ï~~uniform, as would be expected from the initialization
procedure, but it gradually "humps up" to look more
normal after several generations. Large intervals tend
to be bred out fairly quickly, particularly when the
note lengths are short. After several generations the
interval distribution is heavily skewed toward short
intervals, and the average interval shrinks from
around seven scale steps to around two. Note and
rest lengths also become skewed toward the short end
as very long notes and silences are bred out.
5 Extensions and Conclusions
Several enhancements and extensions are under
way to improve GenJam's performance. An overhaul
of the chord/scale mapping procedure is being
designed that will apply knowledge-based techniques
to a wider window on the chord progression. This
hopefully will correct a few bad scale choices
currently made in some progressions.
An attempt will be made to train a neural
network to serve as at least a preliminary fitness
function for at least the measure population. The
strategy will be to extract statistical features
correlating with measure fitness to form a feature
vector, which will be the input layer to a quick-propstyle neural network. The output layer will be a
single node containing the fitness value. The training
data will come from the populations that have been
saved from the approximately two dozen controlled
training runs conducted so far.
Another experiment will seed GenJam's initial
population with measures and phrases taken from
existing tunes or transcribed solos. A similar exercise
will be to merge populations generated in separate
training sessions to hopefully get the best of both.
GenJam's two levels also could be extended upward
to section, chorus, and/or tune levels.
To place GenJam in a larger context, I'll
conclude with a brief mention of the role GAs could
play in algorithmic composition. GenJam shows that
GAs can be a useful tool for searching a constrained
melodic space. Other specific compositional tasks
should be easy to find, for example, evolving a bass
line or percussion sequence or chord progression or
series of voicings. These small tasks might even be
done concurrently in a more comprehensive system.
Evolutionary programming techniques [Koza, 1992],
where individuals map to programs or program
fragments, also promise to broaden the uses of GAs
in music. Finally, the fact that "Genetic Algorithms"
is a special topic area at ICMC 94 indicates that
others are discovering GAs. It would seem that
GenJam is not alone!
References
[Ames and Domino] Cybernetic Composer: An
Overview. In M Balaban, K. Ebcioglu and 0.
Laske (Ed.), Understanding Music with Al,
AAAI Press, Cambridge, MA, 186-205, 1992.
[Coker, 1964] Jerry Coker. Improvising Jazz.
Prentice-Hall, Englewood Cliffs, NJ, 1964.
[Dannenberg, 1993] Roger B. Dannenberg. The
CMU MIDI Toolkit, Version 3. Carnegie Mellon
University, Pittsburgh, PA, 1993.
[Fry, 1984] C. Fry. Flavors Band: A Language for
Specifying Musical Style. Computer Music
Journal 8 (4) pp. 20-34, 1984.
[Gannon, 1991] Peter Gannon. Band-in-a-Box. PG
Music, Inc., Hamilton, Ontario, 1991.
[Giomi and Lagabue, 1991] Francesco Giomi and
Marco Ligabue. Computational Generation and
Study of Jazz. Interface 20, pp. 47-63, 1992.
[Goldberg, 1989] David Goldberg. Genetic
Algorithms in Search, Optimization and Machine
Learning. Addison-Wesley, Reading, MA, 1989.
[Haerle, 1989] Dan Haerle. The Jazz Sound. Hal
Leonard, Milwaukee, WI, 1989.
[Haerle, 1980] Dan Haerle. The Jazz Language.
Studio 224, Miami, 1980.
[Haggerty, 1991] Michael Haggerty. Evolution by
Esthetics, an interview with William Latham and
Stephen Todd. IEEE Computer Graphics and
Applications 11, pp. 5-9, 1991.
[Holland, 1975] John H. Holland. Adaptation in
Natural and Artificial Systems. University of
Michigan Press, Ann Arbor, 1975.
[Homer, et al., 1993a] Andrew Homer, Andrew
Assad and Norman Packard. Artificial Music:
The Evolution of Musical Strata. Leonardo 3,
pp. 81, 1993.
[Homer et al., 1993b] Andrew Homer, James
Beauchamp, and Lippold Haken. Machine
Tongues XVI: Genetic Algorithms and Their
Application to FM Matching Synthesis.
Computer Music Journal 17 (4) pp. 17-29, 1993.
[Homer and Goldberg, 1991 ] Genetic Algorithms
and Computer-Assisted Music Composition. In
Proceedings of the Fourth International
Conference on Genetic Algorithms, Morgan
Kauffman, San Mateo, CA, 1991.
[Iverson and Hartley, 1990] Eric Iverson and Roger
Hartley. Metabolizing Music. In Proceedings of
the 1990 International Computer Music
Conference, ICMA, San Francisco, 1990.
[Koza, 1992] J. R. Koza. Genetic Programming.
MIT Press, Cambridge, MA, 1990.
[Levitt, 1981] David Levitt. A Melody Description
System for Jazz Improvisation. Master's thesis,
MIT, Cambridge, MA, 1981.
[Penneycook et al., 1993] Bruce Penneycook, Dale
R. Stammen, and Debbie Reynolds. Toward a
Computer Model of a Jazz Improviser. In
Proceedings of the 1993 International Computer
Music Conference, ICMA, San Francisco, 1993.
[Russell, 1959] George Russell. The Lydian
Chromatic Concept of Tonal Organization for
Improvisation. Concept Publishing, NY, 1959.
[Sabatella, 1992] Marc Sabatella. A Jazz
Improvisation Primer. USENET, 1992.
[Sims, 1993] Karl Sims. Interactive Evolution of
Equations for Procedural Models. The Visual
Computer 9 (8), pp. 466-476, 1993.
ICMC Proceedings 1994
137
Genetic Algorithms
