information
In these cases (and elsewhere, as in Fourier transform-based analysis and resynthe-
sis) a compelling feature is that the channels of analysis (
p
) and synthesis (
q
)are
equivalent, such that if
f
were simply bypassed
f(x,p)
p
then the input data
would be recovered in some shadowed form. In the extreme case,
Q
=
P
−
1
,the
input sound comes back unaltered.
The
Swarm Music
and
Swarm Granulator
systems of Blackwell (
2001
) and
Blackwell and Young (
2004
) explore both MIDI and timbral domains using swarm
dynamics. Incoming analysis parameters are mapped onto attractors in an internal
space. A swarm of musical events explores this space, on occasion becoming drawn
to attractors. The positions of swarm members are mapped directly onto synthesis
parameters so that if the individuals were sitting directly on the attractors, the out-
put would precisely mirror the input. Novelty arises from the exploration of phase
space, and participation is coded in the tendency to move towards any discovered
attractor.
Swarm Music/Granulator
is a direct realisation of the dynamical systems
programme advocated in Sect.
6.3.5
.
The behaviour of the above prototypes is not always human-like, and indicates
how machine improvisers are already super-human in certain respects. Humans have
an evolved capacity for vocal imitation, extended to imitation on musical instru-
ments, but the mechanisms for perception and action are far from equivalent, the
latter using learnt motor movement to make sound. An individual learns the rela-
tionship between action and perception through practice. The capacity for shadow-
ing cannot be taken as given for humans, therefore, not only because our response
times are too slow, but because we do not have inherent mechanisms to generate
sound in equivalent ways to how we perceive sound. That said, the way we hear
sound is deeply influenced by the salience of spoken language to us, a fact which
matters in modelling human music perception, and the cognitive methods used by
human to achieve this mapping may also turn out be useful to further Live Algo-
rithms research.
A number of Live Algorithm systems also demonstrate a lack of equivalence be-
tween input and output interfaces by combining standard hard-wired audio analysis
tools (
P
), domain-general AI techniques or dynamical systems such as neural net-
works, particle swarms or generative grammars (
f
), and bespoke hand-programmed
generative systems (
Q
). This modular approach provides the opportunity to inte-
grate human and machine decision-making processes by breaking down the be-
haviour of Live Algorithms into a set of problems that can be solved in different
ways.
Lewis' celebrated
Voyager
system is an example of a system that is hand-coded
with complex rule-based behaviour (Lewis
2000
). The system uses standard audio
analysis tools in order to render incoming audio in a MIDI-based form.
Voyager
is
designed in a modular way according to Lewis' introspective detailing of improvisa-
tion behaviours. It encodes musical knowledge, acting as a proxy for Lewis' creative
agency, and achieves each of our four goals through the success of this knowledge
encapsulation. It achieves novelty using simple combinatoric processes.
Young's
Piano_prosthesis
and related prosthesis projects (Young
2008
) demon-
strate a more hybrid approach using standard analysis tools, such as IRCAM's
yin
=
∼
