Virtual Music: Computer Synthesis of Musical Style

551 Pages·2001·33.48 MB·english

by  David Cope

Preface
On November 8 and 9, 1997, as a part of a series of colloquia on computers and
creativity,DouglasHofstadter,inconjunctionwiththeCenterforComputerAssisted
Research in the Humanities (CCARH) and Stanford University, presented a week-
end of papers, panels, concerts, and discussions centered around the works of the
Experiments in Musical Intelligence program. Presenters included Douglas Hof-
stadter,EleanorSelfridge-Field,BernardGreenberg,SteveLarson,JonathanBerger,
Daniel Dennett, and myself. Virtual Music: Computer Synthesis of Musical Style
serves as a document to this colloquium, an extension of many of the thoughts pre-
sented there, and an annotated publication of sample musical output of the Experi-
mentsin Musical Intelligence computer music composingprogram.
With the exception of the last two chapters, the presentation order of this book
closely follows that of the colloquium. Readers will note that both Doug Hofstadter
and I discuss the basic principles that Experiments in Musical Intelligence follows.
This apparent redundancy proved very e¤ective at the Stanford colloquium in that
Doug’sinformalviewof theprograme¤ectively servesasanintroductory tutorial to
my more in-depth presentation.
VirtualMusic:ComputerSynthesisofMusicalStyleisdividedintofourmainparts.
The first part provides a background of Experiments in Musical Intelligence. It cites
precedents such as the eighteenth-century Musikalisches Wu¨rfelspiel and other
composing algorithms and presents a version of what I call The Game—a reader
participation–style recognition test. This is followed by a general overview of Ex-
periments in Musical Intelligence as seen and heard through the eyes and ears
of Douglas Hofstadter, a renowned cognitive scientist and Pulitzer Prize–winning
authorofGo¨del,Escher,Bach.IthenrespondtoDoug’s commentarywhich leadsto
a description of the fundamental principles upon which the Experiments in Musical
Intelligence programoperates.
ThesecondpartfollowsthecompositionofanExperimentsinMusicalIntelligence
workfromthecreationofadatabasetothecompletionofanewworkinthestyleof
Mozart. This presentation includes, in sophisticated laypersons’ terms, relatively
detailed explanations of how each step in the composing process contributes to the
final composition, with an exampleof ine¤ective as well as e¤ective output.
The third part provides perspectives and analyses of the Experiments in Musical
Intelligence program. These scholarly commentaries include analyses, critical evalu-
ation, and relevant history and documentation, as appropriate. These chapters also
discusstheimplicationsoftheprogram’scompositions.ThescholarsincludeEleanor
Selfridge-Field (musicologist, associate director of the Center for Computer Assisted
Research in the Humanities at Stanford University), Bernard Greenberg (Bach
scholar and co-inventor of the Symbolics Lisp machine), Steve Larson (music theo-
viii Preface
rist), Jonathan Berger (noted composer and theorist), Daniel Dennett (cognitive
scientist and author of Darwin’sDangerous Idea),and Douglas Hofstadter.
The fourth part provides my response to the commentaries presented in the third
part along with thoughts on a variety of implications I see as a result of my work
with Experiments in Musical Intelligence. These implications include reflections on
artificial intelligence, music cognition, aesthetics, intention, and the future of both
Experiments in Musical Intelligence and, indirectly, the use of computers in the new
millennium.
AppendixAincludesthemusicofthedatabasesusedtocreateanewMozart-style
movement which appears in appendix B. Appendix C contains a rejected Mozart-
style movement, as discussed in chapter 10. Appendix D provides extended musical
examples referred to and discussed in Virtual Music. The music covers styles from a
four-hundred-year span of classical music history and includes such composers as
Scarlatti, Bach, Mozart, Beethoven, Schubert, Chopin, Rachmanino¤, Prokofiev,
Joplin,Barto´k,andothers.Eachmusicexampleisprecededbyabriefdocumentation
of its composition or thoughts on its aesthetic value. Appendix E presents the key to
The Game inchapter 1(see figures 1.11–1.13).
IhavewrittenVirtualMusicusingnontechnicalterminologyandinastylewhichI
feel will appeal to the layperson with an interest in classical music, as well as to
individuals knowledgeable about artificial intelligence. Since many of the examples
are musical scores, an ability to read music will be beneficial. Those unable to play
these examples on the piano will find the music available on the accompanying
compact disk or on commercially available CDs of Experiments in Musical Intelli-
gence’s music: Bach by Design, Classical Music Composed by Computer, and Virtual
Mozart (Cope1994, 1997b, 1999).
Like most computer applications, Experiments in Musical Intelligence has had
many incarnations. The program has been revised continually over many years.
Therefore, the music from the mid-1980s was created by a substantially di¤erent
program from the one that exists now. There are definite commonalities among the
various forms of the program, and I tend to emphasize these features in my writings
and discussions. I mention these versions for several reasons. First, individuals who
have followed my work through various writings can become confused by the varia-
tionsinthedescriptionsofExperimentsinMusicalIntelligencetheyencounter.While
I try not to contradict earlier publications, I tend to emphasize important, newer
aspects of the program in current writings, aspects which may not have even existed
in earlier versions. Readers should be aware of the distinctions of the various incar-
nations of the program. Second, I do not want listeners of Experiments in Musical
Intelligence’s music to be searching for compositional processes which, by virtue of
Preface ix
theperiodinwhichtheywerecreated,donotexist.Finally,Imentiontheseversional
di¤erences in Experiments in Musical Intelligence because the approach I take in
describingtheprograminthisbookisacurrentversion,whichhasonlyexistedinthis
form since about 1993. Most of the Experiments in Musical Intelligence program’s
output,therefore,cannot be understood to havebeencomposed using all of the pro-
cessesdescribedinthisbook.However,theworkscomposedspecificallyforthisbook
in appendixes Band Cwere created using this version.
I wish to thank the many individuals who have contributed so very much to this
body of work, particularly Douglas Hofstadter, who created the series of colloquia
oncomputersand creativity atStanfordUniversity in1997which ledtothecreation
of this book. I also wish to thank the Center for Computer Assisted Research in the
Humanities,particularlyEleanorSelfridge-Field,DonAnthony,andWalterHewlett.
Without the moral support and advice of colleagues such as these, this book could
not have been completed.
The CD
The CD which accompanies this book contains two sets of music. The first set con-
tains some of the musical examples ofVirtual Music: Computer Synthesis of Musical
Style. Each example is named and numbered according to the text for easy refer-
ence. The second set contains performances of many of the works presented in the
appendixes of Virtual Music, especially those not already available on commercial
recordings.
The music of both of these sets is performed here by a computer algorithm I
developed called, simply, Performance. Like Experiments in Musical Intelligence,
Performance relies on a database of already performed music for analysis which it
then uses as a model. Performance does not use recombinancy, however, but rather
usesitsanalysesastemporallyflexibletemplates.Performancealtersonlytherhythm,
articulation, and dynamicsof the worksit performs.
Musicon theCD
Figures Track
1.1–4 1
1.7–9 2
1.11a 3
1.11b 4
1.11c 5
1.11d 6
1.12a 7
1.12b 8
1.12c 9
1.12d 10
1.13a 11
1.13b 12
1.13c 13
1.13d 14
4.1–2 15
4.5–7 16
4.9–13 17
xii TheCD
5.1 18
5.3 19
Appendix B
Mozart movement 2 20
Appendix C
Mozart reject movement2 21
Appendix D
Scarlatti sonata 22
Bachinvention 11 23
Bachinvention 13 24
Bachfugue 25
Mozart sonata movement2 26
Beethoven sonata movement3 27
Mendelssohnsong without words 28
Chopin nocturne 29
Brahms intermezzo 30
Music Not onthe CD
Works in this book not on the accompanying CD appear on the following commer-
cial recordings where BBD is Bach by Design (Centaur Records CRC 2184) and
CMCCisClassical Music Composed by Computer (CentaurRecords CRC 2329):
Bach-style inventions
1 appears as 2on BBD
2 appears as 1on CMCC
5 appears as 2on CMCC
6 appears as 4on BBD
7 appears as 3on CMCC
8 appears as 3on BBD
10appears as 5on BBD
12appears as 1on BBD
TheCD xiii
Beethoven-style sonatamovements 1&2onCMCC
Joplin-style rag onCMCC
Prokofiev-style sonata onBBD
Chopin-style mazurka onCMCC
Rachmanino¤-style suite onCMCC
I
FUNDAMENTALS
Much of what happens in the universe results from recombination. The recombination
of atoms, for instance, produces new molecules. Complex chemicals derive from the
recombination of more rudimentary particles. Humans evolve through genetic recom-
bination and depend on recombination for communication, since language itself results
from the recombining of words and phrases. Cultures thereby rely on recombination to
establish and preserve their traditions. Music is no di¤erent. The recombinations of
pitches and durations represent the basic building blocks of music. Recombination
of larger groupings of pitches and durations, I believe, form the basis for musical com-
position and help establish the essence of both personal and cultural musical styles. As
will be seen, recombination also plays a significant role in my Experiments in Musical
Intelligence program.
1
Virtual Music
Virtual music represents a broad category of machine-created composition which
attemptstoreplicatethestylebutnottheactualnotesofexistingmusic(Cope1993).
As will be seen, virtual music has existed in one form or another for centuries. With
the advent of computers. however, the potential for virtual music has multiplied
exponentially.Inthischapter,Iprovideabriefbackgroundofvirtualmusicandthen
ask you to participate in three listening tests whichwill challenge your ability to rec-
ognizehuman-composedvs.computer-composedmusicandtorecognizeactualBach
and Chopin vs. computer-composedmusic in their styles.
EarlyExamples
The figured bass, popular during the Baroque period of music history (1600–1750),
demonstrates how composers and performers use combinations of notated music,
period style constraints, and performer choice to produce a diversity of results and
yet adhere to a composer’s style. As in other examples of virtual music, each perfor-
mancedi¤ers,yet each retains its stylistic integrity.
Figured basses constitute the notation for most Baroque basso continuos, the
combinationofakeyboardinstrument(clavierororgan)andareinforcingsustaining
instrument (bass gamba, violoncello, or bassoon). Typically the keyboardist uses a
notated bass line, or a bass-line and treble-line depending on the ensemble require-
ments, above whichthey freely but stylistically improvise.
Figure1.1showsaverysimplefiguredbassinCmajor.Thearabicnumeralsbelow
certain notes indicate inversions of chords. Performers assume root position or 5/3
intervals above the bass-note unless otherwise instructed. The bass gambist or cellist
plays the line as written. The keyboardist, however, must complete the implied
chords in the proper key in a manner consistent with the style, yet original in spirit.
In essence, the figured bass represents an algorithm or recipe, the realization of
which depends upon the application of performance practice and performer style
improvisation.
Figure 1.2 provides a very simple realization of the figured bass of figure 1.1. The
chords here consist of triads (three-note chords built in thirds) or seventh chords
(four-note chords built in thirds) with some notes doubled in octaves. The Baroque
period constraints governing which notes should be doubled, as well as how notes
should move, one to another, are quite strict and too numerous to present here. The
important thing, at least for our purposes, is to understand that the music in figure
1.2representsonlyoneofmanypossiblerealizationsofthefiguredbassoffigure1.1.
Figure 1.3 shows another possible correct realization of the figured bass shown in
figure 1.1. Again, the music here consists of triads and seventh chords with some
4 I. Fundamentals
Figure1.1
AsimplefiguredbassinCmajor.
Figure1.2
Onepossiblerealizationofthefiguredbassinfigure1.1.
Figure1.3
Anothercorrectrealizationofthefiguredbassshowninfigure1.1.
notes doubled. Comparing figure 1.3 with figure 1.2 demonstrates both their similar-
ity (same note names in each chord) and di¤erences (notes in di¤erent registers). In
essence, then, we have two di¤erent examples of music, in similar block chord style,
derived from the same corefigured bass.
Figure 1.4 presents a more Baroque-style realization of the figured bass in figure
1.1.Infact,themelodyshowninthisexamplemighttypicallybeoneoftheprovided
elements. While this example tends to resemble figure 1.2 in chord spacing it none-
theless represents a third and distinctly di¤erent realization of the figured bass in
figure1.1.Inallofthesecases,themusichasadheredtotheconstraintsoftheperiod
usingacombinationofgivenmusicandperformerchoice,aswellasarecombination
of rightnotes and motives.
The Musikalisches Wu¨rfelspiel
One of the first formal types of algorithms in music history, and another good
example of virtual music, is the eighteenth-century Musikalisches Wu¨rfelspiel, or
musical dice game. The idea behind this musically sophisticated game involved