Information Technology Reference
In-Depth Information
conducted with jazz pianist Mark d'Inverno. An
a posteriori
analysis of the ses-
sion by the two musicians playing is provided. Although subjective, these analysis
show that a sense of
playing together
was achieved, and the music generated by the
system, controlled by a human, was of professional-level quality.
5.7 Discussion
The major claim of this study is that all important decisions concerning virtuoso
performance in jazz can be taken at the beat level instead of note-level. This ex-
plains how virtuosos improvise melodies satisfying so many difficult and contradic-
tory constraints at high speed. By delegating the choice of individual notes within
a beat to a non-conscious, sensory-motor level, they have enough time to focus
on high-level decisions, such as influencing pitch contour, chromaticity, tonality,
etc. Concerning the
memoryless assumption
hypothesised by Longuet-Higgins (see
Sect.
5.2.1
), we invalidate it because of side-slips, which require the memory of the
last phrase played. However, the cognitive requirements remain minimal. In some
sense, most of the work is done by the fingers.
Conceptually, we do not consider Markov models as representations of musi-
cal
ideas
, but as a
texture
that can be controlled to produce meaningful streams of
notes. The mechanisms we propose (transforms and controls) turn this texture into
realistic, sometimes spectacular, virtuoso improvisations.
Concerning the relation of virtuosity studies to creativity studies, we have
stressed the importance of two important dimensions of jazz improvisation (side-
slips and fine-control) that are made possible only by
extreme virtuosity
.Wehave
shown how to model these two aspects in a Markovian context. The first one (formal
transforms) does not raise any difficult modelling issues. The second one (control)
does, and induces a very difficult combinatorial problem. How human virtuosos
solve this problem in real-time remains a mystery. It forms important future work
for virtuosity studies.
Running is not the only locomotion mode of animals. Likewise, virtuosity is not
the only mode of jazz improvisation. Our system is in fact a brittle virtuoso: it knows
how to run, but not so well how to walk. Such brittleness was pointed out by Lenat
and Feigenbaum (
1991
) in the context of expert-systems and attributed to a lack
of
common sense
knowledge. A
musical
common sense is indeed lacking in most
automatic systems, and much remains to be done to build a completely autonomous
jazz improviser exhibiting the same level of
flexibility
as humans: a competence
in virtuoso mode as well as in other modes, and the ability to intentionally switch
between them.
Slow improvisation
, in particular, is a most challenging mode for
cognitive science and musicology, as it involves dimensions other than melody and
harmony, such as timbre and expressivity which are notoriously harder to model.
However, considering melodic virtuosity as a specific mode, we claim that these
automatically generated choruses are the first ones to be produced at a professional
level, i.e. that only a limited set of humans, if any, can produce. A claim we leave to
the appreciation of the trained listener.
