Information Technology Reference
In-Depth Information
virtuosity
. The two features we focus on are only possible thanks to extreme vir-
tuosity: (1) side-slips and (2) fine-grained control. We describe and interpret these
two major contributions to style invention in the context of Markov-based music
modelling.
After a review of the state-of-the art in jazz modelling, we describe a Markov-
based model of jazz improvisation and show that it is well adapted to generate
melodies that fit with arbitrary chord progressions. We then use this basic model
to
state and solve
the two main issues of jazz generation: control and side-slips.
5.3 Modelling Jazz Improvisation Generation
Many studies have addressed music composition and improvisation, so we focus on
those specifically addressing jazz. As is often the case in computer science, these
studies follow the general algorithmic trends of the moment. We handle separately
the case of Markov modelling as this is the core of our proposal.
5.3.1 Non-Markovian Approaches
Ulrich (
1977
) proposed an all-encompassing system that performs chord sequence
analysis and chorus generation using a purely algorithmic approach, reaching a rea-
sonable level of musicality. Walker (
1997
) and Thom (
2000
) built interesting sys-
tems emphasising the
dialogue dimension
of improvisation rather than the musi-
cal quality. A more ambitious case-based reasoning approach was proposed by Ra-
malho and Ganascia (
1994
), emphasising the role of motivic components, and fol-
lowing the 'knowledge level' paradigm. This approach proposes to explicitly recon-
struct a cognitively plausible model of a working jazz memory, and was applied to
the automatic generation of bass lines, yielding some interesting outputs, favourably
compared to Ron Carter's samples (Ramalho
1997
). It relied on a manually entered
set of cases, limiting its scope in practice. Genetic algorithms have been used for
music generation by a number of researchers (Weinberg et al.
2008
, Bäckman and
Dahlstedt
2008
, Papadopoulos and Wiggins
1998
), yielding real time systems used
in concert, and producing interesting improvisation dialogues, like in the
GenJam
system of Biles (
1994
). These systems, again, apply a general paradigm (here, evo-
lutionary algorithms) to chorus generation in a top-down approach, without concern
for harmonic satisfaction and continuity. Their outputs, although sometimes spec-
tacular, are still below the level of professional musicians, and do not display par-
ticular virtuosity. Interestingly, the system described by Grachten (
2001
) was used
as a basis for studying jazz
expressivity
in saxophone solos (Ramirez et al.
2008
).
The studies described in Hodgson (
2006
), also use a genetic algorithm but focus on
detailed characteristics of the Charlie Parker style. Notably, Hodgson shows the im-
portance of dyadic (two-note) patterns in the elaboration of Charlie Parker's melodic
