Chapter 3

Topological Structures
in Computer-Aided Music Analysis

Louis Bigo and Moreno Andreatta

Abstract We propose a spatial approach to musical analysis based on the notion of a

chord complex. A chord complex is a labelled simplicial complex which represents a
set of chords. The dimension of the elements of the complex and their neighbourhood
relationships highlight the size of the chords and their intersections. Following a
well-established tradition in set-theoretical and neo-Riemannian music analysis,
we present the family of T /I complexes which represent classes of chords which
are transpositionally and inversionally equivalent and which relate to the notion of
Generalized Tonnetze. A musical piece is represented by a trajectory within a given
chord complex. We propose a method to compute the compactness of a trajectory
in any chord complex. Calculating the trajectory compactness of a piece in T /I
complexes provides valuable information for music analysis and classification. We
introduce different geometrical transformations on trajectories that correspond to
different musical transformations. Finally, we present HexaChord, a software package
dedicated to computer-aided music analysis with chord complexes, which implements
most of the concepts discussed in this chapter.

Louis Bigo
Department of Computer Science and Artificial Intelligence, University of the Basque Country
UPV/EHU, San Sebastián, Spain
e-mail: [email protected]
Moreno Andreatta
CNRS, France
IRCAM, Paris, France
Université Pierre et Marie Curie, Paris, France
e-mail: [email protected]

57
 Downloaded from www.sciencemag.org on December 28, 2
the available chords or chord types and omitting
A B C D E F singularities and other nontrivial geometrical and
topological features. Furthermore, they have been
explored in isolation, without an explanation of
how they are derived or how they relate (10). Our
model resolves these issues by describing the com-
plete family of continuous n-note spaces cor-
responding to the 32 OPTIC equivalence relations.
Of these, the most useful are the OP, OPT, and
OPTI spaces, representing voice-leading rela-
Fig. 1. Progressions belonging to the same OPT and OPTI voice-leading classes. Each group exhibits the tions among chords, chord types, and set classes,
same underlying voice-leading structure: Analogous elements in the first chord are connected to analogous respectively (4). The OP spaces Tn =S n (n-tori
elements in the second, and the distances moved by the voices are equal up to an additive constant. (A) A modulo the symmetric group) have been
iv6-V7 progression from Mozart’s C minor fantasy, Köchel catalog number (K.) 457, measures 13 and 14. described previously (9). The OPT space
(B) A progression from mm. 15-16 of the same piece, individually T-related to (A). (C) A progression from Tn−1 =S n is the quotient of an (n – 1)-simplex,
Beethoven’s Ninth Symphony, movement I, measure 102, related to (A) by individual T and uniform OPI. whose boundary is singular, by the rigid
1. Le Tonnetz : évolution,
58 applications et dérivations 13 (D) A common voice leading between fifth-related dominant-seventh chords. (E) A common voice leading
Louis Bigo and Moreno Andreatta
between tritone-related dominant-seventh chords, related to (D) by individual T. (F) A voice leading
transformation cyclically permuting its vertices
(4). The OPTI space Tn−1 =ðS n " Z2 Þ is the
between tritone-related half-diminished sevenths, related to (D) by individual T and uniform I. quotient of the resulting space by an additional

A B 048

major key
TM (k) dominant
chordV 037 036
CM(k)
tonic
chord I 027 026 025 024

P(k)
016 015 014 013 012
subdominant
IV chord
tonic
006 005 004 003 002 001 000
!
!"#$% &$! "#$%&'! ()*%&'! #*+! #,-(.*/.+! /&)#+0! #0! /1.2! #33.#&! )*! /1.! %&4)5%6+! &.3&.0.*/)*-! /1&..7
Figure 6 – La spiral array organisant les hauteurs suivant l’intervalle de quinte juste le *%/.!81%&+09!!:.&.'!/&)#+0!#&.!3#&/)8,6#&62!86%0.!/%!/1.)&!(#$%&7/1)&+!/&#*03%0)/)%*09!
long d’une spirale verticale. L’étoile représente une tonalité dont la position est donnée !
;1)*-0! 4.8%(.! 0%(.<1#/! (%&.! 8%(36)8#/.+'! 1%<.=.&'! <1.*! <.! 8%*0)+.&! /1.!
par le barycentre de ses accords
Fig. 3.1de degrés
ThreeI, IVsymbolic
et V représentés par des
spaces triangles pleins.
dedicated to musical +)08&./.!6#//)8.0!
representations: the
2 spiral 2 array (left), the
Fig. 2. (A) T /S 3 is a cone. (B) T /( S 3 " ℤ2 ) is a triangle. Numbers refer to pitch classes, with 0 = C, 1 = C♯, etc. Points represent equivalence classes of
/1#/!&.3&.0.*/!=%)8.76.#+)*-!&.6#/)%*01)30! #(%*-!5#()6)#&!+)#/%*)8!%&!
Tonnetz (centre) and a voice leading space (right) 81&%(#/)8! 81%&+09!transpositionally (A) or
! >%&! .?#(36.'! transpositionally
>)-,&.! @! &.8%&+0!and /1.!
inversionally (B) related =%)8.!
(%0/! .55)8).*/! chords. Thus, (C, D, E) and (D, E, F♯) are both instances of 024.
6.#+)*-0!
#(%*-!+)#/%*)8!5)5/10A<1)81!8#*!4.!&.3&.0.*/.+!,0)*-!#*!)&&.-,6#&'!%*.7+)(.*0)%*#6!
B)-7B#-!*.#&!/1.!8.*/.&!%5!/1.!"C4),0!0/&)3!!DE!D9!!F;1.!B)-7B#-!0..(0!/%!4.!)&&.-,6#&!
www.sciencemag.org SCIENCE VOL 320 18 APRIL 2008 347
Tonnetz comme un complexe simplicial 6 résultant de l’assemblage des triades représentées 4.8#,0.!/1.!5)-,&.!)0!+&#<*!,0)*-!/1.!81&%(#/)8!0.()/%*.!#0!#!,*)/G!<.&.!<.!/%!,0.!/1.!
+)#/%*)8! 0/.3'! )/! <%,6+! 4.! &.-,6#&9H! ! I40/&#8/62'! /1.0.! =%)8.! 6.#+)*-0! 5%&(! /1.! 8)&86.!
comme des 2-simplexes (i.e., des triangles) [Catanzaro 2011]. Cette formalisation du Ton- 01%<*! %*! /1.! &)-1/! %5! >)-,&.! @9! ! ;1.! 5)-,&.! +.(%*0/&#/.0! /1#/! /1.&.! #&.! 3,&.62!
netz permet à Catanzaro d’appliquer cette construction aux 11 autres familles d’accords à
3.1 Introduction 8%*/&#3,*/#6! &.#0%*0! /%! #00%8)#/.! 5)5/17&.6#/.+! +)#/%*)8! 5)5/10J! 5&%(! /1)0! 3.&03.8/)=.!
3 sons équivalents par transposition et inversion 7 et d’en étudier les aspects topologiques. KL'!MN!)0!86%0.!/%!KM'!ON'!*%/!4.8#,0.!%5!#8%,0/)80'!4,/!4.8#,0.!/1.!5)&0/!+2#+!8#*!4.!
/&#*05%&(.+! )*/%! /1.! 0.8%*+! 42! (%=)*-! /1.! *%/.! L! ,3! 42! %*.! +)#/%*)8! 0/.39! ! P*.!
Le chapitre 5 propose une approche simpliciale semblable pour la construction d’es- 5#08)*#/)*-!3%00)4)6)/2A<1)81!<.!,*5%&/,*#/.62!8#**%/!3,&0,.!1.&.A)0!/1#/!#8%,0/)8!
#55)*)/).0!#8/,#662!!"#$%"&'#()!=%)8.76.#+)*-!5#8/0J!)/!)0!3%00)46.!/1#/!/1.!.#&!#00%8)#/.0!
paces d’accords, en fournissant en plus une méthode permettant la généralisation et la
This work concentrates on the harmonic aspect of musical sequences and introduces
systématisation de la construction de complexes, associés à tout type d’accord.
/1.!/1)&+!1#&(%*)8!%5!#!8%(36.?!/%*.!<)/1!/1.!0.8%*+!1#&(%*)8!%5!#*%/1.&!/%*.!#!5)5/1!
#4%=.! )/'! #*+! /1.! 5%,&/1! 1#&(%*)8! %5! /1.! 6%<.&! *%/.! <)/1! /1.! /1)&+! %5! /1.! ,33.&'! )*!
a spatial approach for its analysis. Musical analysis often requires specific tools .55.8/!/&#8Q)*-!=%)8.76.#+)*-!&.6#/)%*01)30!#(%*-!/1.!3#&/)#609!
>)-,&.0! R7S! 3&.0.*/! /1&..! #*#6%-%,0! 0/&,8/,&.0J! >)-,&.! R! 8%**.8/0! /&)#+0! )*! /1.! L!
when it focuses on a particular musical layer. For example, some theorists represent+)#/%*)8! 08#6.! 42! .55)8).*/! =%)8.! 6.#+)*-'! #*+! +.3)8/0! /1)&+7&.6#/.+! /&)#+0! #0! 4.)*-!
3#&/)8,6#&62!86%0.G!>)-,&.!T!01%<0!/1.!3%0)/)%*!%5!(#$%&'!()*%&'!#*+!#,-(.*/.+!/&)#+0!
1.2.3 La Spiral arraymusical objects and their harmonic properties by (or in) symbolic spaces. When used )*! /1&..7*%/.! 81&%(#/)8! 81%&+! 03#8.'! <1.&.! (#$%&7/1)&+*&.6#/.+! /&)#+0! #&.! 86%0.RG!
>)-,&.!S!01%<0!F02(4%6)8#662H!/1#/!5)5/17&.6#/.+!+)#/%*)8!08#6.0!#&.!86%0.!)*!81&%(#/)8!
in an analytic context, these spatial representations can reveal some strategies used
Le modèle connu sous le nom de spiral array [Chew 2000] a été développé par Elaine
03#8.9! ! P*8.! #-#)*'! <.! 0..! /1#/! /1.&.! #&.! 3,&.62! 8%*/&#3,*/#6! &.#0%*0! /%! #00%8)#/.!
5)5/17&.6#/.+!+)#/%*)8!08#6.0!#*+!/1)&+7&.6#/.+!/&)#+09!!!
Chew avec pour objectif deto représenter
compose dans un amême
piece.espace géométrique
The spiralla perception
array (Chew, des 2002), the Tonnetz (Cohn, 1997) and
hauteurs, des accords et de la tonalité au cours de l’évolution d’une œuvre. Le modèle peut
voice-leading spaces (Callender
se construire de la même manière que le Tonnetz mais sans l’hypothèse du tempérament et al., 2008) are examples of such spaces. Among
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
égal. Le Tonnetz ne se plietheirdonc que dans un sens, aboutissant
numerous properties, à une structure
they are cylindrique
well adapted R
for determining key boundaries,
!;1)0!-&#31!<#0!5)&0/!+)08%=.&.+!42!O%,/1.//!#*+!U/.)*4#81!FVWWXH9!!!
et non toroïdale (figure 6). La spirale externe correspond à l’organisation des hauteurs en
representing neo-Riemannian
suivant l’intervalle de quinte juste. Deux hauteurs situées l’une au-dessus de l’autre (dûoperations and voice-leading motions, respectively.
à la rotation de la spirale) sont séparées par une tierce majeure. Les triades majeures et
Such spaces typically represent
mineures peuvent être représentées par des triangles sectionnant la spirale. En considérant
pitches or chords by vertices in a graph.
respectivement les centres de ces The chord
triangles, spaces
on obtient deuxpresented in this study
spirales supplémentaires. Enfin, include elements of higher dimension than
deux dernières spirales, représentant les tonalités majeures et mineures, sont constituées en
vertices and edges. We represent n-note chords by geometrical objects of dimension
calculant le barycentre des accords de degrés I, IV et V de chaque tonalité. La spiral array
a été utilisée pour différentes(n − applications,
1), called dont entre autres la recherche
simplices. The faces automatique
of adesimplex represent all sub-chords contained
tonalité [Chew 2002].
in the chord. Chord sets are represented by simplicial complexes. The possibly high
6. Les complexes simpliciauxdimension ofleachapitre
seront présentés dans complex3. highlights specific neighbourhood relationships between
7. Cette propriété sur les familles d’accords sera présentée plus en détail dans le chapitre 5.
chords and allows the space to represent more advanced musical properties.
A strong motivation of this work is the general desire to represent a collection of
musical objects by a symbolic space governed by a set of neighbourhood relationships
that reflect how represented objects interact. This general idea is inspired by the MGS
project (Giavitto and Michel, 2001), which aims to provide tools for the modelling
and simulation of (not necessarily musical) dynamical systems exhibiting dynamical
structures.
Section 3.2 begins by briefly presenting the historical background of music repre-
sentations, including a discussion of Pousseur’s pioneering attempt to capture musical
logic with geometric representations. We then present the notions of T/I class and
Generalized Tonnetze. Section 3.3 introduces the notion of simplicial complex and
explains how this concept allows for any set of chords to be represented by a symbolic
space called a chord complex. This section finally introduces T /I complexes which
are chord complexes representing chords of a T/I class. Section 3.4 introduces the idea
of representing musical sequences as trajectories in chord complexes. We introduce
3 Topological Structures in Computer-Aided Music Analysis 59

the notion of the d-compactness of such a trajectory as a heuristic for selecting chord
complexes for analysis or classification. We finally present a method to transform
a musical sequence by applying some geometric transformations on its trajectory.
Chord complexes and trajectories have been implemented in the HexaChord music
analysis software, which is presented and discussed in Sect. 3.5.

3.2 Background in Musical Representations

We present two well-known notions in music theory: the Tonnetz and T/I classes.
The first is a spatial organization of pitches that constitutes a well-established analyt-
ical tool in the so-called neo-Riemannian music-theoretical tradition (Cohn, 2012;
Tymoczko, 2012). The second provides a classification of musical chords based on
their equivalence up to transposition and inversion. These two representation tools
constitute the musical starting point of this work.

3.2.1 The Tonnetz

One of the strongest motivations of this work is the wish to formalize a widely
used tool in music theory, analysis and composition: the Tonnetz. The Tonnetz is
a symbolic organization of pitches in the Euclidean space defined by infinite axes
associated with particular musical intervals. It was first investigated by Euler (1739)
for acoustical as well as graph-theoretical purposes1 and rediscovered later by the
musicologists A. von Oettingen and H. Riemann and by the composer and music-
theorist H. Pousseur. More recently, music theorists have shown a strong interest in
this model, in particular to represent typical post-romantic chord progressions (Cohn,
2012) currently called neo-Riemannian transformations. This model has also been
used in musical composition, not only in contemporary musical styles (Chouvel,
2009) but also in more popular styles (Bigo and Andreatta, 2014).
The neo-Riemannian Tonnetz (on the left in Fig. 3.2) is a graph in which pitches
are organized along the intervals of the perfect fifth (horizontal axis), major and
minor thirds (diagonal axes). This representation has the interesting property that
major and minor triads appear as triangles. Many theorists have investigated different
derivations of the Tonnetz, often referred to as generalized Tonnetze. For instance,
the three-dimensional Tonnetz introduced by Gollin (1998) is shown on the right in
Fig. 3.2. This model corresponds to the one on the left side of Fig. 3.2 with some

1 Euler’s Tonnetz (Speculum Musicum) organizes pitches in just intonation along horizontal and
vertical axes associated with the pure fifth and pure major third, respectively. Interestingly, the
kind of music-theoretical problems he suggested could be approached via his Speculum Musicum
deeply resonate with purely mathematical problems, such as the Königsberg Bridge Problem. This
fact supports the idea that not only does mathematics apply to music but, conversely, music often
anticipates some developments of mathematics (Andreatta, 2008).
60 Louis Bigo and Moreno Andreatta

3
4
B F♯ C♯ G♯

7
D A E B

Bb F C G
7 9

Db Ab Eb Bb

Fig. 3.2 On the left, a region of the neo-Riemannian Tonnetz. On the right, a three-dimensional
derivation of the Tonnetz by Gollin (1998)

additional interval axes. Tetrahedra represent dominant seventh and half-diminished

chords. Three-dimensional models are well adapted to the study of 4-note chord
progressions. Similar structures can be built by associating axes with intervals that
are diatonic instead of chromatic. In these diatonic Tonnetze, investigated by Cohn
(2012), vertices and shapes only represent notes and chords belonging to a unique
diatonic scale.
In this work, we limit ourselves to the context of equal temperament and octave
equivalence, i.e., we are dealing with pitch classes. For example, the pitches C]3, C]4
and D[4 are all members of the same pitch class. In particular, what we call Tonnetz
more exactly refers to the pitch-class Tonnetz. In this context, the graph on the left of
Fig. 3.2 repeats infinitely the 12 pitch classes along its axes. The pitch-class Tonnetz
is frequently represented as a toroidal structure by merging the nodes representing
the same pitch classes (Cohn, 1997).
Interestingly, this geometrical structure has also been rediscovered independently
from the neo-Riemannian music-analytical tradition by the composer and music
theorist H. Pousseur, one of the leading figures, with P. Boulez and K. Stockhausen,
of European integral Serialism. Although Pousseur’s model is much less well-known
and rarely cited in neo-Riemannian music analysis, it contains the roots of the spatial
approach that we present in this chapter. The point of departure of Pousseur’s network
theory is Rameau’s (1722) Traité de l’harmonie, in which the latter introduces a
bidimensional representation of the tone-space in just intonation as generated by the
pure fifths (horizontal axis) and pure major thirds (descending vertical axis). This
representation is equivalent to the Speculum Musicum which was introduced by Euler
a few years later. Pousseur first presented his Rameau-inspired geometrical model in
1968 (Pousseur, 1968), but he provides several analytical examples of what he calls
his “Network technique” in a more recent paper (Pousseur, 1998). In this analytical
essay he goes back to the definition of a Network as a spatial distribution of pitches
(and pitch classes) according to at least two axes which are characterized as a chain
3 Topological Structures in Computer-Aided Music Analysis 61

Fig. 3.3 The pitch network introduced by H. Pousseur in the late 1960s and accompanied by a
triangulation showing its equivalence to the traditional Tonnetz

of a single interval. It is easy to show that his pitch distribution is equivalent to

the traditional Tonnetz structure as generated by the fifths and minor thirds axes, as
shown in Fig. 3.3. As clearly pointed out by Pousseur, this geometrical representation
is one of the possible network configurations which can be utilized in music analysis.
The basic principle of the Network technique, as he claims, resides in choosing the
generating axes, and hence constructing the triangulations of the space, in such a
way that the musical relations between adjacent pitches (both harmonic and melodic)
produce the most compact configuration within the given space. This clearly suggests
that his Network technique is deeply rooted in a spatial approach to music analysis,
which we can properly formalize in terms of simplicial complexes as presented
in Sect. 3.3. Moreover, as Pousseur suggests, the music analyst can transform the
underlying geometrical space and move from a given network to a different one by
simply changing the generating axes, which modifies the hierarchy of intervallic
relations as well as the structural proximity between the pitches (or pitch classes).
The next two sections will show how to transform this original compositional and
music-theoretical intuition into a computational analytical model.
62 Louis Bigo and Moreno Andreatta

3.2.2 Generalized Tonnetze and T/I Classes

The highlighting of particular chords (minor/major chords as triangles in the neo-

Riemannian Tonnetz and dominant seventh/half diminished chords as tetrahedra
in the three-dimensional Tonnetz) suggests the idea that the starting point of the
construction of a Tonnetz could be a set of chords rather than a set of interval axes. In
the two examples above, the represented chords in a Tonnetz are all transpositionally
and inversionally equivalent—that is, they belong to the same T/I class (Forte, 1973).
This property comes from the repetition and the invertibility of the intervals on the
axes.
It is common to identify a T/I class by the intervallic structure which is shared by
all the chords of the class. For instance, the 24 major and minor chords all share the
intervallic structure [3, 4, 5] because the row of intervals between the pitch classes
they are resulting from is composed of a minor third (3 semitones), a major third (4
semitones) and a fourth (5 semitones). This notation of the intervallic structure is
defined up to reflection and circular permutation. Intervals are ordered in the opposite
direction (up to a cyclic permutation) for major ([4, 3, 5]) and minor chords ([3, 4, 5]).
Intervallic structures can be diatonic as well. In this case, the intervals add up to
7 instead of 12. For example, the triads on the seven degrees of the diatonic scale
belong to the class identified by the intervallic structure [2, 2, 3]. Even though T/I
classes can be enumerated for any division of the octave (i.e., for any N), we focus in
this work on the diatonic (N = 7) and chromatic (N = 12) systems.2
There exist 224 such classes in the chromatic system, also known as Forte
classes (Forte, 1973). In the diatonic system, which divides the octave into seven
(non-equal) parts, there exist 18 such classes. Following this line, we are interested
in building the generalized Tonnetze associated with the 224 T/I chromatic classes
and 18 T/I diatonic classes.

3.3 Generalized Tonnetze as Chord Complexes

This section presents a method to represent any arbitrary set of chords by a multi-
dimensional structure called a chord complex. Figure 3.7 illustrates an example of
a chord complex representing a set of four chords. A chord complex is a simplicial
complex whose components, called simplices, are labelled by chords.3 In a chord com-
plex, the dimension of the simplices represents the size of the chords. Furthermore,
the intersections of the simplices represent the common pitch classes of the chords.
Chord complexes representing the chords of a T/I class are called T/I complexes and
relate to Generalized Tonnetze.

2 For enumeration and classification purpose, T/I classes can be associated with the orbits of the
action of the dihedral group DN on the subsets of ZN , as described by Andreatta and Agon (2003).
3 Simplicial complexes belong to a more general family of spaces called cellular complexes, which

have already proved to be useful in music theory (Bigo et al., 2011).

 3 Topological Structures in Computer-Aided Music Analysis 63

0-simplex 1-simplex 2-simplex

2-s implex
Fig. 3.4 From left to right: a 0-simplex, a 1-simplex, a 2-simplex
Figure 1. Some simplices. 1-simplex and a 3-simplex es.
represented with
their closure simp c
li
0-simp
lex
1. Some
Figure plices
e d b y its sim
form
complex K is defined 3.3.1by the Simplicial
sub-complex Complexes Sn (K) of this complex formed byhiits mplex
s cosimplices leton
K ) o f t
a nd i t s 1-ske
mplthe ex S n ( left)
of dimension n or less. The figure 2 illustrates ah
t sub-co(onspace
complex
e left) and (on the more
itspl1-skeleton
ex together
A simplicial complex
db y
is a multidimensional built
sa
by
c o mgluing
elementarylespaces x definesimplices.
K is called A simplex 2 illu(morestrateprecisely a d-simplex) is the
(on the center). m p fi g u r e
co s s. T he d. As illustrated in Fig. 3.4, a 0-simplex corre-
abstraction of a space of
n or l e
dimension
sponds ensiona 1-simplex corresponds to an edge, a 2-simplex to a triangle, a
diamvertex,
of to
.
center)
3-simplex to a tetrahedron, etc. In a simplicial complex, a d-simplex is necessarily
Simplicial collections
incident (ontothde+ 1 simplices of dimension d − 1. For instance, a 1-simplex (i.e., an
ed by
is label
edge) is incident to two 0-simplices (i.e., 2 vertices), a 2-simplex (i.e., a triangle) is
o n s p le x
incident to three 1-simplices cti 3 edges), etc. This recursive property defines
le(i.e.,
l colcomplex every s
the i m
A simplicial collection ofiis l i ciaconsisting
apsimplicial of allinitswhich every simplex isx labeledwhichby In a ection
closure K
S amsimplex, incident simplices
c o
ofplower
m l e i ndimension.
o p o lo g ical coll
cial
impliconstitutes ion of t ongly
the ot
simplicial complex, a subset of simplices thatsitself a simplicial ncomplex
an arbitrary value. The term “collection” comes from
io n K is anotion
the of topologicalrocollection
f m c h has str
is called a sub-complex. l collect o m e s w h i
ia
simplic Giavitto on” c the highest dimension
tiwhere (2001)
A simplicialA
used in the MGS programming language
d-complex and
is a simplicial
t e r m “collec(2001)
Michel
complex which has o a n d Michel of
strongly
Th e vit
Gianatural t
luae.simplicial
rary vacomplex K is defined ng ge
any simplex is d. A graph is 1-complex.ua For any number n, the
inspired this work.n-skeletonaof n a r b i t i n g l a
a simplicial r a m m by the sub-complex Sn (K) of this
M G S progof dimension n or less. Figure 3.5 illustrates a an
complex formed iby the
its simplices
es from
used n and its 1-skeleton. e s v a l u
simplicial 3-complex k.
waorfunction sociat
is is
thsimplicial that as es to
More formally, a The simplicial collection
f -vector
in s p i r
ofe da d-complex that associates
corresponds values
to
i s
the
a f
from
u n c t
finitei an
o nsequence
n K ( ) enabl
tion io
( f1 , . . . , fd+1 ) where f p+1 is the number of p-simplices l collec included in the complex. e notat pport
arbitrary set with the simplicies of a simplicial complex. s i m p licianotation
The K( ) enables
c o m p lex.11Thver-
to
e |K | the su
The f -vector of the complex inalFig. l y, a 3.5 is (11, 17, 7, 1) because i c i a l
it includes e n o t
ore foin rm of adenotesimpl K. We
d
address the label associated with theMcell the collection s i m p licK.iesWe |K| thee supportc o l l ection c tion K
0 is a

it h t h e i n t h c o l l e
ry set w cell label. A
arbitra complex without with A thecollection K0leisx awithout uity is
of the collection K which is the simplicial
s s o c i a t e d label.
m p e5!n no ambig
el a c o h
the lab plicial 0!f K“e.uW
0 ” !
x”
sub-collection of K if |K0 | ⇢ |K| and adK d0r(es)s = K( ) for every i c h i softhK e 0s.im
When no ambiguity fo r e v
isery o “b
e
”r!m “comple
w h K ( ) t
se “sr”!h e t
n K
ftebnc”u
ollectio
0( ) = “ 21! 3!
possible, the notation | · | will be omitted. of the c Similarly, we|K will
0 | ⇢often|K| an d Kthe
use term “complex” , w e will o !
ls.
o f K if e d . S i m ila r ly
x a n d its labe
ection tt comple
“vs”!

b-coll be omi plicial

to designate a simplicial collection,suthat · | will and its labels.
”!

is the simplicialn complex s i m

“a

ta t i o | t h e
, the no , that is
“Z”!
s s ib l e l e c t i o n 9!
po col 35!
a s i m plicial “or”!
gnate
to desi
Fig. 3.5 A simplicial 3-complex (on the left) and its 1-skeleton (on the right)

4 4
64 Louis Bigo and Moreno Andreatta

A A A
1! 2!
{ D , A , Bb} A D C
C D

{D,A} { Bb, A } B B A
B

3! C D
D Bb
{ D , Bb}
B

Fig. 3.6 On the left, the 2-simplex S(C) representing the chord C = {D, A, B[} and all sub-chords
and notes included in it. On the right, the identification of boundaries illustrating the self-assembly
process on the 2-simplices representing the chords {C, A, B} and {D, A, B}

tices, 17 edges, 7 triangles and one tetrahedron. The term f p of the f -vector of a
complex K is denoted by f p (K).
Simplicial collections are simplicial complexes in which every simplex is labelled
by an arbitrary value. The left side of Fig. 3.6 illustrates a simplicial collection of
dimension 2. A simplicial collection can be built from a set of labelled simplices
by applying a self-assembly process (Giavitto and Spicher, 2008). This process is
based on the identification of the simplex boundaries that share the same labels. This
topological operation holds in all dimensions. The right side of Fig. 3.6 illustrates
the process on two 2-simplices. In step 1, nodes A and B are merged. Then, in step
2, the resulting edges {A, B} are merged. The final structure is a connected, two-
dimensional, simplicial collection in which every simplex is labelled with its own
value (step 3).

3.3.2 Chord Complexes

We use a method described by Bigo et al. (2011) to represent a set of chords by a

simplicial collection that we call a chord complex. In this representation, chords are
reduced to pitch class sets. This requires some abstraction since some properties of
chords as they actually occur within a musical context (e.g., the octave and duration
of each note) are not represented. An n-pitch class set (i.e., a set of n pitch classes) is
represented by an (n − 1)-simplex. In particular, a 0-simplex represents a single pitch
class, a 1-simplex a 2-pitch class set, etc. We denote by S(A) the simplex representing
the pitch class set A. The simplices constituting the closure of S(A) represent all
the sub-chords of A. The 2-simplex representing the pitch class set {D, A, B[} is
illustrated on the left of Fig. 3.6. A chord complex is built by:
1. representing each chord of the collection by a simplex as described above,
 test
3 Topological Structures in Computer-Aided Music Analysis 65

C
! ! ! ! ! !
" 44 $$$ %%% &&& $$
$
%% &&
%&
$$
$
%% && ' $$ %% &&
%& $ %& B
B
3

# 4 $$$ %& $$ %& $$ %& ' $ %&

D
%& %& %&
Piano
%& %& %& $$ %& ! ! ! ! ! !
4

$ $ %&
2

4 1
G
1 2 3 4
E
" 3.7!First measures
! ! ! ! ! ! ! ! ! ! ! !
11

Fig. of Metamorphosis by Philip Glass (left) represented by a chord complex

(right). The three arrows provide the order in which the four chords are played
Pno.
# ! ! ! ! ! ! ! ! ! ! ! ! !

2. applying the self-assembly process to the resulting collection of simplices.

! ! ! ! ! ! ! ! !
" method ensures that a given chord cannot be represented more than once in the
24
This
simplicial complex. As an example, on the right in Fig. 3.7 is shown the complex
Pno.
# ! ! Philip Glass,
! ! on the
! left in the
! same !figure. The
! four chords
!
resulting from the assembly of the four introductory chords of the piano piece
Metamorphosis by shown
are respectively associated with the pitch class sets {E, G, B}, {D, G, B}, {C, G, B}
and {C, E, B[}. Each chord includes three pitch classes and is represented by a 2-
simplex, that is, a triangle. The whole complex exhibits the intersections between the
four pitch class sets.
This representation method provides a topological signature for any chord se-
quence. The musical interpretation of topological properties of chord complexes
(dimension, size, connectedness, morphisms, etc.) is discussed by Bigo (2013). In
the following, we will focus on chord complexes representing sets of chords related
by some theoretic properties (for example chords belonging to the same T/I class)
rather than sets of chords included in a given musical sequence as illustrated by the
example in Fig. 3.7.

3.3.3 T/I Complexes

We represent a generalized Tonnetz as a chord complex composed of n-simplices

representing the chords of a given T/I class. In the following, we consider in particular
KTI [a1 , . . . , ai ], the complex associated with the T/I class identified by the intervallic
structure [a1 , . . . , ai ]. To build such a complex, we first enumerate all the chords
belonging to the T/I class, we then build the corresponding complex, as explained in
Sect. 3.3.2. As a consequence, the dimension of a T/I complex corresponds to the
size of its intervallic structure minus one.
Figure 3.8 illustrates the diatonic complexes KTI [2, 5], KTI [2, 2, 3] and
KTI [1, 2, 2, 2] associated with the scale of C major. These complexes were described
by Mazzola et al. (2002) as interpretations of the diatonic scale. They have the
topology of a circle, a Möbius strip and a torus, respectively. An alternative represen-
66 Louis Bigo and Moreno Andreatta

G
E B
!! !!
" # ! !! !! !! !!
!
C D
{C,E} {D,F} {E,G} {F,A} {G,B} {A,C} {B,D}
A F

!!!
E B F C
!!! !!! !!!
" # !!! !!! !
!!
C G D A E
CM Dm Em F G Am B°

!! !! !!! !!
# " !!! !! !! !! !! ! !!
! !! !!
CM7 Dm7 Em7 FM7 G7 Am7 Bm°7

Fig. 3.8 Construction of the diatonic complexes KTI [2, 5], KTI [2, 2, 3] and KTI [1, 2, 2, 2] by assem-
bling degrees of the tonality of C major (Mazzola et al., 2002)

tation of the complex KTI [1, 2, 4] in which all pitch classes of the diatonic scale are
neighbours has been used for music analysis by Hook (2014).
Figure 3.9 illustrates the chromatic complexes KTI [2, 5, 5] and KTI [3, 4, 5] which
result from the assembly of suspended chords (Sus4 chords) and major/minor chords
respectively. The organization of pitch classes in KTI [2, 5, 5] corresponds to the well-
known Wicki–Hayden note layout, which is used for the key layout on some keyboard
instruments, such as the bandoneon. The 1-skeleton of KTI [3, 4, 5] corresponds to
the Tonnetz. Topological properties of T/I chord complexes of dimension 2 were
studied by Catanzaro (2011). Although the present study focuses on T/I complexes
(i.e., KT I [·]) due to their relation with Generalized Tonnetze, this approach can be
generalized to enumerate chord complexes defined by any equivalence relation, not

C

C

D
E

Music engraving by LilyPond 2.14.1—www.lilypond.org
B♭

G

F
A
F
A

B

E

A♭
E♭

Music engraving by LilyPond 2.14.1—www.lilypond.org C♯

G

F♯

B

C♯
F♯

B♭

Music engraving by LilyPond 2.14.1—www.lilypond.org
Fig. 3.9 On the left, the complex of the suspended chords (KTI [2, 5, 5]) of size 12. On the right, the
complex of major and minor chords (KTI [3, 4, 5]) of size 24, whose 1-skeleton corresponds to the
Tonnetz. The first complex is a strip and the second is a torus
3 Topological Structures in Computer-Aided Music Analysis 67

just transposition and inversion. For example, complexes of chords equivalent up

to transposition only (KT [·]), up to transposition and interval permutation (KT P [·])
and up to affine transformation (KM [·]) have been enumerated and classified in a
catalogue (Bigo, 2013).4

3.3.4 Unfolded Representations of T/I Complexes

The neo-Riemannian Tonnetz was originally designed as a triangular graph embedded

within a two-dimensional Euclidean space, as illustrated on the left in Fig. 3.2.
One may imagine the triangulation as being generated by the choice of three axes
representing three privileged intervallic directions starting from a referential initial
point. In the case of Fig. 3.2 (left), we start from the note C and we consider three
axes corresponding to the minor third, perfect fifth and major third intervals. Note
that two axes suffice to generate the triangulation of the plane, since the interval
corresponding to one of the axes is simply the sum of the intervals corresponding
to the two remaining axes. In our case, for example, the horizontal axis, going from
left to right, represents a fifth interval which is the sum of a minor third and a major
third, corresponding, respectively, to the two diagonal directions within the graph.
This property is easily generalized to the case of the three-dimensional Tonnetz, as
represented on the right side of Fig. 3.2. In this case, three axes are necessary to
generate the three-dimensional model and they correspond to the minor third, major
sixth and perfect fifth intervals. Note that, in this case, the reference point is G, as
indicated on the right side of Fig. 3.2. In both representations, pitch classes are
replicated infinitely along the axes by applying successive transpositions of the same
interval.
These spaces can equivalently be thought of as the result of two T/I chord com-
plexes which have been unfolded in a Euclidean space. Starting from a given
n-note chord considered as belonging to a T/I class, one can first represent it by
an (n − 1)-simplex. The simplex is then embedded in an equilateral manner in the
(n − 1)-dimensional Euclidean space. For a 3-note chord, this space is the Euclidean
plane and the chord is embedded as an equilateral triangle. The directions given to
the 1-simplices (i.e., edges) define axes associated with particular intervals, as in
the Tonnetz. Then, simplices are naturally replicated along these axes, in such a way
that the represented chords respect the transpositions induced by the intervals of
the axes. By starting with a major chord and a dominant seventh chord, one obtains
the complexes KTI [3, 4, 5] and KTI [2, 3, 3, 4] which are then unfolded as shown in
Fig. 3.10. A consequence of this generic method of construction is that two com-
plexes associated with chord classes of the same size are represented by identical
structures (they are said to be isomorphic). Figure 3.10 shows the unfolding of the

4 This “paradigmatic approach”, where groups act as “paradigms” in the enumeration and clas-
sification of musical structures, corresponds to well-known catalogues of chords (respectively,
the Vieru–Zalewsky catalogue of transpositional chord classes, Forte’s pitch class sets catalogue,
Estrada’s permutohedron and Morris–Mazzola’s affine orbits catalogue).
68 Louis Bigo and Moreno Andreatta

B F# C# Ab Eb EB F#
B C#
F# Ab
C# Eb
Ab

D A E B DD AA EE BB

Bb F C G D Bb CF GC G
D D
F A

C# Ab Eb Bb C#
Eb Ab
Bb Eb
F Bb
C

Fig. 3.10 Unfolded representations of 2-dimensional chord complexes KTI [3, 4, 5] (left) and
KTI [2, 3, 7] (centre) as isomorphic structures. On the right, the unfolded representation of the
3-dimensional chord complex KTI [2, 3, 3, 4] composed of dominant seventh and half-diminished
chords

complexes KTI [3, 4, 5] and KTI [2, 3, 7], both corresponding to a two-dimensional
infinite triangular tessellation.

3.4 Analysis and Transformation of Trajectories in T/I

Complexes

In this section, we represent musical sequences by trajectories in T/I complexes.

We first perform a harmony-based analysis and classification by studying the shape
of these trajectories. We then present a variety of spatial transformations on these
trajectories which result in musical transformations on the represented sequences.

3.4.1 Representation of a Piece in a Chord Complex

In this work, a trajectory corresponds to a temporal sequence of regions in a chord

complex. Each successive region represents a temporal segment of the piece. We
use a very simple segmentation method, based on appearance and disappearance of
pitch classes. Each time a pitch class enters or leaves the set of played notes, the
current segment stops and a new one begins. This principle is illustrated in Fig. 3.11.
Note that a musical sequence could be represented by a trajectory constructed using
any other segmentation process, including more sophisticated algorithms involving
automatic harmonic analysis. Each segment is characterized by its relative duration,
compared to the other segments. A musical sequence P is thus reduced to a sequence
of pitch class sets, each associated with a relative duration. We thus have P =
[(A0 , d0 ), . . . , (AN , dN )] where Ai is the set of active pitch classes during a duration di .
We represent the chord A in the chord complex K by the sub-complex KA of K
consisting of all the simplices in K which are labelled by a pitch class set included in
A. A trajectory in a chord complex K is a sequence of sub-complexes of K, each of
3 Topological Structures in Computer-Aided Music Analysis 69

1 1/2
)
1/2

'
1/2
)
1/2

'
1 di

! "
! ! ! ! ! ! !
$ % !* ! !
! ! ! ! ! ! ! #!
31.
"
…

&% ! +
( ! ! !
{C,A} {A} {∅} {C,E,G} {G} {E,G,B} Ai
!
"
! ! ! ,,
3

$ ! ! ! ( !
Fig. 3.11 Segmentation of a sequence depending on the set of played pitch classes. The duration of

!
! !
a segment (grey box) is calculated with respect to the quarter note taken as a unit
! -! ! #! !
& is associated !with a duration. The musical sequence [(A0, d0), . . . , (AN , dN )] is " ,,
! [(K0, d0!), . . . , (KN!, dN )]. !
which
! in a complex K by a trajectory !
Figure 3.12 shows the sub-complexes constituting the trajectory of the sequence!
represented
!
in Fig. 3.11 in KTI [3, 4, 5]. For more clarity, the trajectory is represented in the
"
! ! ! ! ! ! ! ! ! !
$
unfolded representation of the complex. Note that the 1-skeleton of this representation

"
corresponds to the Tonnetz depicted on the left of Fig. 3.2.
-! ! ! ! ! ! ! #! !
!
Figure 3.13 shows the representation of the chorale, BWV 326, by J. S. Bach, by

"
a trajectory in the T/I complexes KTI [3, 4, 5] and KTI [1, 4, 7]. The trajectory on the
&
left constitutes a shape that is more compact than the trajectory on the right. This

! ! ! ! ! !
! ! !
observation suggests the idea that the different T/I complexes might not be equally

! ! ! ! !
!B "
! ! !
7
! ! !
F# C# Ab Eb
$ ! ! ! !
#
! -! #! ! ! ! ! #! ! !
D A E
! B

& "
#! ! ! ! #! ! ! ! ! ! #!
"
Bb F C G D

! ! !
9

$ ! ! ! ( !
! !C# # ! Ab- ! ! Eb ! Bb ! #! !
!
& "
! # !in K [3,# !4, 5]. Note!thatconstituting
Fig. 3.12 Union (in dark grey) of the subcomplexes
- ! a -trajectory
! A and Crepresenting
!is also included
!
the
!
sequence illustrated in Fig. 3.11
in the union
TI the edge between
! !

Music engraving by LilyPond 2.16.2—www.lilypond.org

70 Louis Bigo and Moreno Andreatta

Fig. 3.13 First measures of the chorale, BWV 326, by J. S. Bach, represented as a trajectory in
KTI [3, 4, 5] (on the left) and in KTI [1, 4, 7] (on the right)

well adapted to representing and analysing a particular musical sequence. In the next
section, we propose a method for computing the compactness of a trajectory, in order
to estimate how well a complex is adapted to representing a given musical sequence.

3.4.2 Computing Trajectory Compactness

In this section, we propose a method for computing the compactness of a trajectory

representing a musical sequence in a chord complex.

3.4.2.1 Chord Compactness

Let K be a chord complex and KA the sub-complex of K representing a pitch class

set A. We define the compactness at the dimension d (or the d-compactness) of A in
K to be
fd+1 (KA )
Cd (K, A) = .
fd+1 (S(A))
In other words, the d-compactness corresponds to the number of d-simplices included
in the representation of A in K divided by the number of d-simplices included in
the simplicial representation of A.5 Note that the value of the d-compactness varies
between 0 and 1 whatever the dimension d is. The value chosen for the parameter

5 This ratio recalls the global clustering coefficient (Holland and Leinhardt, 1971) in graph theory,

which measures the degree to which nodes in a graph tend to cluster together. Our approach is more
general because it applies to a set of arbitrary elements of the space (not necessarily neighbours of
3 Topological Structures in Computer-Aided Music Analysis 71

B
E

Fig. 3.14 The pitch class set A = {C, E, G, B} represented by the 3-simplex S(A) (left), by a sub-
complex of KTI [2, 3, 7] (centre) and by a sub-complex of KTI [3, 4, 5] (on the right)

d in the computation of the compactness depends on the type of study which is

being performed. 1-compactness will be calculated for a study based on intervals. If
the study relates to the inclusion of chords of size 3, one may prefer to work with
2-compactness.
As an example, let us compute the 1-compactness of the pitch class set
A = {C, E, G, B} in KTI [2, 3, 7] and KTI [3, 4, 5]. As explained in Sect. 3.3.2, a 4-
pitch class set is represented by a 3-simplex which includes six 1-simplices (i.e.,
f2 (S(A)) = 6). As Fig. 3.14 shows, the representation of A in KTI [2, 3, 7] includes
three 1-simplices, whereas its representation in KTI [3, 4, 5] includes five 1-simplices.
We thus have C1 (KTI [2, 3, 7], A) = 3/6 = 0.5 and C1 (KTI [3, 4, 5], A) = 5/6 = 0.83.
Analogously, by focusing on 2-simplices, we have C2 (KTI [2, 3, 7], A) = 0/4 = 0 and
C2 (KTI [3, 4, 5], A) = 2/4 = 0.5. These results show that the representation of A in
KTI [3, 4, 5] is more compact than in KTI [2, 3, 7] (as can be seen in Fig. 3.14).
Figure 3.15 illustrates the 2-compactness of four types of chords commonly used
in tonal music: major, major seventh, dominant seventh and diminished seventh.
This diagram shows that the compactness of a chord can be more or less equally
distributed among a set of different complexes. For example, the equal compactness
of the dominant seventh chord in KTI [2, 3, 7], KTI [2, 4, 6], KTI [3, 3, 6] and KTI [3, 4, 5]
shows the harmonic diversity of this chord. On the other hand, the diminished seventh
chord is represented with strong compactness in KTI [3, 3, 6] only. This is due to the
fact that the 3-note sub-chords included in this chord all belong to the T/I class
[3, 3, 6].
This type of diagram provides an original description of the constitution of a
chord and can be exploited to represent Z-relations between chords (Mandereau et al.,
2011).

the same element) and in a simplicial complex of any dimension (graphs are simplicial complexes
of dimension 1).
72 Louis Bigo and Moreno Andreatta

0,75
2-compactness

0,5

0,25

0
9]

8]

7]

6]

8]

7]

6]

5]

6]

5]

4]
]
10

2,

3,

4,

5,

2,

3,

4,

5,

3,

4,

4,
1,

1,

1,

1,

1,

2,

2,

2,

2,

3,

3,

4,
1,

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[
K[

diminished seventh chord dominant seventh chord major chord major seventh chord

Fig. 3.15 2-compactness of diminished seventh, dominant seventh, major and major seventh chords
in the twelve two-dimensional T/I complexes

Figure 3.16 illustrates the average 2-compactness of the set of all 3-pitch class
sets. The compactness is computed in the twelve two-dimensional T/I complexes.
The irregularity of the histogram shows that T/I complexes do not tend to represent
arbitrary chords with the same compactness. This property depends on the size of the
complex. For example, the probability of a random chord being represented compactly
in KTI [3, 4, 5] (which includes twenty-four 2-simplices) is six times higher than in
KTI [4, 4, 4] (which includes only four 2-simplices). The average value E(Cd (K, A))
of the d-compactness Cd of chords of size greater than or equal to d in a complex
K is equal to the number of (d + 1)-simplices in K divided by the total number of
chords of size d + 1:

0,25
2-compactness

0
9]

8]

7]

6]

8]

7]

6]

5]

6]

5]

4]
]
10

2,

3,

4,

5,

2,

3,

4,

5,

3,

4,

4,
1,

1,

1,

1,

1,

2,

2,

2,

2,

3,

3,

4,
1,

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[
K[

Fig. 3.16 Average 2-compactness of the set of all 3-pitch class sets in the twelve T/I complexes of
dimension 2
3 Topological Structures in Computer-Aided Music Analysis 73

fd+1 (K)
E(Cd (K, A)) = N

.
d+1

3.4.2.2 Trajectory Compactness

The global d-compactness of a trajectory TP can be defined as the average of the

d-compactness of its successive chords weighted by their duration di :
N
1
Cd (K, TP ) = × ∑ di Cd (K, Ki ) ,
D i=0
N
where D = ∑ di represents the duration of the whole piece.
i=0
In the following, the compactness of the trajectory of a piece in a complex will
systematically be displayed in comparison with the average compactness of chords
in that same complex.

3.4.3 Computing the Compactness of Trajectories for Musical

Analysis and Classification

We now analyse a number of pieces in terms of the compactness of their trajectories

in different T/I complexes.

3.4.3.1 Comparing Spaces Regarding a Piece

The histograms in Fig. 3.17 illustrate the 2-compactness of the trajectories of three
different pieces in the 12 chromatic T/I complexes of dimension 2. In each complex,
the compactness of the trajectory of the piece (in black) is compared to the average
compactness of its chords (in grey). The histograms have been generated with the
software HexaChord which will be described in Sect. 3.5.
The first piece is the chorale, BWV 328, by J. S. Bach . The high compactness of
the trajectory in KTI [3, 4, 5] results from the strong use of major and minor chords
which is typical of tonal music in general and Bach’s chorales in particular. The
high compactness of the trajectories in KTI [2, 3, 7] and KTI [2, 5, 5] is due to the
use of dominant seventh and suspended chords which have particular functions in
this style of music. The compactness in the complexes KTI [2, 2, 8], KTI [2, 4, 6] and
KTI [4, 4, 4] of Claude Debussy’s prelude, Voiles, highlights the predominant use of
the whole-tone scale in this piece. Finally, the piece Parodie from Schoenberg’s
Pierrot Lunaire illustrates for each complex a compactness relatively close to the
average compactness of chords. This results from an almost equally distributed use
74 Louis Bigo and Moreno Andreatta

Johann Sebastian Bach - BWV 328

0,5

2-compactness

0,25

0
9]

8]

7]

6]

8]

7]

6]

5]

6]

5]

4]
]
10

2,

3,

4,

5,

2,

3,

4,

5,

3,

4,

4,
1,

1,

1,

1,

1,

2,

2,

2,

2,

3,

3,

4,
1,

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[
K[

Johann Sebastian Bach - BWV 328 random chords

Claude Debussy - Voiles

0,5
2-compactness

0,25

0
9]

8]

7]

6]

8]

7]

6]

5]

6]

5]

4]
]
10

2,

3,

4,

5,

2,

3,

4,

5,

3,

4,

4,
1,

1,

1,

1,

1,

2,

2,

2,

2,

3,

3,

4,
1,

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[
K[

Claude Debussy - Voiles random chords

Schönberg - Pierrot Lunaire - Parodie

0,5
2-compactness

0,25

0
9]

8]

7]

6]

8]

7]

6]

5]

6]

5]

4]
]
10

2,

3,

4,

5,

2,

3,

4,

5,

3,

4,

4,
1,

1,

1,

1,

1,

2,

2,

2,

2,

3,

3,

4,
1,

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[

K[
K[

Schönberg - Pierrot Lunaire - Parodie random chords

Fig. 3.17 2-compactness of the trajectories of three pieces in the 12 chromatic T/I complexes. In
each complex, the compactness of the pieces (in black) is compared to the average compactness of
chords (in grey)
3 Topological Structures in Computer-Aided Music Analysis 75

of pitch class sets of size 3 or more throughout the piece, as one might expect in
twelve-tone or atonal music.
Three drastically different pieces have been chosen in this example in order to
illustrate how the compactness of trajectories can depend on the musical style in an
illuminating and informative way. However, this method can also be used to reveal
more refined nuances between pieces in the same style.

3.4.3.2 Distance and Classification

The previous examples show that calculating the compactness of a piece in a set
of complexes provides an abstract description that can highlight certain aspects of
its harmonic structure. This description can be used to compare pieces as well. We
propose a notion of distance between musical sequences based on compactness
regarding a set of complexes. For a set of complexes E = {K1 , K2 . . . }, we define the
d-distance between two pieces P and P0 by the Euclidean distance:
s
0

2
DE,d (P, P ) = ∑ Cd (K, TK ) − Cd (K, TK0 ) ,
K∈E

where TK and TK0 are the trajectories representing, respectively, P and P0 in K. The
d-distance is computed by only taking into account chords whose size is greater than
or equal to d + 1.
The mean compactness of the pieces constituting a corpus allows us to calculate a
distance between any piece and this corpus. This notion of harmonic distance has
been used for music classification (Bigo, 2013).

3.4.4 Transformations on Trajectories

The method described above allows any musical sequence to be represented by a

trajectory within a given chord complex. Thanks to the symmetry properties of
some T/I complexes, it is sometimes easy to have an immediate intuition of the
musical interpretation of some geometrical transformations on trajectories or their
embedding in different support spaces. In the first case we can, for example, interpret
a transposition or an inversion operation as a translation or a rotation, respectively, of
a given trajectory.
Figure 3.18 displays the trajectory representing the first measures of Bach’s
chorale, BWV 332, which is rotated by 180◦ . This operation transforms major chords
into minor chords (and vice versa). A second type of transformation consists of
the embedding of a given trajectory in a new support space. This corresponds to a
relabelling of the notes attached to a trajectory according to the pitch content of the
new underlying space. Although the trajectory remains the same, its embedding in a
new space dramatically changes the intervallic relations between the notes. As an
 76 Louis Bigo and Moreno Andreatta

point
reflection

Herr Jesu Christ, dich zu uns wend

&
BWV 332
Herr Jesu Christ, dich zu uns wend
#
BWV 332
#) ! ! ! ! " ! & ! #
* !! !! ! !! ! ! ! !! # !! ! ! ! $ ! ! ! !! ! !! +!! " !!""" *!! ! !! !! !!!! !! ! ! !! % !! " !! !! % % !! ) ! % ! % ! !! ' ! % !! % !! %% !! % !! % !! ' !!
% " !
136. !
% % & & &
! ! ! ! ! ! ! ! !
136. " " ! " ! ! ! ' ! %! '!( '! %! ! %! %! !
+# ) ! ! ! ! ! !! ( !! ! ! ! # ! # !! ! ,!! ""!!" !*( ! ! ! !! !!! !! ! !! ! ! !! % ! % ! ! % ! % ! ( ( ( %% !! % !
! ! ! ! ! " " ! !! $ ! % ! ! %!'! $
' '
& & %! %! # #
# ! 3.18 "!""""! % % !! !of!! the! % !chorale, ! %! %! %! ! %!
# ! !! representing
! !! On!!the!!left, a!!trajectory ( ! the '332,
! by
5 5

* Fig. ! ! !first
!
+ ! !! ' ! % ! ! ' ! % ! % ! % ! % ! ' ! ! ! % % !! ' !! !!
! ! !
" "!
! ! ! ! !
measures BWV
J. S. Bach in KTI [3, 4, 5]. On the right, a point reflection% is applied % to this trajectory, & &
%! %! %!
producing a
! sequence
! ! ! !! !! ! ! ! ! ' ! % ! % ! ' ! % ! % ! ! ! % ! ' ! %!
' ! ' ! ! %! %! %!
! ! ! ! % ! % ! ! ! % !
new
+# ! ! ! ! ! ( ! !! ! ! # , ! #"!""""! % % !! ! ( ! ( ! ! ! % ! ! % ! % ! % ! !$ ! % ! % ! % ! % ! ! ! % ! % ! % !!$
! ! ! #! ! ' '
example, Fig. 3.19 shows the result of embedding the trajectory in KTI [3, 4, 5] of
the first measures of the chorale, BWV 332, in the new support space, KTI [2, 3, 7].
This embedding produces a new musical sequence which will sound more exotic,
due to the prominence of the pentatonic scale in the new support space. From a
mathematical point of view, transformations of trajectories within chord complexes
can be formalized as morphisms (Bigo et al., 2014).

3.5 The HexaChord Software

HexaChord is a freely available computer-aided music analysis software package

based on the spatial representations presented in this chapter.6 The software provides a
3-D visualization of the complex representing any arbitrary set of chords. To improve
intelligibility, chromatic and diatonic T/I complexes of dimension 2 (i.e., constituted
of 3-note chords) can be unfolded as infinite two-dimensional triangular tessellations,
in the same style as the planar representation of the Tonnetz (see Figs. 3.2 and 3.12).
The 12 chromatic T/I complexes together with the four diatonic T/I complexes, each

• Copyleft 2011/01:lbigo/hexachord
6 http://www.lacl.fr/ cette oeuvre est libre, vous pouvez la (photo)copier,
• Copyleft 2011/01: cette oeuvre est libre, vous pouvez la (photo)copier,
la diffuser ou la modifier, selon les termes de la Licence Art laLibre, voir:
diffuser ou la modifier, selon les termes de la Licence Art Libre, voir:
http://www.artlibre.org/licence/lal/ [en, de, es, pt, it] http://www.artlibre.org/licence/lal/ [en, de, es, pt, it]
Créé avec GNU LilyPond 2.14.1 http://www.LilyPond.org Créé avec GNU LilyPond 2.14.1 http://www.LilyPond.org
par Ph. Hardy. http://superbonus.project.free.fr par Ph. Hardy. http://superbonus.project.free.fr
Free Art License Free Art License
 3 Topological Structures in Computer-Aided Music Analysis 77

embedding

K[3,4,5] K[2,3,7]
Herr Jesu Christ, dich zu uns wend Herr Jesu Christ,
BWV 332dich zu uns wend
& "
BWV 332

#) !! !! !! ! # ! ! !! !! $ ! ! ! ! !)) !* ! ! !! &!! ! !! ! ! !! !! ! !! "

* !! ! !
! ! ! ! ! ! ! !, ! ) ! ! ! !! ! ! ! & !! ( ! & ! ! !! ' ! & !! !! & !! & !! !! ' !!
!
% % % " ! $ $ $ %
! ! ! ! ( !! ! ! !! ! ! +!") !*!" !!! !!" ! ! ! !! ! ! !! ! & !! ' !! !! ' & !! & ' !! ! ! % ! % & ! % & ! !
136.
136.
! ! !
+# ) ! ! ! ! ! ! ! # ! # ! ! !) ) ( ! ! ! ! ! ! ! !! $ & ! ! ! '! !
! ! ! ! ! ' # #
' "
& )) & ! 332, & "
# ! 3.19 & !! ' !!! by& !!a trajectory
' ! ' ! in& ! & ! & ! ! & !! ' ! ! ! ' ! !
5
! !of Bach’s
! the!first# !measures ! !,!!)BWV
( ! chorale, !! !! ! represented
5

!
* Fig. ! On!!the left,
! ! ! &! !
! ! ! ! ! !
! % resulting ! from the embedding of the!
! ! ! ! !
" trajectory in KTI [2, 3, 7]
! ! ! ! ! $ $
%
KTI [3, 4, 5]. On the right, the transformation of the sequence
+ '! & ! ! &! !! !
& ! ' ! & ! ' !! ' !! !! & ! & ! & ! !
+#
! ! ! ! !! ! ! !
! ! ! ! ( ! ! !
! !! ! )))! !!!! & ! & !! !' ! ! &!! ! ! & ! & !! & ! ! !! ! !& ! & ! !
! ! ! ! ! #
! ! ! #! ! '
#! #! ! (! (! ! #
'
available in the 12 diatonic scale, result in a total of 60 available planar organizations
of pitch classes.
In the unfolded representation, pitch-classes and chords are repeated multiple times
along interval axes in the complex. As a consequence, a piece can be represented by
an infinite number of trajectories. Different strategies can then be applied to build
a satisfying trajectory from a musical sequence (Bigo, 2013). An algorithm that
minimizes the distance between every pair of successive chords in the complex will
be more efficient for analysing chord progressions in time. On the other hand, an
algorithm maximizing the compactness of each chord (independently of the other
chords) will be more efficient for analysing the harmonic content of a piece.
Musical pieces are imported as MIDI files. A trajectory is automatically computed
for any given piece–complex pair. The trajectory is represented as a path which
evolves in real time within its complex while the piece is being played. The com-
pactness of the trajectory over time (and on average) is automatically computed
for any T/I complex and in any dimension. The calculation of compactness reveals
the complexes which are the most harmonically related to the piece. Following this
hypothesis, this functionality should suggest to the user which spaces can be used in
order to reveal some interesting harmonic properties of a piece via its visualization.
HexaChord also allows the geometrical transformations described in Sect. 3.4.4 to
be carried out on trajectories. A trajectory in a complex K can be translated, rotated,
or even embedded in an other chord complex K0 . Every spatial transformation causes
• Copyleft 2011/01: cette oeuvre est libre, vous pouvez la (photo)copier,
• trajectory
Copyleft 2011/01:
thelavalues labelling the sub-complexes constituting the
diffuser ou la modifier, selon les termes de la Licence Art
to be cette
Libre, voir:
la diffuser
oeuvreThe
refreshed. est libre, vous pouvez la (photo)copier,
ou la modifier, selon les termes de la Licence Art Libre, voir:
harmonic and melodic content of the original
http://www.artlibre.org/licence/lal/ piece
[en, de, es, pt,can
it] be transformed according
http://www.artlibre.org/licence/lal/ [en, de, es, pt, it]
Créé avec GNU LilyPond 2.14.1 http://www.LilyPond.org Créé avec GNU LilyPond 2.14.1 http://www.LilyPond.org
par Ph. Hardy. http://superbonus.project.free.fr par Ph. Hardy. http://superbonus.project.free.fr
Free Art License Free Art License
78 Louis Bigo and Moreno Andreatta

to the new labels of the selected trajectory, thus leading to a new musical piece
which can be exported in MIDI format or analysed and transformed using the spatial
techniques described in this chapter.

3.6 Conclusion

In this chapter, we have presented a method for identifying any set of chords by
a labelled simplicial complex. The family of T/I complexes relate to Generalized
Tonnetze and can be used as support spaces to represent musical pieces for analysis
and classification purposes.
The analysis method proposed here relates to the instantaneous verticality of
chords. However, studying harmonic properties within a piece requires a horizontal
approach as well (i.e., to study transitions between successive chords). In future work,
we intend to address this issue by investigating methods for computing distances
between sub-complexes within a complex.
We believe that the visual and intuitive aspects of this spatial approach constitute
a strong pedagogical advantage for the understanding of harmony. For example, T/I
complexes offer an original and intuitive reformulation of the notion of inversion-
ally/transpositionally related chord classes. Moreover, we believe that the spatial
reformulation of musical problems has a heuristic advantage. Thanks to its intuitive

Fig. 3.20 Graphical user interface of HexaChord

3 Topological Structures in Computer-Aided Music Analysis 79

aspects, spatial structures frequently suggest unexpected questions, thus opening new
research areas.
Finally, methods for automatically computing space-related descriptors (com-
pactness of sub-complexes, distances within complexes, etc.) should allow for the
construction of statistical models given a corpus of a particular style. Such models,
based so far on viewpoints without spatial considerations, have been shown to be use-
ful in various areas including music prediction and classification (Conklin and Witten,
1995). In future work we therefore intend to carry out an in-depth investigation of
some statistical models built from spatial descriptions of musical pieces.

Acknowledgements The authors are very grateful to Antoine Spicher and Olivier Michel from
LACL - Université Paris-Est, Jean-Louis Giavitto from IRCAM - Université Pierre et Marie Curie,
Jean-Marc Chouvel from Université de Reims, and Darrell Conklin from Universidad del Pais Vasco
for numerous fruitful discussions. This research is partially supported by the project “Learning to
create” (Lrn2Cre8). The project Lrn2Cre8 acknowledges the financial support of the Future and
Emerging Technologies (FET) programme within the Seventh Framework Programme for Research
of the European Commission, under FET grant number 610859.

References

Andreatta, M. (2008). Calcul algébrique et calcul catégoriel en musique: Aspects

théoriques et informatiques. In Pottier, L., editor, Le calcul de la musique, pages
429–477. Publications de l’Université de Saint-Etienne.
Andreatta, M. and Agon, C. (2003). Implementing algebraic methods in OpenMusic.
In Proceedings of the International Computer Music Conference, Singapore.
Bigo, L. (2013). Représentations symboliques musicales et calcul spatial. PhD
thesis, LACL/IRCAM, Université Paris-Est.
Bigo, L. and Andreatta, M. (2014). A geometrical model for the analysis of pop
music. Sonus, 35(1):36–48.
Bigo, L., Ghisi, D., Spicher, A., and Andreatta, M. (2014). Spatial transformations
in simplicial chord spaces. In Proceedings of the Fortieth International Com-
puter Music Conference and Eleventh Sound and Music Computing Conference
(ICMC/SMC 2014), pages 1112–1119, Athens, Greece.
Bigo, L., Giavitto, J.-L., and Spicher, A. (2011). Building topological spaces for
musical objects. In Mathematics and Computation in Music: Third International
Conference, MCM 2011, Paris, France, June 2011, Proceedings, volume 6726 of
Lecture Notes in Artificial Intelligence, pages 13–28. Springer.
Callender, C., Quinn, I., and Tymoczko, D. (2008). Generalized voice-leading spaces.
Science, 320(5874):346.
Catanzaro, M. (2011). Generalized Tonnetze. Journal of Mathematics and Music,
5(2):117–139.
Chew, E. (2002). The spiral array: An algorithm for determining key boundaries. In
Music and Artificial Intelligence: Second International Conference, ICMAI 2002,
Edinburgh, Scotland, UK, September 2002, Proceedings, pages 18–31. Springer.
80 Louis Bigo and Moreno Andreatta

Chouvel, J.-M. (2009). Traversée du vent et de la lumière. Six remarques

pour une phénoménologie de la création musicale. Available online at
http://jeanmarc.chouvel.3.free.fr/textes/Traversee0.0.pdf.
Cohn, R. (1997). Neo-Riemannian operations, parsimonious trichords, and their
“Tonnetz” representations. Journal of Music Theory, 41(1):1–66.
Cohn, R. (2012). Audacious Euphony: Chromatic Harmony and the Triad’s Second
Nature. Oxford University Press.
Conklin, D. and Witten, I. H. (1995). Multiple viewpoint systems for music prediction.
Journal of New Music Research, 24(1):51–73.
Euler, L. (1739). Tentamen novae theoriae musicae ex certissismis harmoniae
principiis dilucide expositae. Saint Petersburg Academy.
Forte, A. (1973). The Structure of Atonal Music. Yale University Press.
Giavitto, J.-L. and Michel, O. (2001). MGS: a rule-based programming language
for complex objects and collections. In van den Brand, M. and Verma, R., editors,
Electronic Notes in Theoretical Computer Science, volume 59, pages 286–304.
Elsevier Science Publishers.
Giavitto, J.-L. and Spicher, A. (2008). Simulation of self-assembly processes using
abstract reduction systems. In Krasnogor, N., Gustafson, S., Pelta, D., and Verdegay,
J. L., editors, Systems Self-Assembly: Multidisciplinary Snapshots, pages 199–223.
Elsevier.
Gollin, E. (1998). Some aspects of three-dimensional “Tonnetze”. Journal of Music
Theory, 42(2):195–206.
Holland, P. W. and Leinhardt, S. (1971). Transitivity in structural models of small
groups. Small Group Research, 2:107–124.
Hook, J. (2014). Generic sequences and the generic Tonnetz. Oxford Handbooks
Online.
Mandereau, J., Ghisi, D., Amiot, E., Andreatta, M., and Agon, C. (2011). Z-relation
and homometry in musical distributions. Journal of Mathematics and Music,
5(2):83–98.
Mazzola, G. et al. (2002). The Topos of Music: Geometric Logic of Concepts, Theory,
and Performance. Birkhäuser.
Pousseur, H. (1968). L’apothéose de Rameau. Essai sur la question harmonique.
Musiques Nouvelles. Revue d’esthétique, 21, pages 105–172.
Pousseur, H. (1998). Applications analytiques de la technique des réseaux. Revue
belge de Musicologie, 52, pages 247–298.
Rameau, J.-P. (1722). Traité de l’harmonie, reduite à ses principes naturels. Jean-
Baptiste-Christophe Ballard.
Tymoczko, D. (2012). The generalized Tonnetz. Journal of Music Theory, 56(1):1–52.