Sonic Composition in "Tongues of Fire"Author(s): Trevor WishartSource: Computer Music Journal, Vol. 24, No. 2 (Summer, 2000), pp. 22-30Published by: The MIT PressStable URL: http://www.jstor.org/stable/3681924 .Accessed: 10/07/2011 09:22

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=mitpress. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Computer MusicJournal.

http://www.jstor.org

Trevor Wishart Department of Music University of York Heslington, York YO1 5DD, UK

Sonic Composition in Tongues of Fire

Editor's Note: The sound examples cited in this article can be found on the Computer Music Jour- nal CD volume 24 that accompanies the Winter 2000 issue.

In composing, I differentiate between the proce- dures I use to generate compositional materials and the structures I lay out for the listener. It's not important for the listener to know anything about my compositional procedures, but I hope they in- form the formal and dramatic structure that I do hope the listener will perceive and appreciate, though not necessarily consciously. Hence, much of this article is for the benefit of composers rather than potential listeners.

My composition Tongues of Fire (1994) is a mu- sical work created for the recorded medium using the computer. It relies on the computer's signal- processing power to metamorphose one kind of sound material into another, thereby making au- dible connections between different kinds of sounds and enabling a musical structure to be de- veloped in the sonic domain.

Because sonic space is multidimensional and the choice of possible starting sounds is unlimited, we need some way to navigate through this fascinat- ing space of possibilities. In the early- and mid- 20th century, composers became transfixed by permutational procedures, deriving ultimately from the serial method of Schonberg. However, these procedures relied on a finite set of discrete elements to permute. In contrast, sound-space is unbounded and continuous. A more detailed dis- cussion of these ideas can be found in my books On Sonic Art (Wishart 1985) and Audible Design (Wishart 1995).

We can proceed seamlessly from one point to an- other (a continuous transformation) or by discrete steps (a sequence of discrete metamorphoses, more akin to traditional motivic variation, perhaps). I use the term metamorphosis to refer to the sonic

manipulation of a sound to produce related sounds, while the term transformation refers to a process of sonic development through time; that is, the use of sonic relationships between events to build musical structures in time. This is also re- ferred to as sonic modulation by analogy with mo- tion between keys in the tonal system. We can also proceed in several dimensions at once, and in fact the discrete parameterization of sounds in terms of pitch, duration, etc., does not sit easily with our perception and recognition of sounds in the real world.

The method I therefore adopt is quite similar to the computer-graphics model of evolution pro- posed by Richard Dawkins in The Blind Watch- maker (1986): I begin with a starting sound, and apply any number of different small metamorpho- ses. A metamorphosis must lead to a perceptually similar sound, as discussed later. I then transform many of these sounds a little, and so on. In this way, I build a tree of interconnected sounds with its nodes and branches. As this process of sound generation proceeds, I select particular sounds to further metamorphose, or to use finally in the piece, on the basis of their intrinsic aesthetic qualities and their audible relationships to one an- other. However, some sounds may be used purely because they form a perceptual bridge between two other more notable sounds, like passing tones in a pitch-based musical organization.

This tree-generation approach is a generative method, not a compositional rationale. In many cases, sets of related sounds used in the piece may be derived from connected points along the tree, but this is by no means necessary. Ultimately, my perception of the relationships among sounds de- pends on my listening to them, and when I make such judgments, I have no interest in how, or even whether, the sounds are generatively related. I have criticized elsewhere the confusion of composi- tional methodology with audible design (see, for example, chapter 9 in Audible Design). Sounds originating from very different generative processes may have significant audible similarities, while

Computer Music Journal, 24:2, pp. 22-30, Summer 2000

? 2000 Massachusetts Institute of Technology.

22 Computer Music Journal

sounds from the input and output of some process may have no notable audible similarities at all.

In practice, however, the generative processes I use (and which I have developed) tend to produce sounds that are perceptually related to each other in some way. Otherwise, the generation of mate- rial would become arbitrary, and the search for connections hopelessly complex. Hence, the pro- cess of evolving materials is not strictly random, as in Dawkins's sense. I use the analogy to stress that it is also not deterministically rational. Part of the excitement of composing with sounds in this way is the often-unpredictable consequences of applying metamorphic processes to new types of sounds-consequences to which one must be con- stantly open.

This quasi-genetic process of variation and se- lection combines rational exploration with per- sonal aesthetic choice. This balances two important aspects of composing for me: the things that we know that we know (explicit perceptual connections and rational processes of metamor- phosis and transformation), and things that we don't know that we know (cultural and personal implicit preferences that may be far more interest- ing to future listeners than all our intellectual ra- tiocinations, because they carry information about cultural context and our relationship to it that we have simply taken for granted).

This process forms the basis for the next stage, in which the materials are arranged in temporal sequence, or overlaid contrapuntally (strictly speaking, as counterstreams) to emphasize, mask, or complicate their perceptual connection, form- ing an evolving structure in time. It is primarily this temporal unfolding of the materials that I hope the listener will perceive and appreciate.

In reality, there is not a simple dividing line be- tween the two stages. Often, sets of metamor- phosed material are worked into complete phrases at an early stage. Some continuous metamorpho- ses call out to be treated as musical phrases in their own right, as they extend over a considerable time (for example, the voices to water transforma- tion discussed below, lasting for around 90 sec) and provide a sense of musical motion within themselves. Also, any concatenation of materials,

at any stage of sequencing or layering, can itself become the subject of further metamorphosis.

The "theme" of Tongues of Fire is a rapid solo vocal utterance less than 2 sec in length. This theme was chosen both for expressive reasons (it is recognizably human, but slightly grotesque, slightly comical, and without any linguistic con- tent in any existing human language) and for sonic structural reasons (it is a sequence of several spec- trally complex and different sounds, thus making excellent raw material for many kinds of sonic metamorphoses). The theme of Tongues of Fire can be heard, repeated, in Sound Example 1.

The theme is immediately repeated. Structurally speaking, I tried to make it clear to the listener that this surprising and-on first hearing-appar- ently arbitrary event is significant for the piece. In fact, all of the sound material in Tongues of Fire develops from this seed. In this sense, Tongues of Fire does not differ from a traditionally conceived instrumental work that proceeds by the metamor- phosis and development of notated pitch motives. However, Tongues of Fire deals with sonic meta- morphosis and development. Using the computer, we can apply many processing procedures both di- rectly to the waveform of the sound or to the time- varying spectrum of the sound, and we can reassemble the sound in many different ways, at many perceptually different timescales (from phrase frame to rhythmic frame to grain frame to imperceptibly short sonority frame). Further dis- cussion of this procedure may be found in Audible Design (pp. 16-19).

In this sonic space, traditional categories and hi- erarchies of musical parameters break down. In particular, "timbre"-essentially an umbrella term for all those features of sounds not captured in tra- ditional notation, including stationary spectra, spectral changes, formant structures, formant changes, vibrato/tremolo and their evolution, graininess, jitter, etc.-provides a multidimen- sional arena in which pitchness is but one dimen- sion. We can navigate and explore the timbre of sounds in detail for the first time, now that com- puter analysis and manipulation provide us with the power to do so. (Pitchness and pitchal are terms used in Audible Design to designate the

Wishart 23

spectral quality of a sound that gives it a clear sense of possessing a single pitch, as opposed to be- ing noise-like, inharmonic, or multipitched. The term is used to distinguish it from the harmonic sense of pitch. Thus, a rapidly sliding sine tone, es- pecially one that crescendos from inaudibility and then diminuendos into inaudibility, may never have any specific pitch in the harmonic sense, yet it is perceived as having a pitch-like spectral qual- ity, to have pitchness, or to be pitchal.)

The signal-processing power of the computer means that the starting sound becomes almost in- finitely malleable. In these circumstances, we need to make a distinction between what is tech- nically a metamorphosis and what is perceptually a metamorphosis. As an extreme example, any starting sound can be converted to noise by multi- plying each successive sample by a different ran- dom number. Technically speaking, this could be regarded as a metamorphosis (the algorithm in- volved is quite straightforward), but perceptually, there is no relationship between the starting sound and the resultant sound (and in fact the same or very similar perceived resultant could arise, re- gardless of the starting sound). However, even in this case, it is possible to conceive of creating a se- quence of increasingly noisy versions of the origi- nal sound, spanning the timbral space between it, and noise (compare, for example, the metamorpho- sis of the boy's singing voice to a pure sine tone in Karlheinz Stockhausen's Gesang der Jiinglinge [1958]). Perhaps in this way we might create a set of perceptually acceptable transformational steps for the listener, linking starting sound to noise. Moreover, when presenting this material in a piece, we don't need to sequence the increasingly noisy sounds in that temporal order. As long as it is possible for the listener to make the connection at some stage of a repeated listening, we have a perceptible transformational structure, an audible connection between two quite distinct sounds: our starting sound and noise.

In practice, I am more perceptually demanding than this example suggests. Quite subtle changes in the nature of a sound (for example, changes to the attack characteristics) can radically alter our source attribution (what we imagine the source of

the sound to be). No matter how abstracted from reality our sounds might be, we will still tend to attribute material and dynamic causation to them that is based on our experience of sound in the real world. (See for example the chapter entitled "Is There a Natural Morphology of Sounds?" in On Sonic Art.) Hence, sonic metamorphosis can be a multifaceted and subtle art.

However, occasionally I allow myself to "run with" the result. For example, just beyond the pulsed-rhythmic climactic section of the piece (at ca. 20:20), there is an extreme metamorphosis. I usually refer to this as the "fireworks" metamor- phosis. I always attach names to sounds, giving some hint of their audible substance. This, for me, is a vital organizational procedure when dealing with hosts of sonic material. Simple numbering or technical naming makes it difficult to locate a sound again when one has a musical, rather than purely technical, goal in mind. One advantage of composition with computers is that one is forced to give a name to all soundfiles, whereas in the analog studio, one is faced with segments of tape, looking more or less identical (apart from their length), and elaborate cataloging procedures must be followed to keep track of sound materials.

The immediate starting sound (or sonic node) for this metamorphosis is the sound voismetal (dis- cussed in more detail later), which has a percussive vocal attack leading to an extended "metallic" inharmonic sound whose tessitura descends slowly. The metamorphoses use waveset averaging: the duration and shape of the starting sound's wave between each zero crossing is averaged over time. Applied to such a complex starting sound, waveset averaging produces many unpredictable and noisy artifacts. The only perceptibly retained feature of the starting sound is the descent in tessitura. Al- though it is on the edge of perceptual justifiability, I felt that this extreme metamorphosis fit well at this particularly dramatic moment in the piece, where the energy of the pulsed rhythmic material subsides. Here, the piece has reached its transfor- mational outer limits, and is about to tie up the loose ends in a traditional coda, using mainly reca- pitulated (rather than further metamorphosed) ma- terials from throughout the piece. The

24 Computer Music Journal

time-stretched voice and the fireworks metamor- phoses can be heard in Sound Example 2.

The larger-scale structure of Tongues of Fire does not differ markedly from that of a comparable instrumental work. At the largest time frames, we are no longer dealing with source recognition or sonic detail, but with overarching connections be- tween blocks of material. Hence, in some ways, there is no reason to abandon existing procedures, though of course new means of formal organiza- tion become possible, particularly those using source recognition to project overall sound land- scape or narrative.

Just as the initial theme is immediately repeated, the whole first extended phrase, in which that theme is developed in a number of ways, is followed at ca. 1:00 by an echoing phrase, the theme now de- veloping differently. Similarly, on an even larger scale, the first major section ends at ca. 10:00, when the progressive granulation of the voice slows to a regular clock-like tick, leading to a pause. After the pause, the theme returns but is then followed by four cycles of a rhythmic variant of that theme, which becomes crucial to the continuing evolution of the piece. It recurs at 11:50, 13:10, 16:50, and 17:00 (where it already takes on some of the pitched character of the later development). This leads ulti- mately to the material at 18:20, where it forms the rhythmic basis of an extended pulsed-rhythm sec- tion in which changing pitchfields become impor- tant, and which forms the climactic conclusion of this part of the piece. After the dying away of activ- ity (from 20:40 to ca. 22:20), short segments from all over the piece are juxtaposed amid silences at the start of the coda. At the end of the coda, the theme is recapitulated in a truncated version of the open- ing. The piece concludes with a fragment of the theme as a cadential event.

Three other theme variants play an important role in the piece. The first is a metamorphosis of the almost percussive vocal attack of the theme, in which the tail of the sound is greatly stretched in time. (The initial attack is not stretched, in or- der to preserve the sense of a vocal origin for the sound.) With extensive stretching, the noisy spec- trum, a spectrum that varies rapidly in a brief space of time, becomes slow moving and takes on

an inharmonic, semi-metallic quality. This tail glides down slowly in tessitura. This vocal-attack- to-metallic-extension sound is the starting point or sonic node for many sonic processes, and I refer to it as voismetal.

The second theme variant is a texture of gabbling voices, extending the solo voice theme material into a disgruntled-crowd-like event by su- perimposing several different variants through time. I refer to this as gablcrowd. This variant is also an important sonic node.

Last of all, the percussive vocal attack men- tioned before is stacked (several tape-speed-like oc- tave transpositions superimposed) with attacks synchronized, and a kind of reverberant extension added to the higher-and hence shorter-compo- nents. This brings out the pitch (D) of this event, which otherwise is difficult to focus upon in the sonic context of the theme. This variant appears very close to the start of the piece, and its pitchness sets it apart in the nonpitchal context. I refer to it as pichstak. It also becomes the seed of further developments.

The original theme, its rhythmic variant, voismetal, gablcrowd, and pichstak can all be heard in Sound Example 3. The development of the rhythmic version at the climax of the piece can be heard in Sound Example 4.

Let's now return to the opening of the piece. Sonic development begins immediately. The vocal theme itself starts with a percussive vocal attack, and ends with a "slurp" sound. These elements are immediately developed. After the two statements of that theme, a third statement begins with the percussive attack, but is truncated directly to the slurp, which is then repeated, semi-overlaid, at lower pitches and speeds, leading to the pichstak, the strongly pitched version of the percussive. One of the few steady-pitched sound events in the piece, pichstak becomes an important marker, an- nouncing the start of short or long phrases, some- times transposed, until it becomes the pedal point of the harmonic sequence of the pulsed-rhythmic climax previously mentioned.

At its first occurrence, pichstak introduces a de- celerating sequence of the original percussives. This deceleration prefigures the accelerating

Wishart 25

"bouncing" sounds to come. The deceleration is achieved with a nonstandard time-stretching tech- nique, giving stretched duration proportions of 1:2:3:4:5:6:7. The stretching process divides the original signal at every other zero crossing, and treats the signal segments thus produced as indi- vidual waveforms. Time stretching is achieved by the immediate N-times repetition of each wave- form before proceeding to the next. Clearly, with a simple waveform such as a pure, fixed-frequency sine tone, this procedure produces perfect time stretching, but applied to complex or rapidly changing signals, it generates strange artifacts. In particular, once a waveform has been repeated about four times, it will generate a unique pitch. (This is based on my own experience. It may be that more cycles are needed with very high fre- quencies.) It is therefore possible, for example, to start with a complex signal and proceed, step by step, to a 256-times repetition of each component "waveform," thereby generating a melodic se- quence of arbitrary timbres and arbitrary rhythms. In our case, as the number of waveform repetitions increases, the original signal gradually breaks down into an ultra-rapid stream of brief pitch cells. Toward the end of our sequence, therefore, this pitch bubbling becomes apparent.

Near the end of the sequence, the same percus- sive vocal sound reappears, grouped into an accel- erating and decrescendoing set, suggesting a bouncing object as its source. This rhythmic/loud- ness motive is repeated with sonic variations in which the waveforms are altered to a different shape for each repetition of the motive. Perceptu- ally, we seem to hear bouncing on or in different physical materials, for example, sand. The whole phrase concludes with a repetition of the resonant D, sustained, with a slight tremolo as it decays to nothing. The complete opening phrase, as de- scribed above, can be heard in Sound Example 5.

Hence, sonic development is intrinsic to the piece even within its opening phrase, and we could con- tinue to examine the whole piece in this amount of sonic detail. However, I intend to look at just a few example phrases to illustrate some of the multifari- ous sonic processes used. Note that a sonic meta- morphosis may proceed in the continuum, a sound

seamlessly becoming something different, like the voice-to-bees metamorphosis in VOX 5 (Wishart 1990), or as a sequence of discrete steps, like the metamorphosis from singing voice to pure sine tone previously referred to in Gesang der Jiinglinge.

In the continuous case, and also where discrete metamorphic steps lead gradually away from one type of sound to a recognizably different type, I have compared this sonic process to key modula- tion in the tonal system. Clearly, sonic space does not have the strictly symmetric and cyclical struc- ture of the tempered scale's cycle of fifths. We do not even have an analogy of the octave in sonic space. But it is possible to create a similar sense of movement from one part of musical space to a rec- ognizably different one, for example, from voice- like to wood-like, or to return to a sound sonically close to the original starting sound and to have a sense of "distance" from our starting sound. Thus, just as F-sharp major is much further from C major than is G major, so an inharmonic resonant bell sound is further from the sound of pouring sand than from an inharmonic nonresonant bell sound. As sonic space is multidimensional (Wessel 1979), it is difficult to impose any simple measure of sonic distance (a metric) on that space; whereas in tonal space, G major is the same distance from C major as it is from D major, measuring around the cycle of fifths. However, I am not concerned with draw- ing an exact parallel, but rather with merely dem- onstrating a certain analogy in the sonic domain with tonal motion, tonal return, and tonal distance.

We begin by examining the phrase that starts with the pichstak D attack at ca. 2:40 and lasts un- til ca. 4:00 (as in Sound Example 6). The node sound from which this phrase develops is voismetal, and is heard immediately after the pichstak attack. It is developed in various ways. In the first case, we produce a rather artificial tremolo in the tail of the voismetal sound by im- posing a rapid sequence of brief, shallow, loudness dips. The first of these is heard at 2:40 (where the phrase begins), and is the first of a sequence of dis- crete metamorphoses that gradually leads else- where: a sonic modulation. Thus, on each recurrence of the voismetal sound in the ensuing overlayed texture, these loudness dips cut deeper

26 Computer Music Journal

into the sound, and the tail begins to separate per- ceptually into a sequence of discrete struck-wood events. (Descriptive adjectives like "metallic," "struck wood," or "cicada noise" are meant to suggest the type of sound event being referred to, rather than to give an accurate description of their physicality.) Using a grain-detection program, we can isolate these brief events and duplicate and/or transpose them, and this approach is used to pro- duce a further variant that more strongly empha- sizes the perceptual distinctness of the struck-wood events. By now, we have sonically "modulated" from stretched voice to struck-wood- like (ca. 2:50). The metamorphosis from voismetal to wood can be heard in Sound Example 7.

From here starts a texture in which time-variable pitch shifting is used to create variants of the origi- nal voismetal sound, which pitch-slide down or up over larger ranges, or which oscillate wildly from the mean pitch. In the context of the piece, I hear these more as variations on the voismetal original than as true metamorphoses of that original, but there is no sharp dividing line between the two, and to some ex- tent it depends on where the time sequence of vari- ants, as laid out in the piece, leads. In a similar way, a pitch may be judged to be a chromatic passing note or a pivot of tonal modulation, depending on what eventually happens to the surrounding sequence of pitch events. This texture forms a bridge to a final strong statement of the struck-wood sequence, the separateness of the wood events being even more emphasized by the ritardando of the wood sounds (as in Sound Example 8).

Here the wood sounds accelerate and rise in pitch, leading into a texture of higher-pitched, shortened, repeated sounds, over a small pitch range, vaguely like cicada noise, an even more distant sonic modu- lation (3:07-3:17, in Sound Example 9).

We also hear here a dramatic extension of the voismetal sound, a stack of transposed versions of it. This is also prominent elsewhere, and it serves as the penultimate sound of the entire piece. This then becomes the second part of a repeating phrase unit (3:17 and 3:28), whose first part is an acceler- ating sequence of attacks. This accelerating mate- rial also emerges from the struck-wood events. The waveform of one of those events is modified

by waveset distortion. In this process, the signal is divided at every other zero crossing into wave- forms, and these waveforms are replaced by differ- ent waveforms but of the same duration and amplitude. We can then apply a process of in- betweening, a term I have borrowed from cin- ematographic animation, where principal picture frames are drawn by the chief artistic designer, then the many in-between frames that produce the sense of movement linking one key frame with an- other are produced, originally by assistants, but more recently using computerized methods. In the sonic case, beginning with the pure struck-wood event, we produce a sequence of sounds in which more and more of the new waveform (and hence less and less of the original "wood") is mixed. As the two waveforms being mixed are synchronous at their zero crossings, we are effectively generat- ing a gradual sequence of metamorphoses of the waveform itself, and hence of the sonority. Putting these mixes together in a progressive sequence, we produce a sense of sonic motion from struck wood to "drum-like." The sequence also accelerates, like the bouncing sound referred to earlier, but now the sound sequence is a crescendo (rather than a diminuendo), suggesting, perhaps, inten- tional force, rather than the energy dissipation of free bouncing (as in Sound Example 10).

The phrase terminates (from ca. 3:52) with yet another variant of voismetal. This applies the brassage technique, as used in the "harmonizer," to voismetal. Standard harmonizer brassage chops up the starting sound into segments about 50 msec in duration, sufficient to preserve the instanta- neous pitch of the event, but insufficient to give much of a clue to any changes in pitch that might be taking place. By splicing these segments back together again, in the same order but with a greater or lesser degree of segment overlap, we can reproduce the original event with an altered dura- tion, but with no shift in pitch. (In reality, we have to slightly randomize the durations of the cut segments to avoid creating a spurious pitch from any regular sequence of splices used.)

This process can be varied in many ways. In par- ticular, we can adopt the following routine. First, we cut up the original sound as usual. Then, for the

Wishart 27

first segment of the reconstructed sound, we choose the first segment of the original sound, but for the second segment of the reconstructed sound, we choose the second or the first segment of the origi- nal sound. For the third segment of the recon- structed sound, we choose the third or the second, or the first segment of the original sound, and so on.

As a result of this selection procedure, there is always a possibility that some of the early, percus- sive-attack segments of the original sound will ap- pear anywhere in the reconstructed sound. Hence, this attack quality gets distributed randomly over the time-stretched resultant sound, producing the "gargling" quality that we hear after 3:52 (as in Sound Example 11).

Thus, by putting all these variants of the voismetal sound node together in an appropriate temporal sequence, we have constructed a com- plete phrase based on the sonic development of the original sound node (as in Sound Example 12). This is an example of phrase construction mainly using discrete sonic metamorphoses. However, we can also generate entire phrases by the continuous or semi-continuous metamorphosis of the starting sound. In the two examples considered here, we are using gablcrowd as the sound node from which to begin.

The first example applies sound shredding to gablcrowd. The shredding process takes a given duration of sound material, and cuts it at a given number of randomly chosen positions; for ex- ample, seven cuts divide the sound into eight dis- junct segments, each of random length, though the sum of these lengths equals the original duration. These segments are then shuffled into an arbitrary order, and rejoined to produce a new sound of ex- actly the same length as the original. This process is then repeated many times. In general, at each shred, the new cuts are unlikely to coincide with any existing cuts, so any preserved continuous stretch of the original sound will tend to be cut again into even smaller segments.

At each shred and shuffle, the starting sound is increasingly fragmented and increasingly ran- domly shuffled. Of all our sound experiences, vo- cal sounds retain their recognizability, even under extreme deformations. Thus, even after a hundred

superimposed shreds, the shredded gablcrowd sound retains some vocal clues. Eventually, after thousands of shreds, the dominant events in the resulting sound will be the splices themselves, and any originally complex starting sound will be re- duced to low-level noise. However, there is a point along this path of successive shreddings where the vocal quality of the starting sound is lost, but not its sonic diversity. At this stage, the sound be- comes akin to the sound of water falling gently around stones in a mountain stream.

Our phrase (13:20-15:10) is constructed by splic- ing together successive, cumulative shreds of gablcrowd so that the voices gradually dissolve into the water-like texture. This basic phrase structure is counterstreamed by other events: deep attacks initiating rising noise-bands derived from the shredded voices. It also continues, seamlessly, into a second continuous metamorphic process, evolving a pitchal upward portamento out of the noise texture to lead us into the pitch attacks of the next phrase.

The technique used here is end-synchronized de- lay. First, several copies of a sound are made at slightly different speeds, analogous to tape-speed variation. For example, we might decimate a wave- form of 48,000 samples to 47,600 samples, thereby changing both duration and pitch very slightly. These copies are then superimposed so that they synchronize at their beginnings. The resulting pre- cise beat frequencies caused by the precise delay times between the copies will be heard as a pitch portamento descending from "infinitely" high at the outset, where the copies are all synchronous, through the audible-pitch range as the copies get out of phase and the increasing delay gap generates a falling pitch, into the sub-audio range, resolving into a sequence of decelerating echoes. If instead we synchronize the ends of the copies, we will be- gin by moving into accelerating echoes, leading into a rising-pitch portamento. This is what we hear at our phrase end. This effect occurs no mat- ter what the nature of the original sound: it is a process-determined (rather than a source-deter- mined) effect, so we can use it to move from nonpitchal sounds into pitchal sounds. This entire phrase can be heard in Sound Example 13.

28 Computer Music Journal

As a second example of phrase structure based on a continuous metamorphosis, we again start from gablcrowd at 17:07. The metamorphosis begins by making the crowd texture increasingly dense, by superimposing randomly delayed, inexact copies of the starting sound. The copies are not exact copies, to avoid artificial-sounding delay or pitch effects between copies. With complex material like this at sufficient density, the spectrum becomes satu- rated, and we hear just a band of noise. At the same time, this band is made to rise in tessitura. As the process evolves further, the original gabbling voices re-emerge occasionally, retaining the con- nection with the starting sound. The noise is next put through a filter bank of increasing Q prior to the application of portamento, so that it gradually becomes a set of parallel portamentoing pitches. The continuing portamentoing means that the "chord" is never able to settle in any particular harmonic space (therefore it is not a chord in any traditional sense, technically speaking). The phrase continues by reversing these processes, returning first to noiseband and then back to gablcrowd. A counterstream of portamentoing parallel pitches splits away during the course of the phrase and de- scends slowly in pitch, in contrast to the up-and- down wave-like motion of the principle stream. This phrase is thus based on a classical arch form using continuous sonic metamorphoses (as in Sound Example 14).

An arch form might be regarded as a sequence of events or sounds followed by its retrograde. The idea of a retrograde, however, is not as straightfor- ward as it seems. In traditional notated music, a retrograde of a sequence of musical events means that we play the original events in reverse order. We are retrograding only the starting times of the events. We do not reverse the time flow inside the events themselves.

We could extend this notion of retrograde to dif- ferent time scales; for example, a dance piece of overall form ABBA might be thought of as having a retrograde structure. However, if we reverse the flow of time itself, by running a tape backwards, or reversing the order of a sequence of samples, the result is often surprising to us, as we reverse the events themselves and hence their flow of causal-

ity. As the evolving loudness and timbral envelope of a sound is often a vital recognition cue, revers- ing it can completely alter the perceived quality- and hence the recognition characteristics-of that sound. In general, however, continuous sonic metamorphoses can be retrograded in this way without too many surprises. But if our phrase structure is built out of discrete metamorphoses, we must use the event-order reversal form of retro- grade to achieve a comprehensible sense of retro- gression. Or we may need to use some combination of the two.

In the next example (7:40-8:30), we begin with gablcrowd, raising the tessitura and then applying spectral tracing to the sound. Spectral tracing pre- serves the most prominent partials in the sound, moment to moment (window by window). With simple, unchanging sounds, this process produces first an elementary noise reduction and then a gradual simplification or impoverishment of the spectrum. With complex sounds, however, the set of N most prominent partials changes on a mo- ment-to-moment basis; some partials leave the set while some enter, resulting in the revelation of complex weaving "melodies." (An example of spectral tracing can be heard in Sound Example 15.) In this example, the originally vocal texture gradually becomes pitchal in the timbral sense, rather than noise focused. After somewhat mecha- nistic echoes, we reach a metallic-clang event gen- erated by stacking transposed copies of exponential decay-enveloped versions of the spec- trally traced sound (as in Sound Example 16).

When this material is later recapitulated in ret- rograde form (10:50-11:40), we hear the continu- ous metamorphosis evolve in truly reversed time, from weaving melodies to transposed gablcrowd to original gablcrowd. However, the metallic-clang event is presented again in its original, time-for- ward form, and it subsequently develops through transposed repetitions of this original form. Fur- thermore, as we already know this material (we have heard it previously in the piece), I have placed a counterstream against it. Here, "breath- ing" sounds shorten and accelerate, becoming click-like and evolving into a regular pulse, like the "clock" first heard prior to the pause at 10:00,

Wishart 29

and leading into a recapitulation of the pulsed- rhythmic version of the theme at 11:45. Further- more, the clock and metallic-clang "bells" are briefly associated in a tangential reference to some external reality-but this brings us into a different realm of analysis. This process may be heard in Sound Example 17.

We could continue examining Tongues of Fire phrase by phrase, but I hope I have given sufficient insight into the musical processes at work. I have concentrated on the use of sonic metamorphosis in phrase building, and have said little about crite- ria for counterstreaming or for the more general sequencing of events. Nor have I discussed issues of sonic landscape, which in Tongues of Fire are dealt with in a more tangential manner than in my other compositions Red Bird (1977) or Fabulous Paris (1999). These are of equivalent importance in putting together the piece, but they are probably easier to relate to traditional musical form-build- ing techniques; therefore I'll leave these matters to other analytically inclined listeners to tease out. But in conclusion, I must add that Tongues of Fire

stands or falls by its impact in the real-time aural experience of its listeners.

References

Dawkins, R. 1986. "The Blind Watchmaker." New York: Norton.

Stockhausen, K. 1958. Gesang der Jiinglinge. Cologne: Stockhausen-Verlag.

Wessel, D. 1979. "Timbre Space as a Musical Control Structure." Computer Music Journal 3(2):45-52.

Wishart, T. 1977. Red Bird. Albany, NY: Electronic Mu- sic Foundation.

Wishart, T. 1985. On Sonic Art, new ed. London: Harwood Academic.

Wishart, T. 1990. VOX 5. In The Vox Cycle. York, UK: Orpheus the Pantomime.

Wishart, T. 1994. Tongues of Fire. York, UK: Orpheus the Pantomime.

Wishart, T. 1995. Audible Design. York, UK: Orpheus the Pantomime.

Wishart, T. 1999. Fabulous Paris. In Or Some Computer Music. London: Touch.

30 Computer Music Journal