Patrick Greussay

Jacques Arveiller Musical Software:

Marc Battiler
Chris Colere Descriptions and
Gilbert Dalmasso
Giuseppe Englert
Abstractions of
Didier Roncin
Groupe Art et Informatique Vincennes
Sound Generation
and Mixing
Universit6 Paris VIII
Paris, France

1. Introduction tions have been based on the Vincennes version of

LISP, VLISP (Chailloux 1975) and the low-level lan-
Software research for musical applications at the guage INTELGREU.
Groupe Art et Informatique Vincennes (GAIV) has The set of process descriptions we will present
been devoted almost exclusively to the interaction has not been based on theoretical presuppositions.
between the software system and the interpreter/ Rather, it has been formulated on the basis of expe-
composer (IC). We are especially interested in the rience gathered from the use of our systems in a
real-time aspects of this interaction. Naturally we performance setting, which has allowed us to grad-
have been led to consider this interaction in terms ually correct and perfect these systems, as well as
of: from analysis of musical works from the past in
1. Process terms of data flow and data base (Greussay 1973).
2. Hierarchies of and interactions between Performance experience seems indispensable to us
for the verification of software tools, abstract or
processes
3. Data flows real. Indeed, contrary to the studio situation, we are
faced immediately with-the consequences of our
4. Paths for and operators on data flows
decisions: from the outset, unrealistic or unfeasible
We feel that the IC itself represents a process which viewpoints must be eliminated. Furthermore, we
encounters, interacts with, and controls other pro- can experiment with the actual representation of
cesses. These processes can be other ICs, or system musical processes employed by professional musi-
processes defined in the software. cians, following the lead of experiments on the
This type of approach to musical software is representation of musical processes in children car-
related to other current research, especially that ried out in the LOGO laboratory at M.I.T. (Bamber-
dealing with data flow languages (Dennis 1974), ger 1976). Analysis by computer of musical works
communication among parallel processes (Atkinson from the past allows us to verify our concepts in
and Hewitt 1976; Hewitt and Baker 1977), and the compositions. One of the theories supported by
description and formalization of coroutine net- GAIV is that a successful composition is, in a cer-
works (Kahn and McQueen 1977). We will describe tain sense, an improvisation (Dalmasso 1980). In
new kinds of operators and configurations of pipe- this way we are trying to think, in computer sci-
line stages that have been implemented at GAIV as ence terms, of the processes of (score) reading,
a result of the musical and interactive character of foreshadowing of later events in a piece (or the lack
the processes we have studied. Our implementa- of it), improvisation, planning, and composition.
Finally, we should add that we are not trying to
Computer Music Journal, Vol. 4, No. 3, Fall 1980, establish a theory of musical/instrumental pro-
0148-9267/80/020040-08 $04.00/0 cesses. To be sure, we shall, without exception, use
@ 1980 Massachusetts Institute of Technology. a schematic representation to describe data and pro-

40 Computer Music Journal

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 1. Total dependence tween system processes Fig. 2. Two possible pat-
(top) between an interpre- and sound generation de- terns of total dependence:
tericomposer (IC) and one vices (SG). (a) the noninteractive case;
or more system processes (b) the interactive case.
(SP), and (bottom) be-

IC IC IC

SP SP SP ... SP SP SP SP

SP SP SP ... SP SGSP

(a) (b)
SG

tion are determined in advance. The control of SP

by IC, and of SG by SP, is total; no problems with
cesses. But the formal descriptions we will discuss scheduling or synchronization can occur. The SPs
represent nothing more than abstractions of soft- are conceptually independent and cannot commu-
ware tools that are being planned or have actually nicate among themselves. Whatever the number
been completed. Our only purpose, in fact, is to (necessarily finite) of SPs, the potential relation-
make it easier for musicians at GAIV to approach a ships can be reduced to two simple patterns. Figure
computer science that deals with the interactive 2(a) shows the noninteractive case (e.g., for a com-
use of computers. position realized in the studio). For total depen-
Thus in the following brief presentation we pre- dence in the interactive case (Fig. 2[b]), the IC
fer to emphasize the fundamental concepts on control should operate on no less than two SPs be-
which our software is based, rather than specific de- cause at least two decisions are necessary. This
tails of implementation. The latter are described in kind of control was used in the KRWTH system
detail in the internal reports of the University of (Chailloux 1975). Note that for the sake of sim-
Paris VIII and in our journal Artinfo/Musinfo. plicity we show only one SG. Actually, several SGs
are connected by the abstract mixing operators,
which we will describe later (Section 3).
2. Abstractions of Dependency Relations in
Sound Synthesis
2.2. Levels of Hierarchical Relationships (IC SP)
We will first describe certain kinds of abstract de-
pendency relations that can occur between (1) an We have here the case where the layer of SPs at
IC; (2) one or more system processes (SP), for exam- level ie[1, n] can control or create the layer of SPs
ple, processes for choosing overall frequency, den- at level i+ 1 (cf. Fig. 3). Examples of this would in-
sity, or amplitude; and (3) one or more devices for clude software for aleatoric synthesis, and pattern
the generation of sound (SG). generators (cf. Fig. 4). We see that the dependency is
always hierarchical, that is, it operates from layer
to layer. Furthermore, the SPs of a single layer can-
2.1. Total Dependency not interact with one another. Finally, the number
of layers is determined by the program in a fixed
This kind of relationship is a characteristic feature and definitive way, with only one IC at the top
of noninteractive musical software. As can be seen level. Just as in Section 2.1, hierarchical relation-
in Fig. 1, all of the processes for the entire composi- ships are not limited to interactive musical situa-

Greussay 41

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 3. Hierarchical rela- Fig. 4. An elementary ex-
tionships among various ample of a pattern genera-
levels of SPs. tor in VLISP. All possible
recursive compositions can
be created using these two
elementary patterns.

IC
(DE GENPAT (PAT A B C D E F)
(LET ((PAT PAT)) (COND
((NULL PAT) (NIL)
((ATOM PAT)
SP SP Level 1 (APPEND (EVAL PAT) NIL))
((NUMBP (CAR PAT))
(GENCOP (NEXTL PAT) (SELF PAT)))
((EQ (CAR PAT) ',)
SP SP SP SP Level 2 (GENCOP (EVAL (CADR PAT))
(SELF (CDDR PAT))))
(T (APPEND
(SELF (NEXTL PAT))
(SELF PAT))))))
(DE GENCOP (N SQ)
(LET ((N N))
SG (IF (= N 0) NIL
(APPEND SQ (SELF (SUB 1 N))))))
(GENPAT '(A B A) sq, sq2) sq, sq2 sqi
(GENPAT'(n. A) sq) sq sq... sq
<- n TIMES -*
tions and can be summarized in the two patterns
shown in Fig. 5.

cial IC-SP communication terminals featuring a

2.3. Mutually Dependent (Horizontal) handler for various interrupts and a keyboard, along
Relationships with an inter-IC communication system consisting
of a common clock, display screens, and light sig-
There is also the possibility of communication and nals. Having gathered some experience at a concert
dependency within a single layer, The threevari- in Frankfurt (in April 1977) with a special color TV
eties of horizontal relationships, shown in Fig. 6, terminal constructed at the University of Paris VIII
were used in the program RE COSA MATERIALE by Louis Audoire (1976), GAIV is now planning to
(Battier 1977). generalize this kind of inter-IC communication.
For lack of space, we cannot give specific, de-
tailed examples of each type of dependency. We
should at least point out, however, that formally 2.4. Recursive Dependency
speaking, these relationships among a fixed number
of SPs at a given level constitute a network of finite In this kind of relationship (Kahn and McQueen
automata. If the SPs are ICs, we have the kind of 1977) we are confronted for the first time with a
interdependence characteristic of group improvisa- layer that exists separately from the SP (Fig. 8). As a
tion (Dalmasso 1980). Jacques Arveiller used this matter of fact, an SP in layer i can depend on sev-
kind of organization in the computer program for eral distinct SPs in the layer just above it (i- 1).
Paire-Lacs, created at Tours in April 1976. In this Furthermore, the number of SPs is no longer
kind of organization we must deal with the ques- fixed in advance; a recursive SP call can produce a
tion of synchronization and scheduling due to the new SP network. Since the capacity of memory and
limited number of available resources (Battier process resources is finite, we must also introduce a
1977). In addition, if the upper layer is constructed new kind of process, known as reclaiming, which is
of more than one IC, then the system must use spe- responsible for "garbage collection." This extremely

42 Computer Music Journal

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 5. Hierarchical rela- case (b) at least two IC Fig. 6. Dependency
tionships. The noninterac- decisions are necessary through a "closest neigh-
tive case (a) results in a (Menard 1974). bor" bilateral relationship
single and specific deci- (a). Hierarchical depen-
sion. In the interactive dency (b). Auto-depen-
dency (c).

IC IC

PS PS PS

PS PS PS
SP SP SP . ... SP (b)

PS PS

SP SP SP S P- (c)
a SG
(a) (b)
is abolished, which necessitates the construction of
completely new link terminals. Furthermore, the
interesting organization introduces strict limita- relationships among the SPs are no longer fixed but
tions on the creation of music in real-time or in a can be modified within the network through the
performance situation. We must use techniques of propagation of constraints (as shown in Fig. 10).
"incremental garbage collection" to allocate a new This kind of relationship introduces the concepts of
SP within some given length of time, regardless of distributed monitors, perturbation monitors, and
the number of inactive SPs to be recovered (Baker fracture monitors, which are outlined in Englert's
1977). article (1977).
Finally, we should point out that this kind of re- We can consider this last kind of organization as
lationship between SPs has been found through the the most interactive, compatible with current soft-
analysis of the structure of improvisation (Dalmasso ware/hardware tools. It introduces extremely dif-
1980), as well as analysis of compositions that are ficult and interesting problems of process mainte-
apparently quite removed from our interactive ap- nance. At the same time it is in all probability the
proach (Greussay 1973). model closest to what we know about the interac-
tions among human musicians.
At this juncture, we should point out that the
2.5. Independence of Layers: Total Parallelism interactive aspect of computer composition is char-
acterized by the control of perturbation propaga-
At Vincennes, lacking hardware multiprocessors, tion. Our next step will be to experiment actively
we have only been able to experiment using soft- with more complicated relationships than AND
ware simulation (Greussay 1978). Although the networks and OR networks. Louis Audoire and Di-
possibilities are quite complex, we will discuss the dier Roncin have constructed specialized inter-IC
two simplest organizations, the AND network and terminals adapted to this kind of organization.
the OR network (Fig. 9). In these two cases we see These include a terminal with a color screen and a
first that the logical distinction between SP and IC VLISP system on an LSI-11.

Greussay 43

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 7. Interconnection of Fig. 8. Recursive depen- Fig. 9. AND network (a), in network (b), in which the
ICs. dencies (see text for which the output of the SP adjacent SPs depend alter-
explanation). in the lower right-hand nately (the simultaneity
corner depends on the si- then being a special case)
multaneity of interaction on the SP in the middle.
with the adjacent SPs. OR

[~7=1- - - PS PS opP

(a)

PS

ei e2 e3

e, e2 e3 V V V
PS PS PS

PSPPS
S

PS PS PS

3. Abstractions in Mixing

All of the GAIV programs incorporate the features 3.2. Data Flow Pipeline Stages
discussed in this section. The following description
deals essentially with the relationships between The introduction of a pipeline stage allows us to
SGs (more specifically, the program derived from use SGs in a quasi-parallel fashion. A SG can pro-
the montage Echologique). The same description duce a datum and place it into a pipeline stage
can likewise apply to the relationships among SPs, without necessarily taking into account the SGs
but here (as opposed to the relationships described that are going to use the datum. Introduction of
in Section 2) we are limited to a fixed number of a pipeline stage thus allows for the data synchroni-
SGs. zation of processes. The data-flow pipeline stages
Likewise in contrast with the inter-IC or inter-SP are known as cells. The Bi-Sequencer program
relationships, which are dependency relations, the (Greussay 1978) allows the use of an arbitrary num-
abstract relationships in mixing, which involve the ber of independent processes synchronized through
SGs, operate on streams of sequential data repre- pipeline stages.
senting the characteristics of sound. All of the A most common example of data-flow synchro-
abstractions relevant to mixing can be summarized nization is the time loop, shown in Fig. 13.
in the diagram shown in Fig. 11. We distinguish four configurations, ranked ac-
cording to increasing complexity and shown in Fig.
14. If no pipeline stage is present, the data flow
3.1. Data Flow Operators passes immediately from one SG to the next. With
a single intervening pipeline stage (Fig. 14[b]), SGB
The two forms of data-flow operators are shown in takes its input datum from cell c where the datum
Fig. 12. These are, of course, the usual relationships of SGA is deposited. No time sequence is fixed a
found in oscillator networks. priori between SGA and SGB. As a matter of fact,

44 Computer Music Journal

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 10. Example of a po- Fig. 11. Abstraction of Fig. 12. Data-flow opera-
tential change in a totally mixing. tors: (a) interconnection;
parallel configuration. (b) feedback.

PS PS PS
e-data-flow,

SG s-data-flow
PS P S PS
e-data-flown
Modulator

P PS PS

PS PS PS SGA

SGA SGB ; e (a)

PS P-PS S SGB

PS PS PS

SG M SG (b)

SGA can produce its datum without SGB being ac-

tive, or SGB can be active and SGA inactive. A
request for data by SGB can also be used to activate
SGA.
If an SG is logically decomposed into n sub-SGs,
n cells are necessary, as shown in Fig. 14(c). Each of 4. Conclusion
the SGAS can independently process an output
datum. According to its.needs, SGB will take its in- At the present time we are interested in developing
put datum from cells cl, c2, c3, and/or c4. software adapted to networks of microprocessors.
If a queue is placed between the two SGs, SGA For us the elimination of the simulation of these
can theoretically produce an infinitely long se- networks in software has become a primary goal,
quence of data, which would fill an infinite number not only because simulation is obviously less
of cells. According to its needs, SGB uses the ele- efficient but also because we want to make use
ments of this sequence in first-in, first-out (FIFO) of our special visual terminals. Perhaps a bit para-
order. In practice, since the total number of cells is doxically, we feel that further progress in the de-
limited, SGA must stop producing data and tem- velopment of a computer science.dealing with the
porarily deactivate itself when the limit of the interactive, musical use of computers rests on the
number of available cells is reached. SGA will be re- development of these new visual terminals in all
activated when SGB, by taking an input datum, cases where several ICs must interact during the
makes cells available. performance of a single piece in real-time.

Greussay 45

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
Fig. 13. Data stream syn- Fig. 14. Four ways of con-
chronization. The output figuring SGs with pipeline
datum si+n derives from stages. No pipeline stage
the input datum li, si + n -1 (a). A single pipeline stage
(b). Multiple stages (c).
derives from li- , and so
on. Queue (d).

Fig. 13. 1i-2 Ii-1 i Si+n-2 Si+n-1 Si+n

I--- *----+ *----+-----> SGT ---+--) *----+~ *---i-+ *----+-

Fig. 14.

(a) (b) (c)

SA(a) SGA

S - - - - sequential operator At A2 I A3 A4

S (b) ET
4- - - - - cell

Acknowledgment Bamberger, J. 1976. "Children's Representations of Pitch

Relations." Memo No. 43. Cambridge, Massachusetts:
The editors of Computer Music Journal would like M.I.T. LOGO.
to express their appreciation to James A. Moorer for Battier, M. 1977. "RE COSA MATERIALE et le pro-
gramme compositionnel ICOSA." Artinfo/Musinfo
his help in preparing this manuscript. 27:21-45.
Chailloux, J. 1975. "KRWTH." ArtinfolMusinfo 20:1-16.
References . 1976. UER Informatique. RT 17-76a. Paris: Uni-
versite Paris VIII, Vincennes.
Atkinson, R., and Hewitt, C. 1976. "Parallelism and Syn- Dalmasso, G. 1980. "Musique et improvisation: pro-
chronization in Actor Systems." Draft. Cambridge, cedures de descriptions symboliques d'environments
Massachusetts: M.I.T. Artificial Intelligence Laboratory. dynamiques." Ph.D. disseration, Universit6 Paris VIII.
Audoire, L. 1976. Colorix: un peripherique de visualisa- Dennis, J. B. 1974. "First Version of a Data-Flow Pro-
tion couleur. M6moire de Maitrise d'Informatique. cedure Language." In Colloque sur la programmation,
Paris: Universit6 Paris VIII. ed. B. Robinet. New York: Springer, pp. 362-376.
Baker, H. G., Jr. 1977. "List Processing in Real Time on a Englert, G. G. 1977. "Fragola." Artinfol/Musinfo 22:1-8.
Serial Computer." Working Paper No. 139. Cambridge, Greussay, P. 1973. "Descriptions de proc6dures de lec-
Massachusetts: M.I.T. Artificial Intelligence Laboratory. ture." Artinfol/Musinfo 19:1-30.

46 Computer Music Journal

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms
 . 1978. "Mod1lisation de r6saux de multi-s6quen- Kahn, G., and McQueen, C. B. 1977. "Coroutines and
ceurs." Artinfo/Musinfo. Networks of Parallel Processes." Proc. IFIP 1977.
Hewitt, C., and Baker, G. 1977. "Laws for Communicat- Amsterdam: North Holland, pp. 993-997.
ing Parallel Processes." Working Paper No. 134A. Menard, P. 1974. "Pr6sentation du programme AUTO-
Cambridge, Massachusetts: M.I.T. Artificial Intel- MUSE." ArtinfolMusinfo 15:13-24.
ligence Laboratory.

Greussay 47

This content downloaded from 130.102.42.98 on Sat, 07 May 2016 00:42:45 UTC
All use subject to http://about.jstor.org/terms