Guide tothe Basic Concepts and Techniques 0 95 The same spectrum frequency shifted by 100 Hz : 105.4 230.9 296.4 361.5 427.0 492.2 557.6 623.0 688.8 753.8 2 ——: The same spectrum frequency shifted by -10 Hz : 55.4 120.8 186.3 251.5 316.9 382.6 447.7 513.1 578.6 644.2 modulations (general) ‘The spectra discussed up to this point are based on what physicists refer to as simple oscillations. By this they mean that there is only one periodic element to the wave-form, though that element may be complex. However, in many situations, one sound interacts with a second inde- pendent sound. The most familiar form of interaction is for one sound to modulate the other. Three types of modulation which have frequently been used by spectral composers will be detailed below: amplitude, fre- quency and ring. amplitude modulation This type of modulation, which is used for AM radio, is most familiar in music as amplitude vibrato, such as one finds in flute playing. Amplitude modulation is also important for its role in creating spectral flux (described above) in instrumental timbres. Most partials in nature have amplitudes which are constantly varying, even when the general impression is of a constant level. This aspect is modeled in many synthe- sis applications with jitters (random envelope generators) or low fre- quency oscillators (less then 20 Hz) modulating the main amplitude generator. Auditory rate amplitude modulation (modulations faster than 20 Hz), which generates new audible partials which are distinct from those belonging to the two interacting sounds, have not really been used much either in electronic or spectral music.96 Appenitix 1 frequency modulation Frequency modulation (FM) is the most prominently used modulation for musical applications. In its simplest form, frequency modulation can take the form of pitch vibrato, like that used by string instruments. In the seventies, John Chowning developed the technique of auditory rate FM, which modulates a sine wave carrier by a sinusoidal modulator with a frequency faster than 20 Hz. This type of modulation creates side bands (partials on either ‘side’ — symetrically above and below — of the carrier frequency) in the generated spectrum. These side bands have the advan- tage of exhibiting a great deal of spectral flux as the modulation depth, or index, changes. (Modulation depth and index are ways of expressing the amount of modulation that effects the carrier; the greater the modula- tion depth, or index the more side bands that are present in the modu- lated spectrum. The relative amplitudes of the side bands are also dependent upon this depth or index.) While these changes are quite dif- ferent from those found in instrumental spectra, Chowning demon- strated that in many cases the existence of fluctuations is more important than the exact structure of those changes. Therefore, this technique could generate relatively satisfying sounds with only two oscillators, whereas additive techniques, for example, might require dozens or more. This efficiency led Yamaha to license the technique for their synthesizers, starting with the DX series, In the same way that spectral composers were modeling and analyzing instrumental sounds for the creation of orchestrally synthesized timbres, they also looked to the FM technique. The spectrum produced with this technique is expressed with the follow- ing equation: frequency = carrier + and = (index * modulator); where index equals 0, then 1, then 2, ete. until the maximum index value has been reached, The amplitudes of frequency modulated spectra follow rel- atively complex functions, which are often left out of simple computa- tional models. Spectral composers have used this calculational model to integrate electronic FM sounds with instrumental timbres and to create a new category of spectral models for use in all types of pieces, The following example shows a modulation with a carrier (‘c’) of A, 440 Hz, and a modulator (‘m’) of 100 Hz — slightly above a G2 (notes approximated to the quarter-tone, frequencies marked above): 440 S40 340 640 240 740 140 840 40 940 -60 1040 -160 © cime-me+Qtm) c+Gtm) c+(im) —c+(Sm) c+ (64m) ¢-@*m) — e-Gtm) —e=(4*m)——e-(Stm) ee -(6¢m)Guide tothe Basie Concepts and Teceniques of Spectral Music 97 Note that negative frequencies are heard identically to positive ones, except that the phase is inverted. ring modulation Originally an analog electro-acoustic treatment, ring modulators modify complex sounds. In the original implementations, a sound captured by a microphone was modulated by a sine wave generator (this is used, for example, in Stockhausen’s works Mixtur and Mantra). The major difference from FM modulation is that this type of modulation is not hierarchic: there is not a carrier and a modulator which modifies it, but two equal sounds both of which are directly present in the resultant sound and both of which are modulated by the other. The spectrum resulting from a ring, modulation can be simulated when the frequency of each note of the first spectrum is combined through both addition and subtraction with the frequency of each note of the second spectrum, pro- ducing all the possible additive and subtractive combinations of the par- tials. When the two spectra contain many partials, enormous numbers of combination tones are produced and if the spectra are non-harmonic and rich, the resulting modulation can quickly turn into noise. The number of partials generated will be two times the number of partials in the first spectrum multiplied by the number of partials in the second spectrum, ‘The following example shows a modulation of a first spectrum built on A, 440 Hz, with the first two partials and a second spectrum built on D 3/4 sharp 2, 80 Hz, with the first three partials, (notes approximated to the quarter-tone, frequencies marked above): spectrum 1 spectrum 2 440 880 80 160 240 “ a b 2b 3b = modulation 520 600 680 360 280 200 960 1040 1120 800 720 640 2atb 2at2b 2a+3b 2a-b 2a-2b 2a-3b ath afb at3b ab add asb fo-98 Appenutix 1 virtual fundamentals In the above section on the overtone series, the fundamental was explained essentially as the greatest common denominator of a harmonic spectrum. For a distorted, shifted, non-harmonic, or modulation-based spectrum, however, the ear still tends to find a fundamental. (Note: the ear does this even in some instrumental spectra, such as low piano sounds, where the perceived fundamental pitch is absent from the spec- trum — the strings being too short to actually vibrate at those frequen- cies.) Many psychoacoustic algorithms have been proposed which attempt to model this effect by which the ear creates a sort of ‘virtual’ fundamental in spectra lacking a real one. These algorithms depend on the tolerance of the ear discussed in the above section on approximation. With slight variations, they calculate the greatest common denominator to a given tolerance (which is often user specified). The virtual funda- mental has often been used by spectral composers as an ad-hoc measure of harmonicity (lack of tension) or inharmonicity (presence of tension), equating higher virtual fundamentals with greater harmonicity or less inharmonicity and lower ones with less harmonicity, or greater inhar- monicity. The motivation for this is that harmonic spectra start with a real fundamental and as they are distorted the virtual fundamental moves in various directions; when these distortions become more and more noise-like, the virtual fundamental descends until, for white noise, the virtual fundamental approaches zero Hz. This technique, however, only works within constrained contexts, especially as regards register. It is often necessary to transpose aggregates to a common register prior to comparing their virtual fundamentals, because a lower register will auto- matically result in a lower virtual fundamental for the same harmony. spectra as harmony/timbres Building global orchestral sounds upon the same models that constitute single instrumental or artificial sounds — spectra — (through instrumen- tal synthesis, described above) creates a powerful ambiguity between the notions of harmony and timbre. Once the instrumental mass is perceived globally, as a color or texture, the notion of harmony becomes less rele- vant than that of color or timbre. Where the perception is less that of a fused mass, however, it is clear that the notes of these spectral models have taken on a harmonic role. What truly emerges, in fact, is that in spectral music the line between these two concepts has blurred, practi- cally to the point of non-existence. The aggregates in this music are used simultaneously to control the harmonic movement and the timbral evo- lution. Further, these two types of motion are often indistinguishable. Therefore, in spectral music, at least, it is often more relevant to combineGuide to the Basic Concepts and Teclenigues of Speetrat Music 99 the two concepts into the more general concept of a harmony/timbre; this hybrid concept preserves aspects of both of its component ideas and captures the interdependence and indivisibility that has developed between them spectra as reservoirs Besides building harmony /timbres from the acoustically-based models provided by spectra, many spectral composers use these models as reservoirs. They sometimes treat these reservoirs as modes, from which lines and harmonies can be constructed: the power of this system comes from the fact that acoustic models can generate very large numbers of frequencies (and through approximation, pitches) which can be com- bined with each other while still guaranteeing an overall coherence. This allows a single underlying structural entity and color to create a proliferation of surface manifestations which are coherently related Other times these reservoirs are used to provide a metaphor for the musical evolution they are trying to create; for example, moving to higher and higher partials of the same spectrum as musical develop- ments and other parameters augment the ambient tension. Treating spectra as reservoirs and their treatment as harmony /timbres, described in the preceding section, are not contradictory, but complementary. The reservoir approach is often used to provide surface activity within a slower harmonic rhythm and the harmony /timbre approach lends itself to more harmonic passages; however, many different configurations can be found in the music. Derivation of Pitch Aggregates from Spectral Analysis spectral analysis — Fourier Transforms Much of the information already discussed results from information gleaned from spectral analyses, but the actual mechanisms of these analyses has not been explained. They are based on the work of French mathematician Jean Baptiste Joseph Fourier (1768-1830). Fourier showed that any periodic waveform could be decomposed into the sum of a series of sine-waves whose frequencies are at integer multiples of a fun- damental (though not necessarily a finite series) with different ampli- tudes and phases; in other words, all periodic waveforms can be transformed into some type of harmonic series. This is called a Fourier Transform, since the periodic function is transformed into an equivalent Fourier series. While, in theory, the periodic function must be infinite, in practice, several periods of stability are enough for an accurate, though100 Appendix 1 not perfect (in the sense of being able to reconstruct an exactly identical waveform) analysis. Also while the technique in its pure form can create only harmonic spectra, the use of extremely low ‘pseudo-fundamentals’ allows a good sampling of the spectral energy throughout the auditory range — providing a close approximation of even very non-harmonic sounds. This technique is not well suited, however, for the type of dis- crete numerical analysis required for musical applications, where infinities cannot be part of the input or results and where a certain amount of quantification error is acceptable. The solution for most applied uses of spectral analysis was the development of the discrete Fourier Transform (DFT). This technique essentially samples discrete time positions of the input signal or function and truncates the Fourier series after a certain number of terms; when this sampling is sufficiently dense, a good approximation of the continuous function is created. This technique was extremely calculation-intensive even for computers, until a class of extremely efficient algorithms for calculating these DFTs was developed. These algorithms depend on a factorable number of points which allows the calculations to be broken into separate parts and reordered in ways that dramatically reduce the number of calculations. ‘They are referred to as Fast Fourier Transforms (FFTs). Fast Fourier Transform (FFT) This efficient version of the discrete Fourier Transform is at the heart of all spectral analyses on computers. To perform this calculation on an audio signal, a window of sound must be selected for analysis. This window cannot simply be cut out from the sound without creating major artifacts (frequencies which are created by the mechanism of the analysis rather than as a result of their presence in the sound), but must be extracted with an envelope that creates minimal distortion. The longer the window (temporally speaking), the greater the frequency res- olution of the analysis; temporal resolution, conversely, decreases with window length. This is due to the fact that all the sounds within the window are assumed to be unchanging and averaged throughout the window. This creates a situation which is similar to photos (fast shutter times are needed to capture moving objects, slow ones for low light) where compromises must always be made. When studying spectral analyses it is important to remember the significant effect that these parametric decisions have on the final result. In essence, there is not one analysis which represents the reality of an acoustic signal, but many possible analyses which accurately render certain aspects of the sound while distorting others (this is reminiscent of the effect created by different two-dimensional projections of the map of the globe). WhenGuide to the Basie Concepts and Techniques of Spectral Music 101 selecting these parameters, it is very important to bear this in mind and perform the analyses in a manner well-suited to revealing the aspects that one needs to see. It may even be useful to perform multiple analy- ses of the same sound dynamic FFTS ‘The above description of FFTs deals with single windows, in which all sounds are averaged together. In order to see the changes within a sound over time, a series of FFTs with windows which advance in time is necessary. Devices like phase vocoders are able to analyze a sound with this technique of FFTs with overlapping windows which advance in time and create a representation of the sound as it evol Under optimal conditions, this representation is sufficiently accurate that it can perfectly recreate the sound. While this technique of dynamic FFTs is capable of analyzing almost any sound, a difficulty often resides in assimilating the masses of data that are generated (a set of FFTS pro- duces thousands of amplitudes and frequencies hundreds of times for each second of sound). One of the most familiar and successful means of making this data comprehensible is to generate graphic representa- tions, like sonograms. sonogram Sonograms are a graphical representation of the three dimensions of time, frequency and amplitude in two dimensions, with the third dimension of amplitude represented by a clever graphical use of dark- ness. The advantage of doing this is that the multitude of frequencies with little or no amplitude simply disappear, making the image less cluttered. The program AudioSculpt, created at IRCAM, has provided easy access to sonograms and has had a great influence in making the subtlety of temporally sensitive analysis accessible and manageable. In the past, simplified non-dynamic models, of the sort described in the first section of this paper, were all that composers could easily manage. This sort of representation makes the use of much more sophisticated models possible Sonogram of a note played on the harp (see figure 1): the smaller top rectangle shows amplitude on the y-axis and time on the x-axis, the larger lower rectangle shows time (in seconds) on the x-axis and frequency (in Hz) on the y-axis; in this representation, the amplitude is shown by the darkness (in gray-scale). This type of representation is powerful since it gives a clear and under- standable overview of an enormous amount of data.102 Appendix 1 Untitled Anslusis 2) O02 04 06 08 1012 14 16 1 2.0.2.2 24 26 28 Figure 1 This sonogram was produced with the program AudioSculpt and is used with the permission of IRCAM data reduction Composers and scientists have not simply relied on graphical tricks to reduce the amount of data generated by dynamic FFTs; they needed methods for sorting this data and extracting the precise elements needed for different applications. It would be beyond the scope of this article to describe all of these; however two of them that have often been used by spectral composers need to be mentioned. psychoacoustic algorithms The most widely used strategy for reducing the amount of data pro- duced by FFTs is to attempt to select the partials of the analyzed sound which are the most important to the perception (or salient). The earliest and most simple strategy, and one used in many spectral pieces, is toGuide 1 the Basic Cone migues of Spectral M select the loudest partials (called the peaks). In the eighties, a desire not just to ground the selection in the quantitative parameter of amplitude but also to take into account the psychological side of salience led to new approaches. The most widely used strategy for determining the salience of various peaks (at least in the spectral community) is an algorithm developed by German psychoacoustician Ernst Terhardt, which was implemented at IRCAM by Dan Timis and Gérard Assayag as iana. (It was later ported into real-time by Todor Todorof.) The details of the algor- ithm are beyond the scope of this article, but the goal is to refine the selection of peaks, taking into account both physical principles (critical band, frequency response of the eat, etc.) and cognitive ones (construc- tion of virtual fundamentals); ranking the salience of each peak and elim- inating peaks that may be loud, but that are masked by other partials. This technique has allowed composers to better deal with technical or orchestrational limitations on polyphony, selecting the best partials — even when they are few. partial tracking A more recent technique for data reduction, which is particularly well suited to large series of dynamic FFTs is partial tracking. This technique looks for connections between successive analyses, trying, in essence, to connect the dots. This generates musical lines from the series of analys Itis generally used for resynthesis applications, but has also been used as a model for instrumental realizations. Rhythmic Concepts With the rhythmic and formal concepts described in this and the follow- ing section there are many fewer ideas described than in the previous sections. On a formal and rhythmical level spectral music is more contin- uous with other musical trends of the late twentieth century; whereas, it is more divergent on the level of pitch, harmony and timbre. I will limit myself to ideas which are both important to spectral music and particu- lar to it. Ideas which are part of the lingua franca of contemporary com- position will not receive specific treatment here. absolute duration vs. symbolic rhythm Just as frequencies offer spectral musicians a more direct access to many sonic structures than notes do, absolute temporal durations are often an easier way to conceptualize time and rhythm than the symbolic subdivi- sions of musical notation. This continuous conception has been less104 Appendix ( widely exploited for rhythm than the equivalent one has been for fre- quencies, since the problem of approximation is greater and the accuracy expected from performers and perceived by listeners is much less. Therefore the domain in which durational rhythmic thinking has been widely applied is limited to macro-rhythmic relations along with a few special case relations, in which durations have great advantages. In these situations durationally conceived relations are often more flexible than symbolic ones. An identical temporal structure can easily be stretched or compressed and can have the number of events increased or reduced without changing the framework of its overall perception, whereas this is, often difficult or impossible in a traditionally notated passage without completely re-notating it or changing the tempos (which in certain con- texts may not be possible or desirable). quantification The approximation of continuous temporal events into the discrete units of musical rhythmic notation or any such units is usually referred to as quantification. This name is an explicit reference to the sort of grid that is presented by rhythmic subdivisions. Unlike the grids dividing frequen- cies into the nearest available pitch, rhythmic grids are hierarchical; this makes convincing quantifications difficult to produce. Also, the ear’s ability to discern rhythmic anomalies is very context dependent. For example, during an accelerando, if one of the durations is longer than its predecessor (a micro rallentando) by even a few hundredths of a second, it will be perceptible; whereas a note that lasts eight seconds and one that lasts nine seconds will be indistinguishable when there are no exter- nal beat cues to mark the length. The result is that, unlike frequency approximation, there is not one closest and best approximation for each resolution (semi-tone or quarter-tone, etc. for frequencies, maximum beat subdivisions for rhythms), but different sets of compromises that must be made in consideration of the musical context and constraints. This often leads to the impossibility of using automatic quantifications that have not been manually modified by the composer. As a result, many of the passages in spectral pieces whose rhythmic conception began in the domain of durations were either quantified by hand or were ‘touched up’ manually. A system was developed recently at IRCAM by Carlos Agon, Gérard Assayag and myself which secks to redress this situation by providing an interactive quantification environment called ‘Kant,’ where compositional concerns and calculational strategies can coexist in a user-responsive environment. This system allows the composer to work towards an acceptable quantification rather than simply offering a ingle, perhaps unusable, one.Guide to the Basie Concepts ana Techniques of Spectral Music 105 accelerations/decelerations Among the structures most often modeled as durations in spectral music are accelerations and rallentandi, In order to produce a psychologically convincing impression of speeding up or slowing down, the duration changes must be exponential, not linear. This makes them particularly well suited to approximation by curves. These curves present intuitive graphical representations of the speed changes, that can then be adapted and quantified to fit diverse musical situations. While convincing accelerandi and rallentandi can certainly be written directly in rhythmic notation, they are extremely difficult to alter in this format (for example by adding events within the speed change); whereas curves are extremely malleable and allow the composer to generate entire classes of solutions which create the same sense of speed change. The following example is an acceleration curve, which is generated by an exponential function, then edited and displayed as the following curve: The curve is sampled at discrete points which are then proportionally represented within the Kant quantification editor:106 Appendix? The results are finally quantified into standard rhythmic notation: 2 7 1 Bi d =90 3 4 ‘ An — 1 d gd I Yd |) ATT | ecially noticeable in a very simple quantification, like this, is the loss of information. While we have no rhythms longer than those which precede them, the continuous nature of the acceleration has been lost. In a given context this might or might not be a problem. The only solution short of a tremendously complex rhythm, which is unlikely to receive an accurate rendering, is to adopt some element of proportional notation (like accelerando beams). It is precisely this type of problem which has led spectral composers to mix proportional and traditional rhythmic notation in their works. The attitude used in most spectral works is purely pragmatic: that which can be most clearly expressed through tra~ ional notation uses it, while that which can only be awkwardly or inexactly expressed through this notation turns to the alternate means of proportional or otherwise personalized notation models from electro-acoustic sources There are many other types of durational models that have been used by various composers. Other than curves, the most important class of models are the ones inspired by electro-acoustic procedures like echoes, delays, tape loops, etc. This type of manipulation is simply the applica- tion of widely used techniques from tape music into the realm of rhyth- mically notated instrumental music (via quantification). models from sonic analyses As was the case with harmonic structures, the initial use of relatively simple mathematical models, such as those described above, has been enriched in recent years with more complex models extracted from actual sounds through dynamic frequency analyses. Many different kinds of rhythmic information can be extracted from these analyses (whether it be dynamic contours, spoken rhythms or the pacing of timbral evolutions, etc.). Composers have extracted this information from all sorts of sounds — from crashing waves to recited texts to instru- mental gestures. These models do not just offer rich harmonic materials, but also propose very interesting rhythmic structures, which can be usedGuide to the Basic Concepts and Techniques of Spectral Music 107 along with the corresponding harmonic material or independently from it. Of course, the translation of the often minute rhythmic fluctuations found in nature into an instrumental context requires a very prudent control of the quantification and a judicious mixing of traditional and proportional notation. rhythmic distortions The high degree of perceptual clarity, and the accompanying predictability, that many durational rhythmic structures offer has led many composers to distort these structures by varying degrees. These distortions echo those which are performed on harmonies. The simplest types of distortion pre- serve the relative lengths of the rhythms; a good example of this is rhyth- mic stretching or compression, whereby the relative lengths of each duration is preserved while the total duration is augmented or diminished (this is identical to one of the procedures for harmonic distortion men- tioned above). Another technique is to add a percentage of random rhyth- mic fluctuation to the durations. When this percentage is small, the underlying structure is still very present, but the surface is made less pre- dictable; when the percentage is increased, the randomness can begin to overpower the underlying structure. Combinatorial permutations are also frequently used to disturb the linearity of a rhythmic model (for example, swapping, the positions of two events within a sequence of ten events can provide a moment of surprise and contrast within a strongly directional sequence). These distortions greatly increase the flexibility that can be achieved with a relatively limited number of rhythmic models. Formal Concepts process While not unique to spectral music, the idea of continuous transforma- tion from one state to another, or process, has taken on a special manif tation and played a crucial role in the formal construction of spectral music. The types of processes found in spectral music are significantly different from those of minimalist music, for example, in that they affect all the musical parameters together, rather than acting on only one or two (like phasing). Typical processes of early pieces from the movement were transformations from order and stability (which includes harmonic- ity) towards the disorder and instability of noise, or vice versa. A good example of this kind of multi-parametric process is the beginning of Murail’s piece Gondwana, where an orchestrally synthesized bell —Appenli | with the spectral profile, envelope and instrumental attacks necessary to achieve that structure — is gradually transformed into an orchestrally synthesized brass sound. The kinds of processes used in this music are istinct from many formal processes found in other types of music, in that they function on a perceptible levels. They are not underlying math- ematical structures, but permeate all levels of the piece and are an impor- tant aspect of the perceived musical movement and evolution interpolation The kind of smooth transformation from one state to another described above often makes use of interpolations. These are used in almost all aspects of the music, especially pitches and rhythms. The initial and final states are placed at the two end-point of lines or curves which are then sampled at various points to generate intermediate states. This type of procedure is especially effective in the continuous domains of frequency and time, but is sometimes performed directly on symbolic notes or rhythms. The following figure illustrates the construction of two steps interpolated along lines between the initial and terminal states (these lines could also be replaced by curves or other forms to produce a less direc- tional result). It should be noted that the power of this technique is the novelty that can occur in the interpolated states. Interval or time struc- tures that would have been impossible to abstractly imagine can some- times grow out of these interpolations. In effect, the true interest is not the beginning or the end, but the discoveries that are made on the way. 2 & z 3 5 3 z initial Ist 2nd terminal chord or interpolated interpolated chord or rhythmic chord or chord or rhythmic figure rhythmic rhythmic figure figure figureGuide tote Basie Concepts and Tecniques of Spectral Music 108 limiting directionality — processes of processes The main strength of processes and interpolations is the sense of direc- tion and even inevitability that they impart to the musical evolution. This asset can, however, become a drawback, when it leads to predictability. One of the most significant changes that has occurred in spectral music in the last 15 years is the desire to find strategies which will reduce this predictability while still preserving the previously acquired directional- ity, One technique that has been used to this end is anamorphosis, which is based on the anamorphosis technique of medieval painters. The idea behind this technique is to present a single object from different perspec- tives, which distort the object in various ways — sometimes even making it appear to be a different object altogether. In this way, one object can develop into a rich reservoir of musical and formal material that can sound very different in spite of its high degree of relatedness; creating very different and surprising effects without compromising the coherence of the musical material. Another technique is to skip steps within a process. In a clearly directed process, the composer creates unpredictability, in the rate of change, and contrast, between the non- adjacent steps, through this procedure. As with anamorphosis, the overall direction of the process is strong enough to support these local contradictions without losing its power to provide global direction. ‘The most interesting attempt to subvert the single-minded directional- ity of the process is also the most structurally significant: building processes which use as their elements not harmonies or durations but other complete processes. For example, one might begin a piece with a process which has dense noise-like harmonies in low registers, rising and becoming more harmonic and more definite in their melodic and figura- tive content; this entire process might then be treated as a unit which is gradually bent until it becomes an opposite process of sounds sinking, becoming less harmonic and more diffuse gesturally. The complex inter- mediate states in this type of multi-layer process can often produce novel musical textures and situations. Once these processes of processes are combined with the other strategies of anamorphosis and incomplete pre- sentation, they offer spectral composers powerful tools for building complex, unpredictable forms which, nonetheless, maintain cohesion and directionality. Associated Topics The topics in this section are not precisely about spectral techniques or ideas but are associated issues which have been important to many spec-110 Appendix tral composers. Due to the tangential relationships of some of these issues to our main subject, I will try to be very brief. I provide this infor- mation simply to orient readers unfamiliar with these ideas; those with a deeper interest should look to publications more focused on these topics. Computer Assisted Composition Spectral music uses many procedures which require calculations (such as the frequency calculations described in the first section of this article), but it is not truly algorithmic music. The calculations are required to gen- erate basic material (even the most basic conversions from frequencies to notes, for example, can be very time consuming when performed manu- ally); this material is not, however, used directly, but is manipulated musically by the composer. When these calculations represent a significant investment of time, it is difficult for composers to feel free with the material generated. They are unlikely to throw away weeks of elaborate calculation just because it is not exactly what they sought. They are more likely to perhaps tweak it a bit and then make do. Yet this freedom to experiment and to evaluate (even extremely complex) mater- ial is exactly what the spectral composers needed. The timing was fortu- nate, in that computers were beginning to become prevalent and their usefulness for this application was evident. For the computer, none of these calculations were of significant complexity and, thus, with the proper environment a composer could work freely and intuitively with a material of almost any complexity. This was also useful for demystifying calculation. Once the hours, days and weeks were converted into seconds, it became harder to attribute any abstract significance to the mere fact of calculation and returned the emphasis to the real end product: the musical result. The earliest Computer Assisted Composition (CAC) programs were of relatively limited scope. They were created for or by single composers or small groups. Two notable examples were pro- grams written by Tristan Murail (in the early 1980's) and another program that regrouped various musical functions used by a few com- posers at IRCAM in the package Esquisse (in the late 1980's). The first effort to create a more general environment which would allow many composers to develop the personal environments they would need for their work was the program PatchWork (first conceived by Magnus Larson, then developed at IRCAM by Camillo Rueda, Gérard Assayag and Carlos Agon). In the earlier and mid 1990's, this program spread the use of CAC far beyond previous levels and has also allowed spectral composers to find much more freedom in their daily compositional work. This program is essentially a graphical musical programming lan- guage, which gives the composer the power to make personal functionsGide tothe Basic Concepts and Techiqus of Spectral Misc 111 and environments. OpenMusic, a successor to PatchWork with increased capability and better graphics, is now in development at IRCAM. AudioSculpt The ability to perform spectral analyses used to be the province of uni- versities and research centers. Composers would keep collections of results which they could consult. The Macintosh program AudioSculpt, developed at IRCAM by Chris Rogers and Peter Hanappe in the mid 1990's, has changed this situation. AudioSculpt gives access to sono- grams, spectral analyses with or without Terhardt’s salience algorithm, simple partial tracking, and easy connections for using the data pro- duced by AudioSculpt within the computer assisted composition envir- onment PatchWork. This allows composers to generate and manipulate complex spectral materials at home, in a format that integrates well with the rest of their compositional environment. synthesis techniques Spectral composers often use, either directly or by analogy, different tech- niques developed for sound synthesis. Therefore, it seems necessary to give brief definitions of a few of the most influential of these techniques. additive synthesis This technique creates complex sounds by combining simple sounds with different amplitudes. The mixture stops on organs are the oldest form of additive synthesis. While theoretically capable of producing any sound and conceptually attractive and intuitive, additive synthesis requires an enormous number of partials and complex control to produce even mod- erately satisfactory results. Thus this technique, in the past, was often used for the generation of relatively simple sounds. Now there are pro- grams which generate the parameters for controlling additive synthesis based upon the results of analyses. These analyses can, of course, be modified before the sounds are re-synthesized. This powerful technique, called analysis/resynthesis, has caused a resurgence in the use of addi- tive synthesis. As described in the first section, the concept of additive synthesis has been extremely influential upon spectral musicians. subtractive synthesis The opposite of additive synthesis, subtractive synthi extremely complex sounds, often white noise (sound with equal energy in all frequencies), which is then filtered to leave only the desired portion of the sound. This technique, also theoretically capable of producing any , begins with112. Appendix 1 sound, is easier to use than additive synthesis: since non-controlled para- meters already exhibit complexity and, thus, all sonic richness does not need to be precisely specified. Conceptually, however, this technique is less intuitive and the results, while richer than those produced by addi- tive synthesis, are often much harder to control. Spectral composers have made relatively little use of this technique. FM synthesis ‘The technique of FM synthesis, developed by John Chowning, has been widely used because of its ability to produce fairly complex sounds with relatively little computational cost and relatively simple control parame- ters. The spectral content of FM sounds was explained in the first section of this paper. What must be noted here is that the FM synthesis tech- nique is not (even theoretically) capable of producing all possible sounds. They create a large but finite class of sounds, which often have a certain family resemblance. This class of sounds is very present in the cars of many spectral composers, through Yamaha’s DX series synthesiz~ ers which were based on it. It is especially successful for metallic sounds but much less so for strings. sampling ‘The most common technique in newer commercial synthesizers is sam- pling, This technique uses tiny bits of recorded sound (samples) which are the replayed with modified envelopes, filtered, stretched, looped, etc. to create the final sound. This technique has provided little in the way of conceptual ideas to spectral composers, but as a common technique it has often been used in the electronic portions of their pieces. tape loop, re-injection loop, echo and delay ‘These techniques are all based on the recording of a sound followed by a playback of that sound after an interval of time. The oldest technique was the tape loop. A loop of tape was passed between two open-reel tape machines, one for recording and the other for play-back. The distance between the machines and the speed of the tape determined the amount of delay. When the input is an open microphone, this is called*a re- injection loop, since the play-back is re-injected into the recording through the mixing board: producing a proliferation of sound. These techniques are now generally modeled digitally — with the delay replacing the tape loop and the echo replacing the re-injection loop. The advantage to these digital units is a finer control (for example, the precise amount of sound re-injected into the echo can be regulated with a feed-back control) andGuile to the Basic Concepts an Tecloniques of Spectral Music 113. much better sound quality: The concept underlying these procedures sound copying itself to generate a sonic structure different from the origi- nal sound — has influenced many spectral musicians. software source Many of the computer programs described above were developed at IRCAM and are available through their Forum users group. Those desir- ing more information can contact IRCAM through their web page at http://www.ircam.fr or by mail (Forum IRCAM, 1 place Igor Stravinsky, F-75004 Paris, France).