Using cybernetics as the study of control mechanisms, this paper sketches an analysis of mechanical and computer-based tools for producing music, in terms of their control mechanisms and the relations that they imply between musicians and their tools. After defining control mechanisms and their effect on the use and evolution of tools in general, the paper applies these concepts to mechanical musical instruments and then to computer-based music systems in an attempt to understand the current state in the evolution of computer-based tools for music production.1
Cybernetics, taken in one of its simplest senses, is the study of control mechanisms and their effect on the relations between people and their tools. Using that definition, I have attempted to sketch an analysis of some tools for producing music, in terms of their control mechanisms and the relations that they imply between musicians, instruments, and computer-based music systems.
For the purpose of this analysis, a tool is composed of two parts: an effective mechanism, which directly accomplishes the purpose for which the tool is employed, and a control mechanism, which mediates between the person using the tool and the effective mechanism.
In fulfilling its function, the control mechanism acts as an analogue of the effective mechanism. For example, the effective mechanism of a knife is its edge, and the control mechanism is its handle. Whatever forces are applied to the handle are analogously applied to the edge.
The analogy between the effective and control mechanisms imposes a distance between them, and also between the user of the tool and the effective mechanism. In many cases, this distance makes the tool usable: we cannot grasp the edge of the knife to use it; we are neither large enough nor strong enough to manipulate the parts of a construction crane directly; we are not fast enough to feed data to the CPU of a computer.
While the effective mechanism of a tool imposes limits on its usability through its physical characteristics, such as sharpness, size, or speed, what is far more important for the control mechanism is the extent to which it is a complete analogue of the effective mechanism. A vehicle with wings and a jet engine, but with only a steering wheel, a brake, and an accelerator, would not be an airplane; it would be an automobile, and a clumsy one. Nothing can be accomplished with the effective mechanism that is not provided for in the control mechanism as its analogue.
Every tool is continuously involved with the society in which it is used in a mutual process of shaping and evolution. The evolution of each tool is steered by a socially defined model that defines its use. In order for a new tool to evolve, a new model must first exist in the mind of at at least one person who wants to use that new tool. And the use of a tool according to its model creates a new social environment that may create desires for a change in how the tool is used, and thus in the tool itself.
The examples which are most characteristic of this process are, unfortunately for us all, weapons. But, fortunately for the author, who might otherwise have encountered insuperable difficulty in fitting his ideas into the allotted space, the case is so manifest as to require no explication here.
Mechanical musical instruments: strings, woodwinds, brass, and percussion, can be analyzed in the same terms as other tools. For instance, the effective mechanism of a trombone is a column of air, confined in a brass tube, that is excited by the vibrating lips of the player. The control mechanism for the pitch produced by the instrument has two parts: (1) the lips of the player, acted upon by the tension of the surrounding musculature and by the air pressure applied by the player's respiratory system, excites the tube to vibrating, primarily at the frequency of one of the harmonic partials of the tube's fundamental frequency; (2) the fundamental is determined by the length of the tube, and so the length of the tube must be changed to select different fundamentals.
There are two methods of changing the length of the tube: a slide and a set of valves; and the two methods define two different instruments. That they are two different instruments can be demonstrated by imagining a player attempting to play the scales in the opening scene of Verdi's Othello on a slide trombone, or the glissandi in Bartok's Concerto for Orchestra on a valve trombone.
Musical instruments, like all tools, are involved in a continual process of evolution with the needs and expectations of the society in which they are used. While the evolution of some instruments, such as the strings, has been primarily in terms of the technological improvements of their effective mechanisms, the evolution of many instruments, such as the winds, has been primarily in terms of the conceptual and technological improvement of their control mechanisms.
To take the clarinet as an example: from its historically hazy invention by J. C. Denner of Nuremberg around 1700, the evolution of the clarinet has been primarily through placing and adding holes and keys and levers to permit more correct intonation and to facilitate more rapid and sure passing from any pitch to any other. (Space does not permit the discussion of the two competing control systems that have evolved for the clarinet. Suffice it to say that they are, to everyone but a clarinetist, essentially equivalent.)
The influence between society and the clarinet has been both mutual and circular: with improvements in the clarinet's ability to play more or less in tune in all keys and to play chromatically, composers were encouraged to use it in new ways; the clarinet parts of Til Eulenspiegel could not have been conceived for a clarinet of the early 19th century. Conversely, Til as a composition is inconceivable without clarinets. And the existence of works like Til encouraged, via performers' demands and suggestions to instrument makers, further improvements in the clarinet's control system through the 20th century to today.
As with all instruments, the development of the control systems of the woodwind instruments has been in accord with a model that remained virtually unchanged until the middle of the 20th century. That model considered a woodwind as producing one pitch at a time by exciting its air column with a vibrating reed or by blowing over an open hole. The timbre was to be recognizably uniform throughout the instrument's compass. Any other kind of sound produced by the instrument, including multiphonics or the clicking of the keywork, was considered a regrettable by-product of a necessarily imperfect physical implementation.
Compositions for woodwinds were really compositions for this model, and the technical judgment of a performance was predicated on the ability of the performer to play in conformance with the model. In that sense, the performer performed music using the instrument and, at the same time, performed the social definition of the instrument.
Beginning in the late 1950s, composers and performers recognized that musical instruments were capable of producing sounds that did not lie within the previously accepted social models of the instruments. They began to compose and perform using any sound that a given instrument could be reliably demonstrated to produce. This represented an attempt to explore, not merely to master, the instruments. And these explorations have not only produced new social models of each instrument, but a new meaning of ``musical instrument.''
Depending on the instrument, these explorations met with varying success. This was to some extent a function, for each instrument, of the extent to which its control mechanism, which had evolved under a more restricted model, accidentally permitted the exploitation of these aspects of its effective mechanism. For example, the number of different multiphonics that can be produced on a woodwind instrument is a function of the independence with which the holes in its tube can be closed. In the case of the flute, nearly all of the holes can be independently closed. In contrast, the mechanism of the saxophone precludes opening certain holes while others are closed, to facilitate certain common passages and trills in the keys in which it was normally played.
In the case of key clicks, a characteristic of the implementation of the control mechanism, viewed as a defect under the previous model, is viewed under the new model as a useful function of the effective mechanism of the instrument. The extent to which the bassoon produces the loudest and most varied key clicks, once seen as a necessary evil by builders, performers, composers, and listeners, can now be welcomed as a virtue by all. This rivals the transformation wrought by the marketing experts who turned waxed paper sandwich bags into ``microwave convenience bags.''
Assuming that further evolution of mechanical musical instruments will not be precluded economically by the advent of electronic instruments, what form will this evolution take in response to the new social definitions of the instruments? While it is unlikely that woodwind instruments will be designed to exploit key clicks, is it unreasonable to expect designers to take facility in producing multiphonics into account? Is there a new Adolphe Sax engaged even now in constructing a ``Multiphone?''
In analyzing mechanical instruments, it was reasonably easy to identify and distinguish their effective and control mechanisms. But even if we confine our study of computer-based music systems to delayed performance systems that employ a general-purpose computer to compute waveform samples for digital-to-analog conversion, our analysis encounters inherent difficulties.
The invention of the computer was a principal impetus for the development of the study of cybernetics, because it is a tool that can be used to recursively redefine its own control mechanism (the software). Each layer of control mechanism is built on, and in terms of, the layer below. More important, each layer acts as part of the effective mechanism for the layers above, and as part of the control mechanism for the layers below. And all but the lowest of these layers (the hardware) can be created and modified by the same computer on which they are to operate.
These are the properties that have made the computer, which only a few decades ago was usable by only a handful of mystical practitioners to solve a restricted set of problems, the archetypical tool of our time. But because of these properties, the analysis of a computer system into control and effective mechanisms is by no means obvious.
The key to the analysis is not what is being analyzed, but where it lies in the structure and when we are looking at the process in which it takes part. We can then appeal to our definitions: when a part of a computer system is engaged in directly accomplishing the purpose for which the computer is employed, it is part of the effective mechanism. And when a part of a computer system is engaged in mediating between the person using the system and the effective mechanism, it is part of the control mechanism.
What, then, are the effective and control mechanisms of a computer-based music system? The digital-to-analog converter and the analog electronics clearly are part of the effective mechanism. So too are the software and hardware for storage of the samples and for delivering them to the converter at regular intervals. And the software embodying the algorithms for computing the samples must also be considered part of the effective mechanism. Everything else, including the software and the hardware whereby the user defines, selects, and directs the sample computation algorithms, must be classified as the control mechanism.
The pioneering and still dominant family of delayed-performance computer music systems is of course the ``Music N'' family, initiated in the early 1960s by M. V. Mathews of Bell Telephone Laboratories and epitomized by Music V.2 The Music N family of systems has been described and has been subjected to detailed critique elsewhere by Loy and Abbott;3 the description of the score and of the control mechanism given below is a gross simplification intended to be the minimum necessary to support the discussion.
The input to a Music N system consists of a score: an ordered set of statements that define instruments, generate functions, or specify notes.
An instrument consists of interconnected unit generators including oscillators, envelope generators, adders, multipliers, random generators, and filters, with the output of one becoming the input of another. New instrument definitions can be introduced at any time during the generation of a piece.
A Music N function definition specifies the shape of a waveform to be produced by the oscillators that use it beginning at a given time.
A Music N note specifies the invocation of an instrument, with a given setting of the inputs for the unit generators that compose it, at a specified time for a specified duration.
According to the distinction established in section 4, the Music N control mechanism comprises the software for reading the score, encoding the definitions and specifications, sorting the defining and specifying events in order of increasing start time, and interpreting the sorted list in time order to set up and invoke the computation algorithms at the specified times for the specified durations.
From the composer's point of view, the model of Music N consists of sets of interconnected modules that can come into being at specified times. These sets of modules include oscillators that can change their wave shapes at specified times. Each set can be turned on at specified times for specified durations with specified settings of the modules' controls.
This model has served to make the Music N systems predominant among delayed-performance computer music software. How did this model evolve?
Loy and Abbott4 have cited a number of factors that have contributed to the success of the Music N systems, including the simplicity and power of its interface, and the efficiency of implementation that is implied by the interface model. The evolution of tools suggests additional factors that may have contributed to the widespread initial acceptance of these systems: Music N met an immediate need with a tool that conformed to a familiar model.
In the electronic music studios of the early to mid 1960s, voltage-controlled modules were in the process of becoming the standard. They were particularly successful because of the inter-compatibility of their control systems, which permitted the output voltage of any module to be used as the control voltage for any other module. For instance, a composer could produce an FM signal by simply connecting the output of one oscillator to the frequency control of another, or could produce an AM signal by connecting th output of an oscillator to the amplitude control of an amplifier.
There were three practical difficulties that composers encountered in using a studio build of these modules: (1) there were never enough modules or patch cords to set up the most complex configurations that the composer required; (2) it was tedious to change the setup of inter-connections and knob settings from one configuration to another; and (3) the stability and accuracy of the voltages output by the modules under less than ideal environmental conditions was sometimes a source of frustration.
Whether or not its designers intended it to do so, Music N solved every one of these problems by transporting the voltage-controlled studio model from the analog domain to the digital-to-analog domain. And the very composers most likely to use a computer-based music system were those who were already familiar with the voltage-controlled electronic music studio.
But the Music N family has been predominant for well over two decades, without undergoing substantial change other than in its implementation language and its hardware base. Why has there been so little change in these systems, when computer software systems in general have evolved so radically during the same period?
The factors cited by Loy and Abbot must contribute to this state of affairs; Music N is an elegant, powerful system that is easy to learn. Certainly these are necessary conditions for the evolution of a system to stop. But they hardly seem sufficient, given the rapid evolution and differentiation of other kinds of software systems.
Could it be that the overwhelming prevalence of Music N is in some way responsible? Music N is taught in virtually every university course on computer music. Hardly a book is published on computer music that does not contain a substantial section that treats with it. Just as ``IBM machines'' were once the socially accepted model for all computers, are Music N systems the socially accepted model for all computer-based music systems?
Sawdust5 was conceived by Herbert Brün, designed and implemented by the author, and enhanced by Jody Kravitz and Keith Johnson. The original version ran under the Unix time sharing system on a PDP-11/50 at the Center for Advanced Computation of the University of Illinois, using a D/A interface designed and built by Jody Kravitz.
The composer using Sawdust works in terms of waveforms specified by sample amplitude values and the number of sample intervals during which each is to be held. These waveforms can be made to interact with one another according to pre-defined algorithms to produce sound events with continuously varying characteristics.
The composer interacts with Sawdust via a standard time-sharing terminal. During a session, the composer can define and edit objects, which specify waveforms and how they are to interact, and can interactively play the objects to hear the results of the specifications. As the composer defines each object, he names it so that it can be invoked to be incorporated in the definition of other objects or to be played. The composer can edit, delete, and redefine objects during the session. The state of a session can be saved so that the composer can resume work in another session. Thus the composer can incrementally build up his piece over one or several sessions, and then record it all at once.
The simplest object in Sawdust is the element, which consists of an integer amplitude as accepted by the D/A converter and the integer number of sample periods for which it is to be held. An element cannot produce sound by itself, because the result of playing it would be a DC signal.
Elements can be combined into a link, which defines an ordered list of objects that are to be played one after another and specifies the number of times the list is to be repeated. When a link is played, each object in the list is played in order and the list is repeated according to the specification. For instance, when a link consisting only of elements is played, it produces a static waveform whose duration is the link's repeat specification multiplied by the period of the waveform.
Mingle and merge objects define two levels of interleaving list of objects. When a mingle or merge object is played, each of its component objects remains part of the resulting sound according to its repeat specification.
Grow and vary objects define two ways of producing events that begin with one waveform and end with another. Both are intended to operate on a specified first link and last link, each of which consist only of elements. The duration of either kind of object is dependent on its repeat specification.
When a grow object is played, Sawdust first plays the waveform defined by a single iteration of the specified first link. It then randomly changes, within specified limits, the amplitude values and sample hold values of each of the elements, and plays the result. It continues this process for the number of iterations specified by the repeat count. For the last iteration, the specified last link is played.
When a vary object is played, Sawdust selects polynomials that connect the amplitude value of each element of the first link with the amplitude value of the corresponding element of the last link. It does the same for the sample hold values of the corresponding elements of the two links. After first playing a single iteration of the first link, it computes and plays a new link at each iteration, according to the evaluation of the polynomial for each amplitude and sample hold value, for the number of iterations given as its repeat specification.
The Sawdust control mechanism comprises the software for interacting with the composer, encoding and storing the object definitions, and invoking the appropriate play algorithms for the objects.
Just as key clicks, artifacts of the implementation of a woodwind instrument, have become an accepted part of our model of what a woodwind instrument can do, Sawdust was, in one sense, an attempt to exploit an artifact of the implementation of computer-based music systems. If a computer-based music system produces sound by feeding samples to a digital-to-analog converter, then why should that not be the terms in which the composer confronts the system?
Seen from the point of view of its principal designer and implementor, Sawdust succeeded all too well in answering this question. The determination and entry of all of the amplitude and sample hold values necessary to produce a piece of any complexity has always seemed tedious to one who was not the system's principal user. A more serious drawback of this low-level approach is that a waveform's period is always an integer times the sample period, and, consequently, the frequencies of audible waveforms are limited to the set of rational numbers formed by dividing the sampling frequency (40KHz) by the integers between 2 and 2,000.
At a higher level, Sawdust was a response to a desire to compose directly with waveforms, to produce pieces through the interaction of composer-specified waveforms with each other via predetermined algorithms. Sawdust has been used to compose a number of pieces that it seems unlikely could have been produced in any other way.
Sawdust was also a response to the practical question: How can a usable interactive music system be built with a time-shared 16-bit minicomputer as the computing resource? From this point of view, Sawdust was an unqualified success, with a worst-case compute time/real time ratio of 5 to 1 on a PDP-11/70.
If Music N constitutes the model for computer-based music systems, then what role do systems like Sawdust play in the evolution of computer-based tools for music production? Just as composers' and performers' desires drove the explorations of mechanical musical instruments that resulted in an expansion of our model of what these instruments can do, will similar desires continue to drive the exploration of computer-based music systems to define a new model of what these systems can and should be?