~Proceedings ICMCISMCI2014 14-20 September 2014, Athens, Greece
AutoChorusCreator: Four-Part Chorus Generator with
Musical Feature Control, Using Search Spaces
Constructed from Rules of Music Theory
Benjamin Eva
'benjamin[E
complex. is
hokudai.ac
nsf' Satoru Fukayama:2 Masataka Goto:3 Nagisa Munekatat4
t Hokkaido University, Japan
National Institute of Advanced Industrial Science and Technology (AIST), Japan
at] 2s.fukayama[at] 3m.goto[at] 4munekata[at] 5t
t. aist.go.jp aist.go.jp complex.ist. ho.jp hokudai. ac. jp
Tetsuo Onot5
ono [at]ist.
kudai. ac. jp
ABSTRACT
This paper describes AutoChorusCreator(ACC), a system capable of producing, in real-time, a variety of fourpart harmonies from lead sheet music. Current algorithms
for generating four-part harmony have established a high
standard in producing results following rules of harmony
theories. However, it is still a challenging task to increase
variation in the output. Detailed constraints for describing
musical variation tend to complicate the rules and methods used to search for a solution. Reducing constraints to
gain degrees of freedom in variation often lead to generating outputs which do not follow the rules of harmony
theories. Our system ACC is based on a novel approach
of generating four-part harmony with variations by incorporating two algorithms, statistical rule application and
dynamic programming. This dual implementation enables
the system to gain the positive aspects of both algorithms.
Evaluations indicate that ACC is capable of generating
four-part harmony arrangements of lead-music in realtime. We also confirmed that ACC achieved generating
outputs with variations without neglecting to fulfil rules
of harmony theories.
1. INTRODUCTION
Automatic composition has captivated the minds of both
musicians and scientists for decades and many approaches have already been attempted in the field of information
science [1, 2, 3]. Some of these include constraint satisfaction [4, 5, 6], example based approaches [7], genetic
algorithms [8, 9], probabilistic modelling [10, 11] and
rule based applications [12]. Recently, technologies originally from the field of music information retrieval (MIR)
are also being used to support people who create musical
works [13, 14].
Harmony is an important element in many music styles,
especially in those of classical music. Emura describes an
academic process of musical composition as 1) choosing
a simple cadence of chords, 2) deciding a melody line
which follows the structure of the sequence of chords, 3)
Copyright: ~ 2014 Benjamin Evans et al. This is an open-access article
distributed under the terms of the CevCm;rn. s
~lrpoerd, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original author and source are
credited.
replacing some of those chords with other replaceable
chords and 4) adding harmonic interest to the piece by
adding non-harmonic tones [15]. Harmonisation and reharmonisation are important aspects in each of these steps.
This fact makes the appropriate implementation of harmony theories crucial in developing automatic composition systems that incorporate aspects of tonality as those
found in classical music.
A particular task often dealt with in the study of automatic harmonisation is that of harmonising a classical
four-part chorale from a single melody line. Allan used a
data set of chorale harmonisations to train Hidden Markov Models to create four-part harmony [16]. Suzuki also
used probabilistic models for automatic four-part harmonisation [17], comparing system outputs for when chord
data was and was not used.
Rule-based approaches have also been exploited for
four-part harmonisation. Ebcioglu developed a rule-based
system with over 270 rules and used a logic programming
language for harmonising four-part chorales [12]. PhonAmnuaisuk also created a rule-based system and compared it with a genetic algorithm system which had the
same explicit rules of harmonisation implemented in it
[8]. Biles used genetic algorithms in creating jazz improvisations [9]. MacCallum also used genetic algorithms
- 1016 -
