The Extended Composer - Computers and Creativity - page 191

Hours : On the scale of minutes and hours, we may develop a piece, adding
phrases to sections and sections to movements. These can be replayed to observe
their fit within the wider narrative.
Scaling beyond the length of a single piece of music, we have systems such as the
Continuator (Pachet 2003 ), which reflects back the statistical properties of a user's
musical behaviour over the length of entire phrases. The reward is that, through
listening back to a distorted edition of their original patterns, the player can better
understand their own habits by hearing them recontextualised.
Generative algorithms can be used to apply a similar process of segment organ-
isation, perhaps with generated components or selections from a database. Applied
in interactive composition environments, with an aesthetic fitness function provided
by their human counterpart, such a process can provide an effective heuristic-based
method of exploring musical possibilities (Wiggins et al. 1999 ).
The development of a single work is often achieved through iterated genera-
tion/evaluation with a particular interactive music system. It is also possible that an
artist is able to modify the code of a music system co-evolving the work and the
system. In this case a slower feedback loop can occur: the system is allowed to run,
perhaps repeatedly, and its output observed (evaluation); based on this observation,
the code is modified to alter or enhance the system's behaviour (generation). This
process can be seen quite transparently in live coding performances, where code is
written, run and modified as part of the live performance.
Years : Our personal style may develop with reference to previous works and
external stimuli; a visit to a gallery may prompt a radical departure which causes us
to rethink our trajectory, or consider it along a new axis. A prominent example of a
system that evolved on this scale of feedback is AARON, an autonomous drawing
system developed by Cohen ( 1995 ) over several years.
Developments at this scale can also be observed through data mining of musical
corpus. For example, by matching musical phrases against a large corpus of record-
ings based on similarity measures, Query-by-Example (Jewelletal. 2010 ) enables
its users to reflect on how their performances have developed over long periods—or
relating them to bodies of other musicians' work. We could imagine such tools en-
tering much more widely into the reflective practice of artists, allowing them to more
closely understand their own historical lineage and their position within a wider con-
text, potentially discovering hidden relationships with previously-unknown peers.
Decades : Over decades, cultural fashions ebb and flow. It is this temporal nature
of styles which causes many works to fail to be accepted often for many decades.
Punk, new-wave and dance music are all examples of cultural fashions in UK music
for example.
Centuries : At the timescale of entire eras, we can interrogate historical tenden-
cies through tools designed for H-introspection (Sect. 7.3.5 ).Theworkofempiri-
cal musicologists have laid some groundwork for computational analysis of trends
at this scale, while musical models such as those of Cope ( 1996 ) study cultural