There's a growing trend in the world of electronic-music performance. Next time you go to a show, chances are you'll see a laptop onstage, functioning as a performer. In interactive composition and performance, control of a piece includes a computer that has been programmed to sense significant musical features from a human performer and produce its own music in response.

Nothing about the idea is new — people have been writing and playing interactive works for more than 25 years. But the pioneers worked for institutions that could spend hundreds of thousands of dollars on specialized computer systems. Now that PCs are intertwined with everyday life, interactive music systems have trickled down to the proletarian sphere of individual musicians. In this column, I'll take a brief look at the evolution of interactive music systems and give an overview of some performance approaches that are commonly used. (See the sidebar “References and Recordings” for additional resources.)

SOME RECENT HISTORY

By the end of the 1960s, Max Mathews, the father of computer music, was increasingly dissatisfied with the music that computers were producing. Music created from coded scores was dry and lifeless. In an effort to transmit micromodulations — the uncountable variations in embouchure, bow position, breath pressure, and so on, that give live music a dynamic dimension — Mathews began the pursuit of what he called the intelligent machine that would respond to performers' nuances. Conductor was an early system that Mathews designed for Pierre Boulez, who was the musical director of the New York Philharmonic at the time. Boulez was enthusiastic about electronic elements in performance, but felt constrained by having to follow a tape. The Conductor system allowed electronic elements to be dynamically controlled by external devices such as joysticks and percussion instruments.

In 1977, composer Joel Chadabe snatched the first Synclavier off the production line and had it outfitted with special software that created melodies based on predefined parameters such as harmony and interval content. The Synclavier was interfaced with two modified theremins. One antenna controlled the tempo (note durations), while the other controlled relative volumes of four Synclavier voices (in effect, overall timbre). Chadabe wrote that performing with the system was like having “a conversation with a clever friend.” He could do things like cue clarinet sounds to play slowly; but since he did not know which pitches would play, the notes he heard then influenced his next control gesture.

Meanwhile, at Boulez's research brainchild, IRCAM, in Paris, work was under way on a digital signal-processing computer that was capable of any synthesis configuration as well as real-time audio processing. The 4X workstation was completed in the early 1980s and was like nothing the world had ever seen. Miller Puckette created a Macintosh-based interface for the 4X in which processes and controls were represented graphically. Patches could be created by drawing patch cords between modules, and processing algorithms could be switched on and off by various gates. He named the program Max in honor of Mathews.

Max was later ported to the NeXT personal computer, where it could be run with the help of peripheral hardware processors in a configuration called the ISPW (IRCAM Signal Processing Workstation). Though far more economical than the 4X, the ISPW remained a pricey hardware-software combination. Max was then released commercially as a kind of erector set for MIDI input, processing, and output and is now under active development by Cycling '74 (the sidebar “On the Web” provides URLs for all the developers mentioned in this article) for both the Mac and Windows computers. The tools for interactivity were now within the means of independent musicians.

SAY WHAT?

So what is meant, exactly, by machine responses to a human player? Author-composer Robert Rowe classifies interactions into three broad categories. The first concerns the type of “listening” a computer is doing. The second describes the computer response types. The third describes the nature of the partnership between performer and computer. As for listening, computers can listen generally or specifically.

General listening means that the computer senses general characteristics such as register, loudness, or density. Specific listening can come in two forms. One, score following, involves moment-by-moment estimations of a performer's tempo. One commercial score follower is Smart Music, a practice aid for music students, by MakeMusic Inc. The program has accompaniments to standard repertoire for most solo instruments. A piece's accompaniment plays along with a soloist, whose tempo is tracked with a microphone. A less rigorous form of listening, score orientation, does make not continual tempo estimations but responds to selected highlights, such as a trigger from a pedal or a high note at a given pitch.

So much for listening. Now we can consider three forms of response. Transformative responses create variations on a performance. For example, Max can be configured to invert intervals, play a phrase backwards, transpose notes, arpeggiate chords, sense the current harmony and add a bass note, create chords from a melody, and more. Generative responses are based on material that the computer creates on its own, such as algorithmic creation of melodies from a library of pitches and rhythms (see “Game of Chance” in the November 2003 EM for more on algorithmic composition). Sequenced responses consist of stored musical passages that are kept on hand to be played when triggered. For example, in a score-oriented listening system, certain events in a score, such as a long, loud middle A, might trigger a preset melody. The performer might then create variations on the melody using a continuous-control pedal that changes the sequence's tempo or dynamics.

Finally, we can think of two roles the computer might play in a performance. In one, the computer extends the player's instrument, augmenting a solo performance with features such as filtering, effects, or pitch doubling. In the second, the computer creates another personality, so that it plays a kind of duet with a musician. Sophisticated implementations of duet partnering may rely on techniques of artificial intelligence to perform tasks such as defining phrase beginnings and endings or sensing changes of scale, mode, or key.

MESSAGE IN A CABLE

The previous examples described MIDI responses. MIDI is an effective vehicle for interaction, given its discrete, event-based format. Incoming events can be marked with time stamps, easily cataloged, and complemented by stored catalogs of algorithms or sequences. MIDI, however, provides an incomplete representation of a performance. Notably absent is any description of timbral variation. But an extension to Max called MSP adds the ISPW audio-processing modules to the environment, letting today's computer owners explore what was once only possible with the 4X, at less than one one-hundredth of the cost.

While an audio-based system has the advantage of being more closely tied acoustically to a performance, it lacks many of the flexibilities of a MIDI-based system. Responses such as playing a phrase in reverse or inverting all pitches around a given note are easy to implement with MIDI's unambiguous event types, but much more difficult to perform with a stream of audio samples. Polyphony is another issue that is easy for MIDI: a chord is easily recognizable as a set of discrete pitches. This level of analysis is impossible for an acoustic signal, as no one has been able to create a program that can distinguish between simultaneous pitches and overtones of a fundamental pitch. Acoustic systems, then, are typically based on input from a monophonic instrument.

Pitch trackers can identify the fundamental of a monophonic instrument or signal. With a pitch-tracking module, a signal's frequency can be sent to an oscillator to control its pitch, or the signal may be transposed. Other audio-based applications could include using the volume of an acoustic signal to modify the index of a frequency-modulating oscillator, or mapping MIDI controller values to audio processes such as reverb time, filter frequencies, or stereo placement. Analysis modules can do things like analyze incoming speech, separate noisy sibilants from periodic vowels, and process each differently.

OSC (Open Sound Control) is a protocol introduced by the Center for New Media and Audio Technologies (CNMAT) at the University of California at Berkeley in the late 1990s to enable real-time control of computer-synthesis processes from gestural devices. OSC does not include MIDI messages, but MIDI messages can easily be mapped into OSC, making OSC commands a superset of the MIDI protocol. OSC offers increased resolution and definition of gestures and synthesis parameters, as well as more accurate time control. It is transmitted over networks of computers, which means that it is well suited for broadcast performances of computers and performers interacting with each other from different places. The Gibson guitar company has also developed the MaGIC specification, which sends an electric guitar's acoustic signal over an Ethernet network, giving guitarists the opportunity to participate in these simulcast collaborations.

SUBTLE MANIPULATIONS

Joel Chadabe probably chose the theremin for his original Synclavier system because that instrument is practically unparalleled in its sensitivity to micromodulations. Ironically, as the sound capabilities of electronic instruments have evolved, their player interfaces have become increasingly rudimentary. Interactive performances often feature experimental-instrument types that push the sensing envelope. Instruments like Don Buchla's Lightning allow movements in space to be translated into MIDI control signals.
Massachusetts Institute of Technology Media Lab composer Tod Machover heads the development of hyperinstruments that generate various control signals. The conducting dataglove translates a conductor's left-hand movements into controls by tracking the angle of each finger relative to the back of the hand, as well as the angle of the joints of each finger. Hyperstrings augment the capabilities of string instruments. One commission by cellist Yo-Yo Ma consisted of sensors that tracked bow angle, bow pressure, wrist angle, and left-hand finger positions. Data from the cello motions and an analysis of the instrument's audio were fed into a computer that generated audio in response.

GET WITH THE PROGRAM

FIG. 1: Cycling '74's Max?MSP combines customized MIDI and audio processing. Plain black patch cords carry MIDI-based processes, while black-and-yellow striped patch cords carry audio processes. This example creates simple FM synthesis.

Max/MSP is the software most commonly used in interactive music applications (see Fig. 1). Its graphical front end facilitates algorithm configuration, while the essential issues of event scheduling and input tracking are kept “under the hood.” This allows users to focus on music rather than computer cycles. The Max environment has also spawned two offshoots. Pd (“pure data” or “public domain”) is a version introduced by Miller Puckette that exists in the public domain. It is free, runs on virtually all hardware platforms, and is under continual development by a community of users. Yet another version, jMax, is written in Java and is available from IRCAM's Web site.

Other systems suited to interactivity include Symbolic Sound's Kyma system, an audio processor and sound-programming language for Macintosh and Windows. Like Max, it is visually oriented, but processing and synthesis modules are arranged on a timeline. Kyma includes pitch and amplitude trackers, and it can be configured to wait for a specific event (such as a middle C) before, for example, running a script to generate notes (see Fig. 2).

FIG. 2: In Symbolic Sound's Kyma, processes are dragged onto a multitrack timeline. Processes such as waitForLowG (center-left, track 1) are used to tell the system when to start various tasks.

James McCartney's SuperCollider, a free program for the Macintosh, is a text-based programming environment. Although the absence of a graphical interface makes SuperCollider harder to learn than some programs, it also permits a greater degree of efficiency and flexibility. For example, the number of active oscillators can be assigned to a variable. Changing the number of oscillators in a patch is simply a matter of changing the value assigned to that variable, rather than adding or removing objects and patch cords from the screen.

Kyma's developer, Carla Scaletti, has pointed out that these programs are computer music languages. Most commercial music software falls into the category of a utility, meaning programs that perform common, well-defined functions. It's true that many utilities are quite complex — your average digital audio sequencer is an example. But they cannot match the open-endedness and flexibility of general purpose languages that enable users to configure whatever synthesis and audio-processing algorithms they want, nor can they provide the same ability to tailor these processes to customized input and output routings. You can take all the features of your favorite commercial synths and combine them in one custom environment, provided you have the computer memory (and the patience!) to cobble them together. For those wanting individualized performance environments, computer music languages are the only way to fly.

INTO THE FUTURE

Interactive music raises intriguing questions about musical intelligence, compositional methodology, and collaboration — questions that only become more intriguing as computing power advances. This is a pursuit likely to become an important current of 21st-century music.