Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

SampComp: sample-based techniques for algorithmic composition

Andrea Valle
StudiUm/CIRMA - Università di Torino
andrea.valle@unito.it

ABSTRACT cate reproduction, ”audioparity” may be proposed to indi-

cate a composition mode in which musical data originate
The paper introduces a series of techniques for algorith- from sound. A compositional technique is “audioparous”
mic composition in which compositional data are extracted if it defines a projection between a source sound material
from audio at the sample level. In this way, typical Dig- and an outgoing musical organization. In short, one has
ital Signal Processing operations can act as tools to cre- audioparity when a composition c results from the applica-
ate/retrieve music materials. First, in order to describe tion of an audioparous function Fap on a sound s: thus, c =
audio-driven processes in composition, the notion of au- Fap (s). Instead, “audioparism” could be used to describe
dioparity is introduced. Secondly, various approaches are an aesthetic and practical composition attitude centred on
discussed that map sample values to composition events. audioparity. Such a category does not depend on authors or
The implementation of these processes is then presented, periods, and, in the same historical context, different com-
in particular in relation to music notation generation. Fi- position practices may be more or less audioparous. As an
nally, the paper discusses some applications of the same example, French Renaissance composer Clément Janequin
techniques to physical computing. has often pursued a pervasive audioparism in many pieces
(La chasse, La bataille, Les chants des oyseaulx), explic-
1. INTRODUCTION: AUDIOPARITY, itly inspired by the transcription of acoustic events. Sim-
AUDIOPARISM AND TIME SCALING ilarly, in the 20th Century Olivier Messiaen has notori-
ously integrated the ornithological transcription into many
Algorithmic approaches and techniques to music (regard- of his works, but often including it into a different com-
less of the final medium, whether directly the audio signal, position context [3]. In contemporary music, Spectralism
or the musical notation or messages for multimedia ob- is evidently an area of choice for a pervasive audioparism
jects) are based on a preliminary, general assumption: the (see [4]). In his now classic Les espaces acoustiques cy-
presence of some formally defined procedure able to pro- cle, Gérard Grisey literally re-orchestrates sounds by tak-
duce a number of representations of musical data. Since ing into account data from sonograms: as the French com-
the work of Chomsky on formal grammars [1], this kind poser said, “we are musicians and our model is sound”
of approach is typically labelled “generativism”, and in the (cit. in [5]). On the other side, many composition prac-
artistic domain it is so strictly tied to the use of computa- tices are non-audioparous at all. As an example, Twelve-
tional resources that historically the terms “computer art” tone technique is indeed not audioparous, as it organises in
and ”generative art” are synonyms [2]. In Chomsky’s ap- various ways an abstract source material, i.e. a 12 pitch-
proach, the term has a technical meaning (generativism as class structure. But the same argument applies to most
a formal grammatical theory), but its Latin etymology is polyphonic music from 14th to 16th Century. Without dis-
indeed linked to human/animal reproduction. In this re- cussing further whether the notion of audioparity is rele-
gard, zoology classically distinguishes between viviparity vant or not from a musicological perspective, an interesting
(the offspring is generated alive), oviparity (the embryonic point in many audioparous practices, like the ones intro-
development takes place in the deposited egg), and ovo- duced before, concerns time scale. Messiaen already noted
viviparity (there are eggs, but they hatch in the mother’s that bird songs must be slowed down in order to match
body, and the offspring is born alive). The distinction is in- a comfortable time resolution for human ear [3]. In the
teresting because it concerns the ways in which the repro- same pathway, temporal dilatation in relation to the origi-
ductive behaviour takes place, and is orthogonal to zoolog- nal sound material source is a crucial feature in Grisey’s
ical classes (notoriously, the platypus is at the same time work. This process of time expansion is not irrelevant,
mammal and oviparous). In relation to these aspects, the because of non-linearities in time perception. While dis-
mode of (re-) production and the independence from the cussing such an issue in the musical context and following
class, one could think of a taxonomy of the modes of mu- the seminal reflections by Stockhausen [6], Roads notes
sical creation, regardless of styles, supports or eras. Con- that, as sound passes from one time scale to another, it
sequently, as the suffix “-parity” is used in zoology to indi- crosses perceptual boundaries, as human perception pro-
cesses each time scale differently [7]. In relation to this,
Copyright: c 2018 Andrea Valle et al. This is an open-access article distributed Xenakis has proposed to distinguish four time boundaries
under the terms of the Creative Commons Attribution License 3.0 Unported, which (macro/meso/mini/microstructures), that divide time scale
permits unrestricted use, distribution, and reproduction in any medium, provided into separate perceptual regions, i.e. timbres, notes/events,
the original author and source are credited.

CIM 2 018 P R OC E E DING S 12 8

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

4
1 73

X X # X X X X X X X X X X X # X nX X # X nX # X nX X
& 4 X #X X X #X X X
72

X # X X nX
71
0.6
70
69

0.2 68
67
66
-0.2 65
64 3

& X #X X X X #X
63

X X #X X
-0.6

X X X X X X
62
61
-1 60

0 24 48 0 12 24 36 48

Figure 1. Sinusoidal signal src in a 48-point table (left);

requantized version mld (right). Figure 2. Transcription in music notation of Figure 1 (1
semiquaver per sample).

rhythms, global evolutions [8]. While these considera-

tions are indeed relevant from a perceptual perspective, zation therefore defines the range within which the sam-
still, from the point of view of composition practices, they ple values from src are mapped onto pitches. For ex-
do not prevent to translate a given feature at a certain time ample, if we consider to map 32 bit audio samples onto
level into another one, as it happens in the aforementioned an equally-tempered octave (12 pitches), the resolution of
works by Messiaen or Grisey. The following sections will the source signal must be reduced to less than 4 bits (as
therefore discuss a set of audioparous techniques that trans- 24 = 16 > 12). These two operations structurally intro-
late sonic microstructures (specifically, at the sample level) duce a loss of information and, in fact, a quantity of noise,
into sonic ministructures (at the event/note level). These as they are non-linear. But the exploration of resampling
techniques can be called SampComp, as they relate the and quantization error is indeed a large part of the interest
sample level to the composition one. of the operation. The simplest case consists in resampling
by taking a sample every n. The new sampling rate srn
will be an integer submultiple of the original sampling sr,
2. SAMPCOMP TECHNIQUES i.e. the sampling period sT n will be an integer multiple of
sT . As far as quantization is concerned, a mapping func-
By providing an abstract unified representation for all data,
tion can be defined that associates the value in the sample
the numerical domain allows to easily switch between time
domain ([±1.0]) to values in a certain range in the inte-
scales in sound representation. SampComp processes, while
ger pitch one by linear interpolation. In Figure 1, on the
in some way related to those digital synthesis techniques
right, the src signal is converted into the new mld signal 1 ,
that focus on the time-domain representation and operate
through a mapping onto the octave that has the middle G
directly at the sample level (e.g. Xenakis, Koenig, Brün,
as a pivot (hence, in MIDI notation, in the range [67 ± 6],
see [9]), still have with a different goal, that of a time scale
from 61 to 73) 2 , and with srn = sr (all 48 samples of
expansion. The idea that the signal can act as a control
src are taken into account). From this simple example two
of “notational” parameters is widely used in analog elec-
potentially interesting compositional features emerge:
tronic music, for example in Voltage Control technique.
As an example, Strange’s 1972 work [10] has investigated • to some extent, the sinusoidal waveform of the signal
the subject by discussing voltage control operations in re- src is maintained as the overall form of the melody
lation to standard music notation. However, the numerical in the new signal mld;
domain is more pervasive and flexible and allows to pre-
cisely map sample data into note ones. As a general case, • the quantization to the equally tempered octave in-
the following discussion will focus on the generation of a troduces a distortion that results in a terracing oper-
sequence of pitches (i.e. a melody) from a waveform. ation, so to speak.
The simplest case may take into account a sine wave,
as the most basic and usual signal in audio synthesis. The The transcription of mld into musical notation (Figure 2)
generated digital signal (a cycle) is represented by a vector assumes that every sample in mld has a duration of one
containing the values in normalised range ([±1.0]), as this semiquaver.
is a standard representation, for example widely used for Expanding further the discussion, the src signal in Fig-
waveform tables in digital oscillators. The resulting digital ure 3, left, again counts 48 samples, but this time 3 cycles
signal src, for an array of 48 elements (an arbitrary size), of the sine wave are tabulated. The mapping quantization
is plotted in Figure 1, left. function is the same, but the mld signal is obtained by re-
Two basic digital operations can be applied to the src sampling src with a sampling period sT n = 3 × sT (that
signal. The first is resampling. In particular, since the is, one sample of src is taken every three). As seen in
amount of information that describes a signal over time Figure 3, right, the sinusoidal waveform is still visible, but
is typically very large if compared to the one describing
1 Sample values are traced in light blue for sake of visibility.
a melody (e.g. 1 second of audio at CD quality requires 2 Each float value in the range [−1.0, 1.0] obtained from src is
44100 samples), the signal can be resampled at a lower (v+1)
mapped onto the range [61, 73] by means of ( 2 ) × 12 + 61, then
rate than the original. The second is quantization. Similar rounded and converted to integer. The same mapping will be applied in
to sampling, the information of the source signal src (in all following examples for sake of simplicity. Some audio examples are
floating point format) has typically a much higher resolu- available here:
http://www.musicaelettronica.it/
tion than the one required for pitch description. Quanti- campcomp-lineamenti-di-musica-audiopara-2/

CIM 2 018 P R OC E E DING S 12 9

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

7 X X X X XJ X X Xj X X X
&8X
1 73

#X X X #X X j
X
72
71
0.6
70
69

0.2 68
67
66
-0.2

X X XJ
65

&X X X X X Xj
4

X X X X X
64

#X X X j
X
63
-0.6
62
61
-1 60

0 24 48 0 4 8 12 16

#X X X X X
7

Figure 3. 3 cycles of a sinusoidal signal (left); requantized & Xj #X Xj Xk

version (right).

3 # XJ . Xj. XJ . XJ .
Figure 6. Transcription in music notation of Figure 5 (1
& 8 Xj. j
X. #j
X. Xj. semiquaver per sample).

# Xj. # XJ . # Xj.
5

& # Xj. # Xj. Xj. 1 1

0.6 0.6

0.2 0.2

-0.2 -0.2

Figure 4. Transcription in music notation of Figure 3 (1 -0.6 -0.6

semiquaver per sample). -1 -1

0 20 40 60 80 100 0 20 40 60 80 100

1 73
72
71
0.6
70

the combined effect of the presence of 3 cycles and sam-

69

0.2 68
67
66

pling/quantization produces a set of asymmetries in the re- -0.2 65

64
63

sulting melody. The musical transcription (Figure 4) may -0.6

62
61
-1 60

decide to assign any duration to any sample of mld, but 0 20 40 60 80 100 0 5 10 15 20 25

to underline the relationship between sT and sT n it still

assumes the equivalence between a sample of src and a Figure 7. Sinusoidal signal (top, left), percussive envelope
semiquaver. (top, right), enveloped sinusoid (bottom, left); requantized
Proceeding further, the following example demonstrates version (bottom, right).
the increase in asymmetry by increasing the number of cy-
cles in src (5 in 100 tabulated samples, Figure 5) and re-
sampling with sT n = 7 (a sampling that is not yet critical, electronic music) for the purposes of melodic generation.
but certainly important). Although in some way the result- As such operations are defined in the time domain, a basic
ing pitch sequence still bears a sinusoidal shape, the for- case is the application of an amplitude envelope. In Figure
mer –as a result of the relationship between resampling and 7 the src signal (bottom, left) is obtained by multiplying 5
quantization in mld– includes only 9 pitches, 4 of which cycles of a sinusoid in 100 samples (top, right) by a percus-
introduced in a sort of final ending tail, while the other 5 sive (i.e. exponentially decreasing) envelope signal (top,
are repeated twice (3 times in the case of the C], MIDI note left). The melodic contour mld (with sT n = 4 × sT , 1
61). It can be seen that the resulting pattern is made up of value extracted every 4 samples, Figure 7, bottom, right),
15 elements. As in the previous examples, the transcrip- that result from mapping, clearly shows the relationship
tion into music notation (Figure 6) maintains the equiva- between the signal amplitude of src and the distribution
lence between source samples and semiquavers: the period along the chosen range for pitches. The fading amplitude
sT n = 7 can easily counted as a duration of 7 semiquavers tail of src results in the iteration of the pivot pitch G (MIDI
in each note. 67). Transcription into musical notation is given in Figure
The previous examples have shown that the underlying 8.
logic in SampComp is to explore signal time-patterns at the The application of an envelope can be generalized. The
sample level and exploit them at the mini/meso level. This construction of the waveform thus becomes the construc-
working hypothesis allows then to explore standard Digi- tion of the melodic contour, in relation to sampling and
tal Signal Operations (DSP) techniques originally devised quantization. In Figure 9, on the left, a 100-point tabu-
for signal generation and manipulation (and well-known in lated sinusoid undergoes three operations. The first is a

5 X #X X #X #X
&4 X X X X X # X nX X #X
1 73

X
72
71
0.6
70
69

0.2 68
67
66
4

& X X X X X X X X X X
-0.2 65
64
63
-0.6
62
61
-1 60

0 20 40 60 80 100 0 5 10 15

Figure 5. 5 cycles of a sinusoidal signal (left); requantized Figure 8. Transcription in music notation of Figure 7 (1
version (right). semiquaver per sample).

CIM 2 018 P R OC E E DING S 130

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

19 XkX.. XJ # XJ # XJ . X XK nX .. XJ XJ # Xj. # X
& 16 # Xj. # X Xk# X .. nXj Xj
1 73
72
71
0.6
70
69

0.2 68
67
66
-0.2 65
64
63
-0.6 4

Xk X .. Xj Xj # Xj.
62

& # Xj. nX Xk# X .. Xj Xj # Xj. X

61
-1 60

0 20 40 60 80 100 0 20 40 60 80 100

Figure 9. Sinusoidal signal processed by absolute value

and impulse replacing (left); requantized version (right). Figure 11. Musical transcription of a variable rate Sam-
ple & Hold algorithm applied to one cycle of a sinusoidal
signal.
5 #X X X X X #X X X X X
& 4 X X # X X X X X X X X X X X # X X X # X X X X X X X X X X nX X X X 1 73
72
71
0.6
70
69

# X X # X X X X X X X X nX X X X # X X X X X X # X X X X X X # X X X X X X X X X X X X X X
3 0.2 68

&
67
66
-0.2 65
64
63
-0.6
62
61

# X X # X X X X nX X X X X X X X # X X X X nX X
-1 60
5

&
0 20 40 60 80 100 120 140 160 0 11 22 33 44 55 66 77 88 99 110 121 132 143 154 165

Figure 12. A signal obtained by iteratively chaining aver-

aged 15-sample blocks (left); requantized version (right).
Figure 10. Transcription in music notation of Figure 9 (1
semiquaver per sample).
the transcription into music notation that results from such
a process. Here, a 100-point 1-cycle sinusoid src signal is
sampled along the (cyclic) series [3, 5, 7, 2, 2], i.e. the sam-
form of digital distortion through the application of the ab- ples 3, 5, 7, 2, 2 are taken from src, and so on. By counting
solute value that reverses the negative half-cycle to posi- note durations in semiquavers in Figure 11, the “rhythmic”
tive values (in the audio domain, this process has been pro- sampling becomes apparent.
posed by Puckette [11] as a crude but efficient octaver al- All the previous examples have started from a sinusoid
gorithm that introduces distortion). The second is a rescal- signal as the most basic signal in electronic music. Indeed,
ing of the amplitude (here halved) so that it is compressed the manipulation of signals in the time domain opens up a
in the range [0.0, 0.5]. Finally, a further operation is ap- Pandora’s box of possibilities, far more extensive than the
plied, iterated with modulo 4 on the samples, which re- previous preliminary considerations. Further explorations
places the first and second values of a four-sample block may take into account a large corpus of DSP techniques in
with 1 and 0.7 respectively. Such an operation, while typ- order to generate note contour material. In the following, a
ically numerical, can however be thought of as a form of simple case is thus introduced as an example. An averag-
mixing by replacing, in which samples of the sine-derived ing filter is an operation that calculates the value of a sam-
wave are regularly replaced by an almost impulsive posi- ple x through the average of a certain number n of nearby
tive signal. Such a process can be easily generalized, i.e. samples (before or after x) [14]. The recursive application
in terms of mixing of weighted sum of signal vectors to of a filter is at the base of the notorious Karplus-Strong al-
obtain desired shapes (or, with an experimental attitude, gorithm [15], in which a noisy signal is stored into a table
to explore emerging melodic shapes from various possible that is rewritten iteratively by applying a median filter (the
operations). Even in the compressed selected pitch range so-called “recirculating wavetable”). In the following ex-
(an octave) the result in terms of melodic transcription (for
sT n = sT , Figure 10) is interesting as it shows some
perceptual features related to grouping. The melody in
4 X X X # X X nX X X X X # X X X X X X # X X X # X X X X nX nX
& 4 X X #X X
fact tends to segregate into two streams [12], the pivot top
note and the lower “terraced” arpeggio, in a sort of virtual X X X

X X X nX # X nX X X # X X X nX # X X X # X nX # X X X X X # X nX X X X X X
3

& #X
polyphony (the melodic contour is more clearly displayed
by Figure 9, right). X X
X X
& X X # X X X X X # X nX X X X X nX X # X X X X X X # X nX X X X nX X X X
5
In the previous discussion, a fixed resampling rate for
src has been assumed. While fixed sampling frequency is
X X X X # X # X nX X nX X X X X X X X
7

& X X X X # X # X nX X X nX X X X X X
a standard technique in DSP, SampComp techniques may
benefit from non standard approaches. In this regard, vari- X
able sampling can be applied to a src signal. In electronics, X X X
9

a Sample & Hold circuit polls the signal, extracts the value, & X X # X# X nX X nX X X X X X X X X X # X# X nX X nX X X X X X X X X X # X # XnX Xk
and keeps it until a new polling is asked for [13]. The next
example is based on a Sample & Hold algorithm that polls
the source signal src according to a sequence of variable Figure 13. Transcription in music notation of Figure 12 (1
sample intervals instead of a constant rate. Figure 11 is semiquaver per sample).

CIM 2 018 P R OC E E DING S 131

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018
q»
4 Kalimba
4 1 4 9 2 8 7 3 5
1 3
B >œ
i n œ n œ n >œ Bœ ‰ n œ n >œ Œ nœ Bœ ‰ Bœ nœ
Perc Bœ Bœ Œ Bœ Jœ Jœ µœ Œ ‰ Œ B œ B œ B >œ Œ nœ Bœ
F µœ µœ µœ µœ S
3
3 3

Pf
q»96
4
4 A
m 0 n 0 n >0 n0 n0 B
n 0 b >0 b 0 D D O
B > A
n0 n0 n 0 n 0 n >0 n0 b 0 bB0
1 D
b0 n0 n0 E E Bœ n0 b0
ˆ S n0 n0 m0 n0 n0 n0 ‰ n0 m0 m0 m0 S
Guit Œ ‰ Œ Œ ‰
3 3 3 3
P
3 3

Figure 14. Beginning of Alatia, from Regnum lapideum.

ample, a src signal is purposely constructed in a two-step

process. At the initialisation step, a signal src0 of 15 points GUI GUI
is filled with random values. Then, in the second step, the
signal src0 is filtered (in fact, it is a low pass filter) with a
AudioFunc src MapFunc mld GuidoGen
step of 3 samples: the value of xn is calculated by averag-
ing the samples [xn , ..., xn+2 ]. The new 15-sample src1 ,
that results from filtering, is concatenated with the previ- SuperCollider PSGen
ous one, then src1 is filtered again by the same procedure
to obtain src2 , and so on. The filtering and chaining pro-
cess is repeated 10 times, so that the final src signal has a notation NoteAbility Pro ps guido
size of 165 samples. As shown in Figure 12, the recursive
filtering progressively smoothes the edges of the curve: in
the mapped melodic signal mld, the initial dispersion of Figure 15. Integrated software pipeline for SampComp
the pitches is thus progressively replaced by a more or less and notation generation.
stable curve, centred around a pivot note, as it can be seen
in Figure 13. The final result depends crucially on the ran-
domly filled sample block src0 at the initialization step. notation in the workflow [17]. In fact, an experimental ap-
In order to conclude the discussion with a real musi- proach such as SampComp requires that the generation of
cal example, Figure 14 shows the beginning of the piece the notation is integrated into the compositional process
Alatia, from the work Regnum lapideum by the author and in order to rapidly inspect the final results. To this pur-
Mauro Lanza, commissioned in 2016 by Ensemble 2e2m pose, a light software architecture has been implemented.
(see later). Here the melodic contour is assigned to kalimba In Figure 15, the softwares involved are boxed in bold,
and (prepared) guitar. A variable Sample & Hold tech- the software modules are shown in rounded corners, gener-
nique such as the one introduced before has been used to ated data structures and files are enclosed in rectangles. As
generate the pitch profile. It is easy to recognise a sinu- shown in the Figure 15, SuperCollider implements the gen-
soidal pitch pattern, obtained with a variable (“rhythmic”) eration of the src signal (through specialised modules, in
polling rate. Figure labeled as AudioFunc) and the mapping functions
(MapFunc) that result in data structures for the represen-
tation of pitches (mld). SuperCollider also operates as a
3. IMPLEMENTATION
“glueing” solution. The PSGen module generates text files
The implementation of SampComp techniques by itself does in the PostScript vector graphic format [18] for the dis-
not pose particular problems as the numerical procedures playing of the generated signals (both src and mld). The
involving the manipulation of small vectors are neither par- graphic visualisations of signals shown in the previous Fig-
ticularly complex nor they require relevant computational ures have been obtained in this way. The integration of mu-
resources. All previous examples have been implemented sic notation into an automated pipeline is not a trivial task.
in the SuperCollider environment [16], which features au- An option could be to generate MIDI files as an intermedi-
dio synthesis capabilities and an Object-Oriented language ate format to be imported into music notation software, the
to control them. The use of a programming language specif- main drawback being the loss of control on music notation
ically dedicated to audio is particularly appropriate because typesetting, as the task is left to the importing software.
the latter comes with a large repertoire of dedicated audio On the other side, MusicXML 3 requires the definition of a
algorithms that can be easily integrated into the workflow. large set of typesetting-related parameters. Another option
Moreover, through SuperCollider it is possible to sonify for the generation of musical notation is LilyPond [19],
the obtained signals and visualise them through graphical a music engraving software which features a text-based
interfaces for sake of inspection. A crucial point in algo- input. Rather, in the present work the musical examples
rithmic composition techniques that include musical no-
tation concerns the integration of automatically generated 3 https://www.w3.org/2017/12/musicxml31/

CIM 2 018 P R OC E E DING S 132

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

3 q»150 4 3 4 3 4
1
4 X
1 4 X
4 4 4 Z 4
1
> L >Z > s Z X > s 8 > s s
Vno i ‰ m0 ‰ S Œ ‰ m0 ‰ n0 s Œ . n >0 m 0 Œ . n >0 0 ‰ 0 Œ¿ m 0 Œ . ‰ n0 Ó n >0 ‰
S S > n0 S >
0.8 >
Jƒ Ksempre

3 X
0.6 n >0
8 legno batt. Z n >0 Z
Vla X >
¿ ‰ nœ n’ ‰ ‰ s n >0 ‰ S ‰ ‰ S ‰ s Œ. s Œ i Ó ‰
0.4 S fl . n >0 S n >0 n0 n flœ n ’.
ç Jƒ Ksempre 0
0.2 Y 2
0 8 8 8 Y
f
8
b >0
> s µ >œ µœ ‰ 8
> >¿ ‰ ‰ µ >œ
Y 2
0 Vc ¿ ‰ S Œ. n >0 ‰ ‰ S Ó ‰ Ó s ‰ ‰ s Ó s ‰
S S S S b >0
œ 0 µ >œ µ >œ
ç Jƒ Ksempre
-0.2 8 0 8 8
-0.4

-0.6

-0.8
Figure 18. Beginning and ending measures from the in-
-1 strumental score of Daripessus Yantillipicus.
0 50 100 150 200 250 300 350 400

controlled, electromechanical devices, in the perspective

Figure 16. Recursively filtered signal used as a source for
of physical computing [23].
Daripessus Yantillipicus.
In all the four setups of the cycle, the electromechanical
devices are obtained from low-cost, scavenged/recycled ob-
XXVI: Daripessus yantillippicus jects and materials. In order to produce sounds, they are
girodisco1
0 5 10 15 20 25 30 35 40 45
equipped either with small 12 volt DC motors, that provide
girodisco2
girodisco3 mechanical energy to scrape, beat, pluck, or blow various
molatore1
molatore2
molatore3
objects and materials, or with 12 volt relais, that switch
molatore4
armonica1
other circuits on/off (e.g. in order to scrape metal plates,
armonica2
armonica3 they turn on/off modified electric knives operating at 230
zampogno1
zampogno2
zampogno3
volt). While relais work in a binary mode (on/off), DC mo-
zampogno4
radio1
tors can be fed with a variable amount of current modulated
radio2
radio3 (on a 8 bit scale) by means of Arduino microcontrollers’
segopiatto1
segopiatto2
segopiatto3
Pulse Width Modulation (PWM) circuitry [24]. While op-
segopiatto4
spremoagrume
erating in real-time, in Systema naturae these devices fol-
rasoio
meshugghello low a pre-composed score that allows a strict synchronisa-
lampadina

0 5 10 15 20 25 30 35 40 45 tion with human players, as the latter follow a click track.

In Systema naturae, each of the four works of the cycle is
in itself a collection of short pieces. Among these, various
Figure 17. Visualisation score for electromechanical de- have been composed using SampComp techniques, applied
vices in Daripessus Yantillipicus. both to melodic construction (see the example at the end of
Section 2) and to device control.
were produced via NoteAbility Pro 4 , an advanced music In Regnum animale, the piece Daripessus Yantillipicus
notation software that includes GUI and playback. Note- exploits the recursive noise filtering approach discussed
Ability Pro can import files written in the compact Guido above, to define the “melodic” sequence of activation of
textual description format [20] and has a much more flexi- the devices. This means that values of audio samples are
ble user frontend compared to the tools available for Lily- quantized so that they can be associated to the devices, as
Pond. Thus, a SuperCollider module (GuidoGen in Fig- the latter are ranked in a list. The recursively filtered source
ure 15) has been implemented, that generates files in the signal used to activate deviced is shown in Figure 16 (DC
Guido format, so that they can be imported into NoteAbil- offset is not an issue, as src will be mapped onto mld).
ity Pro, played back, eventually re-edited, finally exported The visualisation score is shown in Figure 17 5 . In this
as graphic notation files. case, activation for each device is on/off. The noisy attack
in the src signal thus results in a dispersion of events, while
filtering reduces the range of activated events. In the same
4. APPLICATIONS TO OTHER TIME-DOMAIN piece, such an activation sequence is orchestrated for tradi-
BEHAVIOURS tional instruments too. Figure 18 shows the first three and
the last two measures of the instrumental parts (separated
Audio DSP techniques provide a vast repertoire of time-
by a dashed line), so that the filtering from noise-based dis-
patterns that can be used to represent various behaviours on
persion of events to a compressed range is apparent.
different time scales. In order to extend the previous dis-
The same mapping approach (sample values onto ranked
cussion, this section presents some of the procedures dis-
devices) is exploited in Ismiosia Papanabuis from Regnum
cussed above in relation to the Systema naturae cycle for
vegetabile, an example interesting also from a general per-
acoustic instruments and electromechanical devices (2013-
17), composed by the author and Mauro Lanza [21, 22]. 5 Score for devices are automatically generated from the composi-

The cycle is made up of four pieces (Regnum animale, Reg- tion data structures by exporting PostScript code, following the approach
discussed before. From top to bottom, the names of the devices are:
num vegetabile, Regnum lapideum, Fossilia) and integrates girodisco1, girodisco2, girodisco3, molatore1, molatore2, molatore3,
various instrumental ensembles with real-time, computer- molatore4, armonica1, armonica2, armonica3, zampogno1, zampogno2,
zampogno3, zampogno4, radio1, radio2, radio3, segopiatto1, segopiatto2,
segopiatto3, segopiatto4, spremoagrume, rasoio, meshugghello, lampad-
4 http://debussy.music.ubc.ca/NoteAbility/ ina.

CIM 2 018 P R OC E E DING S 133

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

4 q»150
4 Z
n œ. ‰ . . . . . n œ. ! n œ. ‰ . .
‰ n k œœ .
1
‰ . nœ Ó
1
Fl i Œ ! nœ ‰ nœ ! ‰ Œ ! nœ ! nœ ‰ Œ Ó
SS SS SS SS SS SS SS SS
0.8 ∏ p f F ø
.5 5
. 5
. . 5
0.6
Ob i Œ Bœ Œ Ó Bœ Œ. Bœ ‰ Ó i ‰ Bœ Œ . Ó
S S S S
0.4 p P F
3
. 89
‰ m l œœ ..
lœ
0.2
Cl i Ó Œ S Œ i Ó w
w

p
0 ø
con Sordina
1 1 s
Œ. ! n œs ‰ n œs ‰ .
-0.2 ss ss ss ss ss ss ss s s
nœ ! ! nœ ‰ Œ nœ ‰ nœ Œ nœ ‰ nœ ! nœ ! nœ Ó
s
Vno i
-0.4 ∏ F π P ∏ π f π
con Sordina
3 3 3 3 3 3 3 3 3 3 3
1 s s s s s s s s s
-0.6
Vla i ‰ mœ ‰ ‰ mœ ‰ Œ mœ Œ Œ mœ mœ mœ ‰ mœ ‰ mœ ‰ mœ mœ ‰ mœ mœ Ó mœ Œ Œ mœ Œ ‰ mœ ‰ Œ
F f F p F ∏ P π Ff P ∏
-0.8
5 5
W s h
..
con Sordina 2 s Ó s Ó 2
-1 Vc i Ó Ó i ‰ n n 0œ
1 Bœ Bœ
p π 0 ø
0 0.25 0.5 0.75 1

Figure 19. Audio sample (1 second) used as a source sig- Figure 21. Beginning measures from the instrumental
nal for Ismiosia Papanabuis. score of Ismiosia Papanabuis.

VII: Ismiosia papanabuis both the PWM and the audio signal are one-dimensional.
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90

fischietta
sirenetto1
While in previous examples, all devices were treated as
ocarino1
zampogno7 operating in a binary mode, such a PWM control signal
zampogno6
zampogno5
zampogno4
can be directly generated via SampComp techniques. The
zampogno3
zampogno2
zampogno1
following two examples are from Regnum vegetabile too.
trombo9
trombo7
trombo5
The average filtering technique is used in the piece Ario-
trombo3
trombo1 lactus usteginsiphillemena with a different approach. In
trombo2
trombo4
trombo6
the piece, filtering is used to manage the delivery of elec-
trombo8
trombo10
armonica1
tric current to group of devices, thus properly implement-
armonica2
armonica3
armonica4
ing a Karplus-Strong-like algorithm that simulates an at-
armonica5
armonica6
tack/resonance acoustic model to generate 8-bit PWM sig-
armonica7
ocarino2
sirenetto2
nals. Figure 22 reproduces the visualisation score for ma-
ancetto
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 chines. In this case, the averaging filter is used to directly
control the values sent by Arduino’s PWM circuitry to de-
vices. Finally, the same approach is at play in the piece Re-
Figure 20. Visualisation score for electromechanical de-
ocerantroma phenaudi. PWM signals for devices are ob-
vices in Ismiosia Papanabuis.
tained by applying the previously introduced technique of
the inverted sinusoid with impulse replacement, yielding
spective. In this case, a short audio fragment is mapped to activation patterns made up of a continuous behaviour
directly onto events, like in Daripessus. The 1-second au- with short spikes (as can be clearly seen in Figure 23).
dio sample (Figure 19) is taken from the recording of an
electroacoustic improvisation made by the author and Si- 5. CONCLUSIONS
mone Pappalardo, and it has been chosen both as a formal
model for composition because of its temporal behaviour Audioparity can be seen as a musical aesthetics meta-cate-
and as a homage to Pappalardo, a frequent collaborator of gory that groups together many historically and technically
both composers. The only operation performed by Map- different approaches to composition, that share a common
Func is the quantization from values in the audio signal to interest in sound as a starting point for music creation.
indices representing the selected devices in a ranked list. Thus, SampComp can be seen as an audioparous approach,
Figure 20 6 shows the visualisation score for devices. De- provided that sound as as a starting point is not considered
vice activations describe a single melodic contour, follow- neither in perceptual nor in acoustic terms, rather in rela-
ing the imported waveform (compare Figure 19 and Figure tion to its technological implementation as a digital audio
20). In the piece, the same principle is applied to acous- signal. This allows to profit from two mutually related as-
tic instruments, so that the overall composition is properly pects. On one side, the digital encoding of information
a hochetus (if attacks, rather that durations, are taken into allows a continuous transition from the sample level to the
account), as Figure 21 shows. event one, due to the common numerical format. On the
As discussed, the electromechanical devices (in case other one, by taking into account digital audio signals, a
of DC motors) expose as a control interface for their be- large corpus of knowledge is available to experimentation
haviour a single parameter, the 8-bit value addressable for from the DSP theoretical domain. In short, the audio signal
PWM signals. In this sense, they fit the audio signal, as is taken into account as an abstract cognitive form to de-
scribe time patterns. Hence, the possible interest of Samp-
6 From top to bottom, devices are: fischietta, sirenetto1, ocarino1,
zampogno7, zampogno6, zampogno5, zampogno4, zampogno3, zam- Comp techniques, which use can be further expanded and
pogno2, zampogno1, trombo9, trombo7, trombo5, trombo3, trombo1, not necessarily limited to pitch contour generation: as an
trombo2, trombo4, trombo6, trombo8, trombo10, armonica1, armonica2,
armonica3, armonica4, armonica5, armonica6, armonica7, ocarino2, example, to manage the harmonic level or, as discussed in
sirenetto2, ancetto. the last section, to generate time forms for generic sound

CIM 2 018 P R OC E E DING S 134

Proc e e din g s of the 2 2 n d C IM, Udin e , Nov e m be r 2 0-2 3, 2 018

[8] I. Xenakis and R. Brown, “Concerning time,” Perspec-

VIII: Ariolactus usteginsiphillemena
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 tives of New Music, vol. 27, no. 1, pp. 84–92, 1989.
fischietta
sirenetto1
ocarino1
zampogno7 [9] C. Roads, The Computer Music Tutorial. Cambridge,
zampogno6
zampogno5
zampogno4
MA, USA: MIT Press, 1996.
zampogno3
zampogno2
zampogno1
trombo9
[10] A. Strange, Electronic Music: Systems, Techniques,
trombo7
trombo5
trombo3
and Controls. Dubuque, Iowa: William C Brown Com-
trombo1
trombo2
trombo4
pany, 2nd ed., 1983.
trombo6
trombo8
trombo10
armonica1
[11] M. Puckette, The Theory and Technique of Electronic
armonica2
armonica3
armonica4
Music. River Edge, NJ, USA: World Scientific Publish-
armonica5
armonica6
ing Co., Inc., 2007.
armonica7
ocarino2
sirenetto2
ancetto
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
[12] A. Bregman, Auditory Scene Analysis. The Perceptual
Organization of Sound. Cambridge, Mass. and Lon-
don: The MIT Press, 1990.
Figure 22. Visualisation score for electromechanical de-
vices in Ariolactus usteginsiphillemena. [13] P. Horowitz and W. Hill, The Art of Electronics. New
York, NY, USA: Cambridge University Press, 3rd
(2015) ed., 1989.
XIV: Reocerantroma phenaudi
fischietta
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
[14] D. Rocchesso, Introduction to Sound Processing.
sirenetto1
ocarino1 Firenze: Mondo Estremo, 2003.
zampogno7
zampogno6
zampogno5
zampogno4
zampogno3
[15] K. Karplus and A. Strong, “Digital synthesis of
zampogno2
zampogno1 plucked-string and drum timbres,” Computer Music
trombo9
trombo7
trombo5
Journal, vol. 7, no. 2, pp. 43–55, 1983.
trombo3
trombo1
trombo2
trombo4
[16] S. Wilson, D. Cottle, and N. Collins, eds., The Su-
trombo6
trombo8
trombo10
perCollider Book. Cambridge, Mass.: The MIT Press,
armonica1
armonica2
armonica3
2011.
armonica4
armonica5
armonica6
armonica7
[17] A. Valle, “Integrated Algorithmic Composition. Fluid
ocarino2
sirenetto2
ancetto
Systems for including notation in music composition
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
cycle,” in NIME 2008: Proceedings, pp. 253–256,
2008.
Figure 23. Visualisation score for electromechanical de- [18] Adobe, PostScript Language Reference. Reading,
vices in Reocerantroma phenaudi. Mass.: Addison-Wesley, 3rd ed., 1999.
[19] H.-W. Nienhuys and J. Nieuwenhuizen, “LilyPond, a
producing behaviours. system for music engraving,” in Proceeding of the XIV
CIM 2003, (Firenze), pp. 167–172, 2003.
6. REFERENCES
[20] H. H. Hoos, K. A. Hamel, K. Renz, and J. Kilian,
[1] N. Chomsky, Syntactic Structures. The Hague: Mouton “The GUIDO Notation Format – A Novel Approach
and Co., 1957. for Adequately Representing Score-Level Music,” in
ICMC’98 Proceedings, pp. 451–454, 1998.
[2] M. A. Boden and E. A. Edmonds, “What is generative
art?,” Digital Creativity, vol. 20, no. 1-2, pp. 21–46, [21] P. Roullier, ed., Mauro Lanza et Andrea Valle:
2009. Systema naturae. Champigny-sur-Marne: Ensemble
2e2m, 2016.
[3] O. Messiaen, The Techniques of my Musical Language.
Paris: Leduc, 1956. [22] A. Valle and M. Lanza, “Systema naturae: shared
practices between physical computing and algorithmic
[4] A. Orcalli, Fenomenologia della musica sperimentale. composition,” in Proceedings of the 14th Sound and
Potenza: Sonus, 1993. Music Computing Conference (J. P. Tapio Lokki and
[5] F. Paris, Le temps de l’écoute. Gerard Grisey ou la V. Välimäki, eds.), (Espoo), pp. 391–398, Aalto Uni-
beauté des ombres sonores, ch. L’empreinte du cygne, versity, Aalto University, 2017.
pp. 51–65. Paris: L’Harmattan, 2004. [23] D. O’Sullivan and T. Igoe, Physical Computing. Sens-
ing and Controlling the Physical World with Comput-
[6] K. Stockhausen, “. . . how time passes,” die Reihe,
ers. Boston, Mass.: Course Technology, 2004.
vol. 3, pp. 10–43, 1957.
[24] M. Banzi, Getting started with Arduino. O’Reilly,
[7] C. Roads, Microsound. Cambridge, Mass. and London:
2009.
The MIT Press, 2001.

CIM 2 018 P R OC E E DING S 135