Hearing Musical Streams Author(s): Stephen McAdams and Albert Bregman Reviewed work(s): Source: Computer Music Journal, Vol. 3, No. 4 (Dec., 1979), pp. 26-43+60 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4617866 . Accessed: 26/04/2012 06:00

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

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

http://www.jstor.org Hearing Musical Streams

Stephen McAdams AlbertBregman Hearing and Speech Sciences Departmentof Psychology Stanford University School of Medicine McGillUniversity Stanford, California 94305 Montreal,Quebec, Canada

Introduction

Theperceptual effects of a soundare dependent upon haveintrigued us withcompositional possibilities. This article the musicalcontext in whichthat sound is imbedded.That is, thuspresents, in a tutorialfashion, a reviewof research(in- a givensound's perceived pitch, timbre, and loudness are in- cludingour own) whichhas directimplications for musi- fluencedby the soundsthat precedeit, coincidewith it, cians,especially for composersworking with computermusic. andeven follow it in time.Thus, this context influences the In all of ourresearch and in most otherresearch cited, com- waya listenerwill associatethe soundwith various melodic, puters(predominantly PDP-11 minicomputers) were used to rhythmic,dynamic, harmonic, and timbralstructures within synthesizethe soundspresented. They were also usedfor the musicalsequence. It thus behoovesthe composerand presentationof soundstimuli, collection of responsesfrom the interpreterto understandthe variousperceptual organizing listeners,and analysisof the data.For thoroughsummaries principlesthat affect the derivationof musicalcontext from andtheoretical treatments of this areaof research,see [2, 5, sequencesof acousticevents. We include the interpreterhere 9.] . becauseseveral musical dimensions, such as timbre,attack and decay transients,and tempo, are often not specified WhatIs An AuditoryStream? exactlyby the composerand are controlled by the performer. In this articlewe shalldiscuss principles that describe Auditorystream formation theory is concernedwith how variousmusical dimensions affect the perceivedcontinui- how the auditorysystem determines whether a sequenceof ty of music.Leon van Noorden has statedthat "insequences acousticevents results from one, or morethan one, "source." wherethe tones follow one anotherin quick succession, A physical"source" may be consideredas somesequence of effectsare observedwhich indicate that the tones arenot acousticevents emanating from one location.A "stream" processedindividually by the perceptionsystem. On the one is a psychologicalorganization that mentally represents such handwe find varioustypes of mutualinteraction between a sequenceand displays a certaininternal consistency, or con- successivetones, such as forwardand backwardmasking, tinuity, that allowsthat sequenceto be interpretedas a loudnessinteractions and duration interactions. On the other "whole."By way of example,two possibleperceptual or- hand,a kind of connectionis foundbetween the successive ganizationsof a repeatingsix-tone sequence are illustrated in perceivedtones." [28, p.1 ]. As for simultaneoussonic events, Figure1. Timeis representedon the horizontalaxis and Bregman[5, 9] has suggestedthat differentsounds are ex- frequencyis representedon the verticalaxis. The dotted lines tractedaccording to variousperceptual and cognitiveorga- connectingthe tonesin the figureindicate the streampercepts. nizationalmechanisms from the superimposedacoustic vibra- In the first configuration,six tones areheard one afterthe tions. Whilesome researchers would like to attributethese otherin a continuouslyrepeating cycle (TapedIllustration phenomenato mechanismsin the peripheralor earlycentral la)l; it is easy to follow the entiremelodic pattern. In the nervoussystem (which are surely involved to some extent, secondpercept, though, one mighthear two separatethree- see [24, 33]), prominentmembers of thismore psychophys- tone patternswhich appearto have little relationshipto icallyoriented "school" are beginning to thinkthe organiza- eachother (Taped Illustration 1b). It is difficultin this case tion is too complexto be so easilyexplained. However, rather to followthe originalsix-tone pattern. Note thatin the first thandelve into theoreticalexplanations of thesephenomena, exampleone streamis heard,and in the second,two areheard. it will sufficefor the presentpurpose to describethem in andto somesalient that general brieflyquantify parameters (1) A tapecontaining sound examples is availablefrom Mr.McAdams. Descriptions of the tapedillustrations ? 1979 by StephenMcAdams and Albert Bregman arefound in Appendix1. Page 26 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 D D D B"B /B B' % F F F - I I . "

II I I I Cr - O ST % %

(a) (b)

Time -, Time- One Stream Two Streams

Figure 1. A repeating6-tone sequence composed of interspersedhigh and low tones can result in different percepts. In Figure la, with high and low tones alternatingat a tempo of 5 tones/sec., one perceptualstream is heard. In Figure 1b, at a tempo of 10 tones/sec., the high tones perceptuallysegregate from the low tones to form two streams (cf. Taped Illustration1).

D

4 F t t D o 0

L. C F C U-U C"=I u

A ,/ A /

Time -+ Time -+

J II JU :I I:JJHigh Stream Ji I:JJJ:Low Stream HighStream ll Low Stream

Figure2. Due to the competition among streamorganizations, tone F may be perceivedas belongingto either the higherstream or the lower stream but not to both. The organizationof the streams changes, among other things, the perceived rhythmic structure,as indicatedunder each diagram(cf. Taped Illustration2).

Stephen McAdamsand AlbertBregman: Hearing Musical Streams Page27 In the rest of this section we will discuss some of the can group with either the higher or lower stream (Taped propertiesexhibited by a stream. For example, it is possible Illustration2). Notice that this regroupingresults in a rhyth- to focus one's attention on a stream and follow it through mic transformation as shown under each configuration. time [6, 28]. In the first taped example one can follow the Some non-musicalexamples of this phenomenonare the face- six-tone pattern without any trouble. Thus a stream must vase illusion and the reversibleNecker cubes; Escher and exhibit a certaincoherence over time. However,in the second Vasarelyhave produced art works based on such principles example, it is difficult to follow the six-tone pattern, but it is of perceptualorganization. easy to follow either three-tone pattern. Notice that one can Finally, this bringsup the relationshipbetween "source" pay attention to either the higheror lower stream,switching and "stream."A streamis perceivedas emanatingfrom a single between them at will, but that it is not possible to attend to source. So, in the first example, the pattern was fairly con- both simultaneously (Taped Illustration ib). Indeed, each tinuous and was easily recognizedas coming from one source; distance between the three-tone patternin Figure lb constitutes a separatestream but in Figure lb, the large frequency and maintainsits own temporalcoherence. While a listeneris two three-note groups introduced a sort of discontinuity paying attention to one coherent stream, other acoustic in- that caused the perceptual system to interpret the sequence formation is perceptually relegated to the background. If as resultingfrom two sources. Since at any given moment the one group of sounds is distinct enough, the foreground-back- composite pressurevariations stimulating the ear result from ground relationmay be almost involuntaryand it may require severalsources, the auditory system needs a battery of heuris- a greatdeal of attentional effort to focus on streamsinitially tics to parse, or segregate,the information into separate relegatedto the background. streams.It thus needs to build a descriptionof the acoustic The information-processingnature of the streamsegre- environmentfrom separatedescriptions of the variousstreams which the gation process is suggestedby the observationthat the segre- and the relationshipsbetween them [2]. Factors of and gation of a sequence into smallerstreams takes time to occur perceptualsystem uses to build descriptions streams, rate of occurrence of [4, 13]. In Figure lb, notice that one can hear a six-tone pat- subsequently sources, are frequency, tern for the first few cycles before it segregatesinto two events (or tempo), intensity, timbre, and attack/decay tran- the of the these will be discussedin some separatestreams (Taped Illustrationlb). It thus appearsthat sients. In rest paper are the perceptualsystem assumes things are coming from one detail. Of course it is obvious that sounds assignedpercep- the sources are at source until it acquires enough information to suggest an tually to different sources when physical this alternateinterpretation. different spatial positions. In case, intensity, spectral, all utilized to the sound into When temporal coherence is lost in a sequence, it be- and temporal cues are parse we will confine our comes more difficult to order the events of that sequence in separate sources. However, primarily of sources which occurs time [6, 11, 14, 28] . In Figure la, it would be easy to tell the discussion to the illusion many a emanationof sound. order of tones A, B, and C. But as this largerstream breaks due to the organizationswithin single down into the smallerstreams of Figure ib, i.e., as temporal coherenceis lost, it becomes more difficult to judge the order Frequencyand Tempo of these tones. In such a case one might notice that tone A comes before tone C but it would be hard to tell whether Considerthat a repetitivecycle of tones spreadover a tone B came before A, between A and C, or after C. This certain frequency range may be temporally coherent, or in- statement should be qualifiedby noting that in the tone se- tegrated,at a particulartempo. It is possible to graduallyin- quences mentioned all of the tones are of equal loudness and crease the tempo until certain tones group together into timbre. The tone sequences are furthermorecontinually separatestreams on the basis of frequency,as discussedabove. recyclingand are faded in so that the listener cannot label tone The faster the tempo, the greaterthe degree of breakdownor A as being the first tone in the sequence. Informationin a decomposition into narrowerstreams until ultimately every musicalcontext, such as the first beat being emphasizedas a given frequency might be beating along in its own stream. downbeat, may give the listener an anchorpoint againstwhich This last possibility is dependent upon a number of other to relate the temporalpositions of other tones. Thus one can factors which will be discussedlater. judge the order of events in time within a given perceptual Figure 3 illustratesthe possible stages of perceptual streambut not necessarilyacross streams. Accompanying this decompositionfor a recyclingsix-tone pattern as one gradual- is the observationthat different streamscan appearto overlap ly increases the tempo (Taped Illustration 3). Note that in time even when they do not. Again in Figure lb, notice streamsper se are not trackedbeyond a certainpoint, but a that one can hear two three-tone patterns apparentlygoing texture or timbre is perceivedsince the ability to temporally on at the same time even though the tones are alternating resolvethe individualtones degeneratesaltogether. Of course, (Taped Illustrationib). one could hold the tempo constant and graduallyexpand the effect If an event is potentially a memberof more than one frequency relationshipsto achieve a similar streaming would be "competing"stream, one may perceiveit as belongingto one [3, 6, 14, 17], but the musicalconsequences vastly streamor anotherbut not to both simultaneously[3, 10, 11] . different as one can hear in Taped Illustration4. Dowling [17] This is not to say that a musiciancannot hear severalsimul- used simple melodies to illustrate this frequency-based taneous lines. The point is that it is impossibleto use several streamingprinciple. He interleavedtwo melodies in the same it without parsing schemes at the same time. Figure 2 illustrates the frequencyrange thereby making very difficult, prior effect on our second example of moving the higher triplet knowledge of the melodies, to separatethem perceptually. the into closer frequencyproximity to the lower triplet. We can But as they were pulled apartin frequency,i.e. when all each find an intermediateposition where the lowest tone, tone F, tones of one of the melodies were transposedupward, melody became apparent(Taped Illustration5).

Page 28 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 A&B A&B f I ? I r r Isolated Absorbed 0 i I I I I I ! I - 0. I A A , • . .2-. , =. CL " 0 ccL 0)0 04 % cm je.-" Q/) -x,0.1 r- IE Y S S O" > B 'O B ca) oo 0 (D L-- CDE O

CC X ------u, --o,)=------, QI - x Time Time------? -I Figure4. This figure illustratesthe effect of frequencycontext I> Time- as opposed to frequency separationon streamformation. In the first part, tones A and B form one perceptualstream which is unaffected by the simultaneousstream composed of tones X Timbre/Texture and Y. Whentones X and Y are moved into the proximity of tones A and B, which remain unchanged, an alternate Figure 3. This figure illustrates the decomposition of an perceptualinterpretation results whereby tones A and B are acoustic sequence into smallerand smallerperceptual streams assignedto separatestreams on the basisof a new frequency as the frequency separationbetween the tones or the tempo of context (cf. Taped Illustration6). the sequence increases.In the latter case, a point is ultimately reached where one can no longer perceive individualtonal events; a texture or timbre is heard instead (cf. Taped Illustra- tions 3 and 4). the rateof alternationof the toneson the otheraxis. Note that the horizontalaxis indicatesincreasing tone repetition Thefrequency range within which the perceptualsystem time,which corresponds to decreasingtempo. One can draw groupstones on the basisof frequencyproximity is not boundarieson this graphindicating the frequency-tempo constant;the groupingcan varywith the particularpattern regionsin whichthe tonescohere as a singlestream and those of frequenciespresented. For example(see Figure4), two in whichthey segregateinto two simultaneousstreams of tones, A and B, are arrangedwith given frequencyand differentfrequencies. There are two such boundaries.Leon temporalseparations such that they will alwaysstream VanNoorden [28] has termedthese the fission boundary and togetherwhen no othertones are present.We can createa temporalcoherence boundary and noted that they areslightly similarorganization with tones X andY in anotherfrequency differentfor eachperson. In Figure5 the uppercurve is the range.These two groupswill each forma streamas can be temporalcoherence boundary. Above this boundaryit is heardin TapedIllustration 6a. By bringingthe pairsinto the impossibleto integratethe two alternatingtones into one samefrequency range, new streamscan be formed,such as stream.Below this boundarylies the temporalcoherence A-X and B-Y, on the basisof an alternativeproximity region,where it is possibleto integrateevents into a single organization[3] (TapedIllustration 6b). Thus,the particular perceptualstream. The lower boundaryis the fission relationshipsbetween frequencies in a tonalpattern, and not boundary.Below this it is impossibleto hearmore than one just the frequencyseparation between adjacent tones, plays stream.Note that the regionsof fissionand coherencealso a vitalrole in the formationof streams. overlap,creating an ambiguousregion where either percept It appearsfrom our examples that there is an essentially maybe heard. inverseand strictly interdependentrelationship between As the tempo decreasestemporal coherence can be tempoand frequencyrelationships among individual tones maintainedat greaterfrequency separations. However, the [26, 28]: the fasterthe tonesfollow one another,the smaller frequencyseparation required for fission remainsfairly the frequencyseparation at whichthey segregateinto separate constantwith decreasing tempo. The temporal coherence and perceptualstreams. Conversely, the greaterthe frequency fission boundariesabove and below a given tone are separation,the slowerthe tempoat whichsegregation occurs. symmetricalwith respectto pitch. Thisindicates that the Thisillustrates yet anotheraspect of musicwhich, with the phenomenaof temporalcoherence and fissionmay occur aid of the computer,can come under the composer'scontrol. dependingon the absolutemusical interval between the tones Supposewe makea graph,as in Figure5, whichrelates but irrespectiveof the directionof pitchchange. While there frequencyseparation of two alternatingtones on one axisto are substantialquantitative differences in theseboundaries Stephen McAdamsand Albert Bregman:Hearing Musical Streams Page29 FrequencyTrajectories

Tempo (Tones/Sec) Another frequency-based effect involves frequency 20 10 5 3 trajectories. These are important on two levels. The first involves trajectoriesbetween tones (see Figure 6). Bregman and Dannenbring[7] have found that tones that are con- nected by glissandiare much less likely to segregateunder given conditions of tempo and frequency separationthan o 15 those which make abruptfrequency transitions.Intermediate situations beget intermediateresults. Yet even when using a Al sine wave for frequencymodulation of a sine tone, it is pos- / Alwaysays effect of the and Segregated, sible to discerna sort of streaming higher O c/ 10 5

_I I

a I Tone Repetition TiAm biguous D = IRegion I I I Always I Fission BoundaryCoen Ramped Semi- I I I 0 50 100 150 200 250 300 350 400 r- I Tone Repetition Time (msec) a., Sta 1 a Tas a Figure 5. Boundaries of temporal coherence (upper curve) and rapeady- fission (lower curve) define three perceptual regions in the stream relationship between two alternating tones each lasting I ______40 msec. This relationship is a function of both the tempo of alternation and the frequency separation between the tones (after [28] ).

Discrete aaTime-+

Steady- Transition Steady- between listeners, the qualitativetrends tend to be similar State State [28]. Note that the difference between boundaries is increasinglysubstantial for tempi below about 10 tones/sec. Figure 6. These are the 3 types of frequency transitions (tone repetition time = 100 msec.). This means that at very between tones used by Bregman and Dannenbring (1973). In slow tempi of, say, five tones/sec., a separationof more than the top section the tones are completely connected by a a minor 10th is necessaryto induce streaming.Below this it frequency glissando. In the middle section an interrupted becomes virtually impossible to induce streaming. In the glissando is directed towards the succeeding tone. No limiting case, if the temporal distance between tones is too frequency glide occurs in the bottom section; the first tone great, they do not seem to be connected at all but sound ends on one frequency and the next tone begins on another as isolated events. (after [71 ). The unconnected tones segregate more readily A musicallyrelevant aspect of these boundariesshould than the others. be mentioned. The region between the two boundariesmay be consideredto be an ambiguousregion since either a segregated or an integrated percept may be heard. The primary determiningfactor in this regionis attention. In other words, lower peaks of the modulation at certainmodulation frequen- it is possible to shift one's attention back and forth between cies (Taped Illustration7). At lower modulating frequencies the two perceptswhen the sequence falls in this region. For one can track the modulation, but higher modulating fre- example, one may focus either on a whole streampercept or quencies result in the effect of a texture or timbre. This con- on smaller individualstreams. It appearsthat the closer the tinuum is found for modulation involving discrete changes sequence lies to one of the boundaries,the easierit is to focus in frequency as well. on the percept which is predominantbeyond that boundary. The second type of trajectorymight be called a melodic Conversely,it is very difficult in this situation to shift one's trajectory. The basic rule goes: large jumps and sudden attention back to the other percept. Once the physical values changes in direction of a melody produce discontinuity in go beyond either of these boundaries,attaining the comple- that melody. In terms of streamformation, one or two tones mentary percept may be consideredto be impossible. The role in a melody that are removedfrom the melodic continuity of of attention will be discussedin more detail later. the rest could be perceivedas coming from a different source

Page 30 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 and would not be integratedinto the melody as a whole, terns presented to opposite ears segregateinto upright and perhapsleaving a rhythmic gap in the phrase depending on inverted V-shaped melodic contours [16]. Each contour the nature of the main sequence. It has been reported,how- is heard as if being presentedto one ear. The same result was ever, that the excluded tones are sometimes noticed in the found by Halpern [22] for simultaneousascending and de- background,with their absencehaving little effect on the main scendingsine tone glissandi,as can be heardin Taped Illustra- melody line [23]. Implied polyphony in a solo compound tion 8; in this case both glissandiwere presentedto both ears. melody line, as in the suites for solo instrumentsby Bach, is a In these two examples a stream boundary is established at compositionaluse of this principle. the pitch where the two lines cross. Schouten [30] reported that if an ascendingand de- scendingmajor scale fragmentplayed with sine tones is con- a a tinually repeated, temporal coherence is maintained up to a tempo of about 20 tones/sec. However, if these same tones High I are arrangedat random,the maximumtempo at which coher- \ Contour I ence still occurs is reduced to about 5-10 tones/sec. It might be inferred that the reduction of predictabilityreduced the t 1 pitch boundary within which the auditory system could --I successfully integrate the incoming information. But van Noorden [28] found that previous knowledge of the order of tones had no effect on the coherenceboundary. It might be that the small in frequencyjumps Schouten's demonstra- Low tion are effective in holding the streamtogether. S L a Low ' Heise and Miller four melodic / [23] investigated con- .Contour _ tours: a V-shaped contour, an invertedV, and risingand fall- ing scale patterns;the V-shaped patterns which change di- rection can be of as less thought being "predictable"than Time-+ the scale patterns which move in only one direction. Each patternwas eleven tones long and the frequency of the middle 7. This stimulus, used by Deutsch (1975), is com- tone was variable.They found that the degreeto which the Figure posed of ascending and descendingscales presented simul- variabletone could be separatedin frequency from the rest taneously to opposite ears. Two V-shaped patterns of the patternbefore it segregatedinto its own streamwas a (high and low contours outlined with dashes)are perceivedrather function both of the shape and of the steepnessof the pattern. than the complete ascending and descendingscale patterns. The steepnesswas variedby keeping the tempo constant and varyingthe intervalbetween successivetones. As the rate of frequency change for the entire pattern increases over time, Loudnessand Continuity Effects so does the amount of frequency separationof the middle tone requiredto produce segregation.Less separationof the Fissioncan be obtainedby alternatingsequences of middle tone is required to produce segregationwith the tonessimilar in frequencyand timbre but differingin intens- V-shaped patterns than with the scale patterns, possibly ity. Figure8 illustratesthe rangeof perceptsachieved as the indicating that the perceptualsystem can follow "predict- amplitudelevel of tone A is variedrelative to thatof tone B able" patterns more quickly than "unpredictable"ones. in the alternatingsequence ABAB... . The referencelevel (However,certain anomaliesin their data, which will not be for tone B usedby vanNoorden in theseexperiments [28] discussedhere, might suggestother interpretations.) was35 db SPL;the frequencyof both toneswas 1 KHzand Van Noorden found similarresults for patternswith as each tone lasted40 msec.If the level of tone A is below few as three tones [28] .He investigatedthe relativetemporal the auditorythreshold (approximately 6 db SPLat 1 KHz), coherence of so-called linear and angular three-tone se- only the B streamis heardat half tempo,as mightbe expected quences. For linear sequences, i.e., those with two tone (see Figure8a). Whentone A is loudenough to be heardand intervalsin the samedirection, the temporalcoherence bounda- is at least 5 db belowtone B, two separatestreams of dif- ry occurs at faster tempi than were found with angularse- ferent loudness can be perceived,each at half tempo, with A quences in which the melodic pattern changed direction at being the softer stream(see Figure8b). WhenA is within 5 the middle tone. In this latter case the first and third tones db of B, a "pulsing"stream is heard and neither the A nor are more contiguous in frequency than are the first and the B streamcan be heardindependently, i.e., temporal second, or second and third tones, which facilitatesa loss of coherenceis inevitablein this range(see Figure8c). As the temporal coherence similarto that found by Schouten. It levelof the A tone is increasedabove that of the B tone, three should be noted that while these melodic trajectory effects differentpercepts may result,depending on the alternation do seem to play a role in musical streamformation beyond tempoof A andB. If thistempo is less thanabout 13 tones/ the simple frequencyseparation effect, they are certainlycon- sec., fissionis the next perceptheard. This time B is the softer founded by other contextual organizations. stream(see Figure8d). Thusa certaindegree of loudness Both of these trajectoryexamples illustrate a principle differenceallows us to focus on eitherstream using only of perceptual organizationthat uses pattern continuity, thisinformation. If the tempois greaterthan about 12.5 -13 in some form or another, as a criterionfor "source"distinc- tones/sec.,the perceptencountered is the "roll"effect tion. Figure7 illustrates,however, that frequency proximity discoveredby vanNoorden. It soundsas if streamA, the may sometimescompete with trajectoryorganization; Deutsch louderstream, were pulsing at half tempoas in the fission found that simultaneousascending and descendingscale pat- percept,but the B streamsounds as if it werepulsing at full Stephen McAdamsand AlbertBregman: Hearing Musical Streams Page31 tempo (see Figure8e). Thus the A streammay be heardin- A Below Threshold dependently, but not the B stream. In other words, it is as if the A tones consisted of two parts: one that combineswith a -K A B ( B the B stream to give a full tempo roll, and another, at half tempo, which can be perceived separately.At a tempo of time about 13 tones/sec. another effect emerges when the level Fission of A is about 18-30 db above that of B. This is the continuity effect shown in Figure 8f, so named because tone B is not heardas pulsingbut ratheras a fairly soft, continuous tone b fA B A I under the louder, pulsing A stream; this is an example of a class of effects that will be discussed below. Finally, if the Coherence level of the A streamis incrementedstill further,this stream completely masksthe B stream(Figure 8g). Again, this set of loudness-based phenomena exhibits an ambiguous region c A B A B between the coherence and fission boundarieswhere one might pay attention to either the A or B streamsindividually. or to the AB stream as a whole. There are thus three Fission per- ceptual regions for alternatingtones at the same frequency where the tempo is above about 12.5 tone/sec.: the roll region, d A B A B the continuity region, and the temporalcoherence region. Van Noorden made quantitativemeasurements of the fission boundary (see Figure 9). For tempi of about 2.5 Roll to 10 tones/sec., the fission boundary is more or less hori- zonal, i.e., the intensity difference (AL) necessary for a segregatedpercept does not change with tempo over this range,but lies about 2 to 4 db on either side of the reference e A B A B tone level. For tempi less than 2.5 tones/sec., the minimum level difference for fission increaseswith decreasingtempo (or longer inter-tone intervalsof silence) and is symmetric Continuity about AL = 0 (no difference in level). For tempi greater than 10 tones/sec. the level difference at which fission occurs increaseswith increasingtempo but the situation is not sym-

Tempo (Tones/sec) 20 0 5 3 2 Masking SContinuity " LA> Masking r 20o Fission LA >LBLB

08., Roll g Inevitable A B, A ev le 35 db SPL I _Coherence --*

20 Fission LA < LB j3 Sub-Threshold ? 8. This illustrates the of Figure figure range possible per- 0 40 200 400 60o 800 cepts found by van Noorden (1975) for two alternatingsine tones differing in each case only in their intensities.Tone B Tone RepetitionTime (msec) was kept at 35 db SPL throughout; both tones were 40 msec. long and had a frequency of 1 KHz. a) The level of tone A is below the auditory threshold. b) Tone A is at Figure9. The perceptsshown in Figure8 fall into different least 5 db below tone B. c) Tone A is within 5 db of tone B. regionsof tempo-intensity combinations.The verticaldotted d) Tone A is louder than tone B with an alternationtempo line correspondsto the durationof each tone. The level dif- less than about 13 tones/sec. e) Tone A is louder than tone ferences between tones A and B are relativeto the level of B with more than 13 tones/sec. f) Tone A is about 18-30 tone B at 35 db SPL. The horizontal axis is shown as tone db louder than tone B and the tempo is still above 13 tones repetition time on the bottom and as its reciprocal,tempo, /sec. g) Tone A is more than 30 db louderthan tone B. The on the top. It should be pointed out that Figure5 and Figure arrows indicate the percepts reported; a more complete 9 cannot be directly compared,because the alternatingtones descriptionis given in the text. in Figure5 were played at different frequencies.

Page 32 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 metrical about AL = 0; for tempi greater than this, there is the complex interactionof level differenceand tempo which resultsin the roll, continuity, and maskingregions. A combination of frequency separationand level dif- ference was used to obtain the roll and pulsation thresholds shown for two different tempi in Figure 10. A great deal of psychoacoustic researchhas been devoted to the boundary between the roll and continuity regions. The "pulsation threshold,"as this boundaryhas been termed [24, 28, 33], EI requiresa greater loudness difference at slower tempi than does the roll threshold, as can be seen in Figure 10. The continuity effect may operate accordingto prin- , ciples very similarto those which governocclusion in vision. The interpretationthat one object occludes anotheris given by informationderived from the intersectingedges of the ob- jects. In Figure 11, the fragmentshave no apparentstructure since informationregarding the shape of the occluding struc- Figure 11. These fragments have little apparent structure and ture is not present. However,in Figure 12, with the appropri- meaning since information regarding the shape of the ate edge information, the fragmentscan be grouped into occluding structure is not present. meaningful chunks. The same process may be postulated to occur in audition. That is, the "edges"of a louder sound and a contiguous softer sound might be used to induce the perceptionthat the less prominentsound is partiallyoccluded, or masked,by the louder. In such a case, the illusion of con- tinuity of the softer sound "behind" the louder sound may result. This has been found to be true for sine tones masking sine noise sine and noise noise tones, masking tones, masking r [8, 14, 24, 28. 32, 33, 34]. 1 Tempo - 21 Tones/sec Tempo a 16 Tones/sec 40 I il/ g i ,? IC 30"\-0continuity +- ro

Figure 12. With appropriate information about the edges of the occluding structure, the fragments of Figure 11 can be Qo\ fission \ fession > 01 grouped into meaningful chunks. _

That the nature of the temporal "edge" between the 0 0-400 -200 200 -400 -200 0 200 two sounds is the importantfactor was illustratedby Bregman and Dannenbring[8]. As Figure 13 illustrates, they varied kHz 1 LB = 35db SPL fB = 1 fB = kHz the transitionalamplitude of a tone both leading into the attack and coming out of the decay of a noise masker. In Frequency Difference (Hz.) reality, the tone stopped the instant the noise began and then began againthe instant the noise stopped. The greatestcon- Figure 10. The boundaries between the fission, roll, and tinuity was found when there was no changein the amplitude continuity regions already discussed in Figures 8 and 9 are of the tone. Whenthe level of the tone was either increased shown here as a function of frequency and level differences. or decreasedbefore the attack of the noise, the illusion of There is an increase in slope for roll and pulsation thresholds continuity was substantially diminished. The amount of and an upward shift in the latter at slower tempi. The change in judged continuity was greater for the initially frequency and level differences are given relative to the decreasingamplitude ramp than for the initially increasing frequency and level of tone B, i.e. 1000 Hz. and 35 db SPL, ramp.The changein the amplitudeof the tone may suggesta respectively. 0 KHz. frequency separation corresponds to the possible terminationof the tone sometime duringthe noise vertical dashed line in Figure 9. (Note that Figure 9 cannot be burst. The length of the rampwas also found to have an effect mapped directly onto Figure 10; in the former illustration, on perceivedcontinuity. The least continuity was found when listeners were asked to try to hear fission, but in Figure 10, the amplitude rampslasted 50 msec., and for longer ramps the test subjects were listening for roll. The different focus on there was less of a break in perceived continuity. This may the part of the listener results in a shift in the boundaries suggestthat a very long rampwould be interpretedas no ramp shown.) since the rate of change of intensity would approachzero.

Stephen McAdams and Albert Bregman: Hearing Musical Streams Page 33 Timbre

Another musically useful dimension that can play a 76 db Tone 91 db role in streamingeffects is timbre. Timbre tends to be the 66 0Noise multidimensionalwaste-basket db-ne psychoacoustician's category for everything that cannot be labeled pitch or loudness, includingshort-term spectral changes such as onset transients, long-term spectra,those dynamicqualities which a musician would term "texture," and so on. The important thing to notice about streamformation is that of simultane- 76 or 66 db Tone 91 db any group which move to each Noise - - ous spectralcomponents roughly parallel other in the pitch (logarithmicfrequency) domain are most likely to result in a fused percept and thus to be interpreted as emanatingfrom one source [22, 25] . So even though the individualspectral envelopes of two instrumentsor sources 76db changeover time, they each follow patternswhich are suffi- \ 91 db ciently consistent internally to be fused and yet different to them one from the other. 66 db Tone / Noise enough, in most cases, distinguish in which sen- -+ There are, of course, many situations especially Time sitive instrumentalistscan blend their tones and become as one voice, but these situations usually involve very slow, smooth changeswhere onset transientscannot give the listener the informationneeded to separatethe sources. As an example, Figure 13. The stimuli used by Bregmanand Dannenbring how often is a performanceof Ravel'sBolero preciseenough (1977) consisted of a pure tone interrupted by noise. so that the instrumentsplaying the various"harmonic" parts The level of the tone leading into and coming out of the are not heardindividually, but fuse into one voice? The fact noise burst was varied as shown. The greatest degree of that one typically hears them separatelypoints to the extreme continuity was found in the center case where there was sensitivity of the auditory system to even minute variations no change in level. in temporaland spectralelements. Let us examine one of the many aspects of timbre listed above in more detail, using a simplistic steady-state timbre It has also been that alternation has almost reported speed derivedfrom summedsinusoids, as was done to test the effect no effect on the continuity perceived [14]. of timbre on stream [25]. If the frequenciesof a An in relation to this segregation important finding phenomenon of sinusoidaltones, as shown in 14, are within was et sequence Figure reportedby Warren, al. [34]. They noted that there close that would form one stream if must be some of in the two proximity (such they degree spectral overlap signals none of them were it is to cause a subset of to obtain the effect. Withsine for enriched), possible continuity tones, example, these tones to into its own stream a the effect is when the tones are identical in segregate by inducing greatest frequency, timbre difference. In this the verticalline indicates that and falls off as the on figure, monotonically frequency separation timbreis created or at least co-exists with, the fusion of either side of the maskerincreases. The effect was even found by, components into a single percept (Taped Illustration11). We to occur througha burst of noise as long as 50 seconds [34] . will discuss the natureof perceptualfusion later. Though some people think the pulsation threshold A musicaluse of timbre-basedstreaming might work as reflects only a neural excitation pattern that decays grad- follows: Figure 15 shows an isotimbralmelodic phrase con- ually, with the perceptionof continuity resultingfrom over- tainingthree sub-melodies that are slowly to emergeas contra- lap of excitation at smallertime intervals[24, 33] , the follow- puntal melodies on their own. Each sub-melody is marked ing demonstrationillustrates that higher-level context-de- with a different symbol under the notes that are to belong to pendent predictivemechanisms may be involved.These seem it. Now we might cause the "X" melody to slowly changein to the predictive nature of the stream formation parallel timbreto a where its timbre differs sufficiently from the mechanismsmentioned In Illustration9, point previously. Taped rest of the stream to induce At this there one hears that has been it is difficult to under- separation. point speech gated; would be two lines their stand the Then the in the are contrapuntal distinguishedby respec- spoken message. gaps speech tive timbresas discussed Wessel several filled with white noise. This effect was first verified by [35]. Alternatively, experi- cause an Millerand Lickliderin 1950 The same simultaneous timbral modifications might equal mentally by [27]. number of streams to as sort of with music and then emerge separate contrapuntal phenomenon gated gated-plus- as is illustratedin the second of 15 music is 10. melodies, part Figure by noise-filled presented in Taped Illustration the "X" and streams.Note that as these two imbedded We have chosen a familiar to dramatize the "O" composition melodies the is since several effect. It be difficult to it the emerge, originalmelody changed may very recognize in gated of its elements have left and other is but should become when the joined groups. (It condition, easily recognizable to note that certain of inharmonicenrichment are filled with noise. In both of these demonstrations important types gaps of a tone the as well as the timbre, which the amount of removed and then or might change pitch information replaced, would be another case An of stream induced the is far too and far too altogether). understanding by perceptualsystem, large formation based on timbre modification is if one to be attributedto interactionbetween local- important complex entirely intends to use multitimbralmelodies compositionally, as ized excitation patterns.

Page 34 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 3f4 3f2

f4 " " .f4"

OneOne Timbre Stream Two TimbresStreams

Time - Time -

Figure 14. Timbredifferences between groupsof tones have been shown by McAdams(1977) to influence segregationof those groups. A repeating4-tone sequence can be decomposed into two streams when two of the tones are enriched with their respectivethird harmonics.The dotted lines indicatethe streampercepts and the verticalsolid lines representthe fusion and timbre of the complex tone (cf. Taped Illustration11).

Varise and Schoenberg,among others, have done. A problem TimbralTrajectories potentially ariseswhen largechanges in both timbre and pitch occur simultaneously.It becomes very difficult for the listener This leads into a discussion of the effects of timbral to follow the "melody" line if the tempo is above the listener's trajectorieson streamsegregation. As was mentioned before, fission boundaryfor that set of events. One's attention may if two glissandiof identical spectralstructure cross in frequen- tend to drift from one timbre set to another and relegatethe cy, it is reportedlyeasier for listeners to hear high and low V- rest to unattended streamsrather than following the sequence shaped contours than to hear the ascending and descending the composer intended. paths actually taken by each glissando. However, if the glissandido not cross, or if they differ in spectralstructure, it becomes possible to segregatethem and hear the risingand fallinglines separately.In Taped Illustration12 and in Figure 16, harmonics3, 4, and 5 of two different "missing"funda- xx o x XXA mentals are modulated exponentially in opposite directions.It is easy to hear them separatelysince neither the spectral components nor the "missing"fundamentals cross. In the next example (see Figure 17) two different spectralstructures are used in the same frequency range.The descendingglissando is composed of harmonics 10, 11, and 12 and the ascending glissando is composed of harmonics 3, 4, and 5 (Taped SI • - -h Illustration 13). One might argue that it is not a timbre I/li ?vi I • I/I # l differencethat allows one to distinguishthese patternsbut the w'_ #t,• .IF fact that the pitches do not cross, as is illustratedin Figure 17 by the dotted lines which show the "missing"fundamentals. Halpern [22] tested for this by using the same harmonic composition (see Figure 18) as in the last example and crossing Figure 15. This figure illustratesthe possible effect of changing the pitches of the glissandi,which would correspondto the the timbre of different sub-groupsof a melody line. The notes "missing"fundamentals, rather than crossing the frequency markedwith X's and O's in the originalstream are shown on components. The rising and falling glissandiwere still per- separatestaves below to indicate the rhythmictransformation ceived, which suggeststhat the segregatingeffect was based on that takes place after segregationoccurs. (Taken from the spectralcontent ratherthan on pitch (Taped Illustration14). A MajorPrelude, Well-Tempered Clavier, Book I, by J.S. For comparison,listen again to the pure tone crossingin Bach). Taped Illustration8. Some pre-tests done for both Halpern's Stephen McAdams and Albert Bregman: Hearing Musical Streams Page 35 4f, 3f,

2f2 5f2 1 f2

LL 4f LL 0 3f, 0

Time Time

Time Time -"

[22] and McAdams'[25] studies bore out the intuition that Figure 16 (above, left). Harmonics3, 4, and 5 of two different it is necessary to use constant ratio relationshipsamong "missing"fundamentals are modulated exponentially in fre- components to attain fusion and constant timbre. Further quency in opposite directions. Neither the frequenciesnor the researchon the fusion of harmonicsfor the buildingof timbre fundamentals of the two glissandi cross each other. Two is currently in progressin Bregman'slaboratory at McGill separate glissandi are heard (cf. Taped Illustration 12). University. Figure 17 (above, right). An ascendingglissando composed of harmonics3, 4, and 5 of one "missing"fundamental and a TimbreMovement Versus Pitch Movement descendingglissando composed of harmonics10, 11, and 12 of another "missing"fundamental are crossed in frequency, but One pitch phenomenonhas a scientific history which is the pitches of the "missing"fundamentals do not cross. Two almost coextensive with that of pitch researchin general.This different glissandi are heard (cf. Taped Illustration 13). phenomenonhas been termed the "missing"fundamental for Figure 18 (below). Here the same harmonic structureas in reasonsthat will become obvious. It is possible to construct Figure 17 is used but now the pitches of the "missing"funda- a tone from severalsinusoidal components whose frequencies mentalscross each other and the partialsdo not. Two separate are integer multiples of some fundamentalfrequency which streamsare still heard (cf. Taped Illustration14). may or may not be actually present in the final signal's spectrum.For example, frequenciesof 1000, 1250, 1500, and 1750 Hz would representthe 4th, 5th, 6th and 7th harmonics, respectively,of a fundamentalat 250 Hz. Whetheror not a spectralcomponent is present at the fundamentalfrequency, the perceived pitch almost always corresponds to the fundamental.Of interest in our discussionof timbre and streamingis the fact that tonal events with various spectral structurescan elicit the same "missing"fundamental as long as the components are integermultiples of that fundamentaland lof2 are within a certain so-called "existence region" [29] of 5f, harmonicsthat can actually contribute to the pitch. Thus C 4 fi different tonal events can elicit the same pitch but with 3fl different timbres. 0, Some of the uL first author'spersonal experience in work- 0) ing with the pitch of complex tones has led him to the conclu- 0 sion that it is possible for changesin pitch and timbre to con- tribute competinginformation to the overallperceived move- ment in tone sequences. Figure 19 illustratesthat is is possible Ts m f for timbre to get "sharper"or "brighter"[see for example 1, Tm- 18, 20], i.e. the spectral components move to a higher fre- quency range,while the pitch moves in the opposite direction. In Figure 19, while both complexes will elicit pitches corre- spondingto the "missing"fundamental, i.e. Fl and F2, the

Page 36 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 "centerof gravity"of the spectralenvelope, TI, of the first toneis lowerthan that of the secondtone, T2. Spectral"cen- ter of gravity"may be roughlydefined as the weightedmean of the spectralcomponents in an acousticevent averaged over the durationof the their derivedfrom ~T28f2 event, weightsbeing theirrelative amplitudes. of 5f, 7f2 Van Noordeninvestigated the possibility streaming basedon pitchcontiguity, as opposedto streamingbased on t TI4f, frequencycontiguity, i.e. continuitydue to the continued presenceof spectralcomponents. Figure 20 showsthe stimuli he used.The first consisted of two differentthree -component 3f, complextones with harmonics 3, 4, and5 for the firsttone Both U- andharmonics 8, 9, and 10 for the secondtone. tones havethe samefundamental frequency. Here an attemptwas madeto get "missing"fundamentals to streamtogether. In the f0I secondstimulus a puretone of frequencyf alternatedwith a complexcontaining harmonics 3 through10 of the same T-f2 fundamentalfrequency; this was an attemptto streama pure tone witha "missing"fundamental of the samefrequency. He Time foundthat it wasnot possiblewith either stimulus to obtain a coherentstream percept. However, it was necessarythat thesestimuli have vastly different spectral structures: if one were to partiallyoverlap the spectralstructures of the respectivetones, it would be impossibleto separatethe Figure 19. It is possible to construct two tones whose respec- perceptof streamingbased on frequencycontiguity from that tive spectral "centers of gravity" (see text) and "missing" basedon pitch contiguity.For example,if the first tone fundamentalpitches move in opposite directionswhen these consistedof harmonics3f, 4f, and5f, andthe secondtone tones are presentedin succession. consistedof harmonicsSf, 6f, and7f, how is one to knowif

lOf- - lOf -

9 f-••- 8f-

t-

C c5f- - 5f - - Tu4f

3fi-

Time -+ Time-

Figure20. Van Noorden (1975) tried alternatingtwo tone complexes of differingharmonic structure but identical "missing" fundamentaland also tried alternatinga sine tone with an 8-tone complex whose "missing"fundamental was at the same pitch as the pure tone. He was not able to obtain streamingof the fundamentalsin the first example or of the sine tone and the "missing"fundamental in the second example. StephenMcAdams and Albert Bregman: Hearing Musical Streams Page37 the streamingeffects reportedare due to a repetition of 5f or and thus attenuate the effect of the difference in timbre to the perceptionof a repeating"missing" fundamental, f. A between the two pairs.But this condition still evoked more primaryresult of this arrangementof harmonicswas that segregationthan two isotimbralcontrol cases, where all of the van Noorden used substantiallydifferent timbresin the two tones were composed of either one or two components. Of tones, and this condition may have contributedsubstantially interest here, again, is the fact that pitch and timbre move- to the preventionof streamformation. Equallylikely is the ment may work againsteach other. possibility that streamformation mechanisms do not use pitch While these timbre changes have proven to be a information but use frequency information. The following consistent irritant to the experimenterpursuing the elusive findingsmay lend support to the formernotion. "missing"fundamental, the implied interrelationshipsbetween Our work on timbre effects in stream formation [25] timbral"brightness" and pitch "height"may prove interesting found an anomalousresult that may have been due to pitch to investigatein terms of stream formationand to apply (both effects interactingwith timbreeffects. In Figure21, two of experimentally and compositionally) to melodic sequence the test cases used in that study are shown. In the first, the construction. Grey [18] has delineated a three-dimensional higher tone pair of the repeating four-tone sequence is timbre space roughly defined by 1) spectral distribution, enrichedwith the third harmonicand in the second case the 2) spectralfluctuation and synchronicityof higherharmonic lower pairis similarlyenriched. The first case was found to transients, and 3) low-amplitude, high-frequency attack segregatesignificantly more than the second. It is McAdams' transients.How does the "distance"between events in this contention that this is the result of an interaction between timbre space affect the temporal coherence and fission perceived pitch "height" and timbral "brightness."This boundaries?Do the effects obtained vary differently along "brightness"seems to be related primarilyto spectral"center these different dimensionsassociated with timbre? Research of gravity."In our example with the enrichedhigh pair, the by Wessel [35] and Grey and Gordon [19, 20] has provided pitch "height"is alreadygreater than that of the low pair and information which would allow the composer to deal with the addition of a highercomponent increasesthe "brightness" timbre as a compositional parameter.This information also of its timbre. The perceptualintegration of these two factors allows a certaindegree of prediction of the perceptualrelation- may facilitate the segregationof the tone pairsby resultingin ships between sounds resultingfrom such timbralvariations. an overallincrease in perceived"height" of the tones in the high pair. However,in the second condition where the low pair Fusion,Timbre, and Frequency Streaming was enriched, the increase in "brightness,"interacting with perceivedpitch "height," apparentlyserved to increase the The idea of competingperceptual organizations is an overall perceived "height" of the lower pair. This would excitingone for musiccomposition and theory. Bregman and bring it into closer perceptualproximity to the higher pair Pinker[10] investigatedthe competitionbetween sequential

High PairEnriched Low PairEnriched Strong Segregation WeakerSegregation

C 0 u, t Lu.

Cr E ca a) a) LL L_

_____- TmtTe %No - 0.0. ..-00:> 00 r ?

Time Time • -

Figure21. One result of McAdams'study (1977) suggestedthat there is an interactionbetween pitch "height"and timbral "sharp- ness" (see text). In a repeating4-tone sequence,one of the pairsof tones was selectively enriched by addingthe third harmonic. A greaterdegree of segregationof the high and low streamswas found for the formercase. The dashed lines indicatethe two- streampercepts and the dotted lines indicatea potential one-stream percept. The verticalsolid lines representthe fusion and timbre of the 2-tone complex.

Page 38 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 (or frequency) organizationsand simultaneous(or timbral) different attack characteristics.In addition, there is a very organizationsin the formation of auditory streams. The low probabilityof severalpeople precisely synchronizingtheir stimulus used by Bregmanand Pinkerwas a sine tone alter- attacks. natingwith a two-tone complex. In Figure22, tones A and In light of this work, one might make the following B would representthe sequentialorganization and tones B and predictionsfor the perception of the stimuli used by Bregman C would represent the simultaneous organization. The and Pinker: 1) Sequential streaming is favored by the harmonicity and synchronicity of tones B and C in the frequencyproximity of tones A and B (as we have illustrated complex were variedas was the frequency separationbetween in the earlierexamples). 2) The simultaneous(or timbral) the sine tone A and the upper component B of the complex. fusion of tones B and C is favoredby the synchronyof their The rationalefor varyingthese two parameterswas as follows. attacks. 3) These two effects "compete" for tone B's member- Tones with frequency relationshipsderived from simple ship in their respective perceptualorganizations. 4) Finally, ratios, i.e. those that exhibit "consonance,"should tend to when tone B is "captured"by tone A, it is removedfrom the fuse more readily than combinations considered to be dis- timbral structureand tone C sounds less rich. Thus, it is sonant. While the evidence for this was very weak in the reasonedthat if the two simultaneoussine tones B and C are Bregmanand Pinker study, work currently in progressin perceivedas belonging to separatestreams, they should be Bregman'slaboratory strongly suggests that this is indeed heardas sine tones. But if they are heardas one stream,they the case. The new evidence further suggests that the effect should sound like one rich tone. of harmonicity itself is relativelyweak and may be over- It would be appropriateto introduce the notion of ridden by strongerfactors such as frequency contiguity and "belongingness"at this point, since we talk of tones belonging synchronicity of attack and decay. Tones with synchronous to streams, and of frequency components and the timbre and identically shaped attack and decay ramps are more resultingfrom their interactionbelonging to a perceivedtonal likely to fuse than those with asynchronousor dissimilar event. "Belongingness"(a term used in the perceptuallitera- attacks and decays [12, 15]. This may be a major cue in ture of Gestalt psychology) may be consideredas a principle being able to parse out the different instrumentsplaying of sensory organizationwhich serves to reconstructphysical together in an orchestra since they all have substantially "units"into perceptualevents by groupingsensory attributes

Stimulus

C B LL t 4-

Time-

Percepts

:3 A A Cr B B B B \ ,

C C C C A & B Stream A & B Segregate C Pure C Rich Time - Time -

Figure 22. The competition between sequential and simultaneousorganizations in the formation of auditory streamsis shown here. Tone B can belong either to the sequential organizationwith tone A or to the simultaneousorganization with tone C but not to both at the same time. Bregmanand Pinker (1978) varied the frequency separationsbetween tones A and B and between tones B and C and also variedthe relativesynchrony of onset of tones B and C. The dotted lines in the figure indicate the streampercepts and the verticalsolid lines representthe fusion of tones B and C (cf. Taped Illustration15). Stephen McAdamsand AlbertBregman: Hearing Musical Streams Page39 of thoseevents into unifiedpercepts. As Bregman[2] points A verycompelling demonstration of the decomposition out, "belongingnessis a necessaryoutcome of any process of a timbreorganization by alternatefrequency streaming whichdecomposes mixtures, since any sensoryeffect must be organizationsis illustratedin Figure23. Whentone A is assignedto someparticular source." In this case,when the presentedby itselfit elicitsa timbre,denoted TA. If this tone simultaneoustones B and C are segregated,the timbre is precededby tone B elicitingtimbre TB, and succeeded by resultingfrom their interaction still existsand can be heardif tone C elicitingtimbre TC, one notices that timbreTA one listensfor it, but it is not perceptuallyassigned to either completelydisappears and is replacedby timbresTB and TC. of the tonesB or C, andthus does not affectthe perception Here the highestand lowest componentsof tone A are of them.The nature of a streamis suchthat its qualitiesare streamedwith those of tone B and subsequentlyassume a dueto the perceptualfeatures assigned to it. timbreidentical to thatof tone B. Also,the innercomponents In TapedIllustration 15 one canhear a casein whichA of tone A streamwith those of tone C and assumea like is closeto B in frequencyand C is asynchronouswith B. Then timbre(Taped Illustration 16). a caseis heardwhere A is furtheraway from B in frequency An importantquestion concerning the assignmentof andC is synchronouswith B. TonesB andC havethe same timbreand pitch to a tonal eventarises. Both timbreand frequenciesin both cases.Listen for both the A-B streamand pitch havebeen foundto be context-dependent [10, 21, the richnessof tone C. Thelisteners in Bregmanand Pinker's respectively].But each may be determinedby different studyreported perceiving C as beingricher when B andC were ongoingcontextual-organizations; as such they may be synchronous,and thisjudged richness dropped off with an consideredto be associatedperceptual dimensions of a sound increasein asynchrony,i.e. as C eitherpreceded or followed but maynot be inextricablybound to one another.How rele- B by 29 or 58 msec.As the frequencyseparation between vantto musictheory, then, are studies that deal with the per- A andB wasincreased, C wasreportedly perceived as being ceivedpitch and timbre of tonesin isolation?This question is increasinglyrich. not meantto insinuatethat sensoryand psychophysical exper- imentationare useless. Far from it. If we thinkin termsof The Role of Context in DeterminingTimbre investigatingthe experienceof musicat differentlevels of processing(Figure 24), we see how importantall of these Thesefindings indicate that the perceivedcomplexity areasof researchare in buildingthe wholepicture. The raw of a momentof soundis context-dependent(see [19] as physicalinput is modifiedby the limitsof the senseorgan another of the example trendtoward viewing timbre as whoseoutput is still furthermodified by cognitiveprocesses. on depending context).Context may be suppliedby a number But by studyingsteady-state, or at least relativelysimple of alternativeorganizations that competefor membershipof signals,we can find the limitsand interactionsof the sense not elements yet assigned.Timbre is a perceivedproperty of a organsand peripheral processes, such as and stream ratherthan temporal spectral organization the directresult of a particular resolution,lateral inhibition, and whichlimits affect and is thus In other masking, waveform, context-dependent. words, the final percept.Beyond these, the centralperceptual two whose and frequencycomponents synchronous harmonic processessuch as pitch extraction,timbre buildup, and relationshipswould cause them to fuseunder isolated condi- coherenceand fission the initialneural result of tions be modify may perceivedas separatesine tones if another stimulationof the sensorysystem. Further interactions in organizationpresents stronger evidence that they belongto higherbrain processes such as attentionalprocesses, memory separatesequential streams. andcomparison of pitch,timbre and loudness, context extrac-

One Timbre Two Timbres

:TA versus II: i. 11 :1 12.

A B A C

Figure23. The first part of this figure shows a repeatingtone (A) consistingof 4 harmonics.This tone would elicit a certain timbre percept,TA. In the second part this tone is precededby tone B, consisting of the top and bottom harmonics,and is suc- ceeded by tone C consistingof the two inner harmonics.Tone B elicits timbre TB and tone C elicits timbre TC. However,due to the streamingof tone A's components with those of tones B and C, TA totally disappearsand is replacedby TB and TC (cf. Taped Illustration16).

Page40 ComputerMusic Journal, Box E, MenloPark, CA 94025 Volume3 Number4 sequenceof tones may segregateinto two or more separate streamswhich are individually coherent. In this case,we would Physical Environment Levelsof severalsimultaneous melodies rather than one. Processing perceive If suchan operationaldefinition of melodyis tenable, we mustthen questionhow it is thatsome of the extensionsof elementsother than pitch for "melodic"material are valid Temporaland Spectral Resolution perceptually.For example,when a composeruses timbreas LateralInhibition "Sensory" thematicmaterial, do we stillperceive the sequenceas main- Masking taininga temporalcontinuity? Sometimes coherence is main- etc. tainedby sensitiveperformers and sometimes it canbe very difficultto perceive.Of course,we havenot includedother elementswhich affect the perceptionof melody,such as underlyingharmonic structure, but we areonly attemptingto convey the notion that temporalcoherence should be Pitch Extraction consideredessential to melodyformation. Conversely, one Timbre Buildup "Perceptual" mayuse the principlesof fissionto developrules for creating Coherenceand Fission Limits polyphonyand counterpoint in sequencesof acousticevents. etc. Attentionand Musical Structure

It hasbecome apparent to the firstauthor that musical Attention structure,as it is perceivedin real-time,is inextricablybound Memory to attentionalprocesses. A bit of introspectionwill revealthat thereare at least two kindsof attentionalprocesses, which Context Extraction we mightcall active or willfulattention, and passive or auto- Form and Texture Integration direct attentionto etc. maticattention. One might willfully one's some object or sequenceof events,such as listeningto particularevents within a piece of music.Or, someunusual * eventmight attract one's attention unexpectedly, such as the honkinghorn of an oncomingcar you hadnot noticedas you steppedinto the streetabsorbed deep in thought.That horn demandsyour attentionand in all probabilitygets it in a hurry.In particular,both kinds of attentionmay participate Figure24. This block diagramsuggests a possible arrangement in the processof listeningto auditorystreams. For instance, of the processingof acoustic informationat different inter- in Figure5 whena sequenceof toneslies abovethe temporal connected levels of the auditory system. coherenceboundary, no amountof activeattention can extractthe perceptof one coherentstream. Here perception is limitedby passiveattentional processes [28]. Thishas tion, andform and texture integration, sculpt the transduced importantconsequences for composerswho intendto use fast informationinto meaningfulpercepts. It is felt thatall of these melodicsequences, since it suggeststhat thereare tempi at levelsof processingfeed into each otherin a sort of heter- whichthe listenermay not be ableto followas a melodythe archical(as opposedto a hierarchical)system. The point being sequenceyou haveconstructed, regardless of the attentional madeis that in the frameworkof musicwhere all of these will powerinvoked. An exampleof theseeffects may be found complexinteractions are of greatimportance, the context in the sequencespresented at differentrates in CharlesDodge's thatis createdmay be the essentialdeterminant of the musical Earth's MagneticFields. Withoutpretending to know the resultof a givensound. One sound is potentiallyperceivable in composer'sintent, it can be amusingto listen to the same a greatnumber of ways,depending on its context. sequencedecompose and re-integrateitself duringvarious tempochanges. In a multi-streamedsequence one can relax Melody attentional effort with the result that attention might randomlyalternate among the availablestreams. Or one might Thisleads us to believethat the fundamentalperceptual selectively focus attention on any one of them individually elementin music be may the "melody"rather than the and even play them againstone another. isolatedtone. Orin the terminologyof auditoryperception, Van Noorden reported that the temporal coherence the fundamentalstructure is the auditorystream. This is not, below which all tones may belong of a new but an boundary (the boundary course, notion; empiricalapproach may allow to one stream)is not affected by previousknowledge of the us to clarifyand delimitthe conceptto the extent that we the sequence, and consideredit to be a function of a passive maypredict perceptualresults. That, in the mindof the attentional that the listener "wants to hear first is the mechanism,given author, primaryconcern of the composer.Let us, what between the boundaries examine andits coherence."However, happens then, melody relationto attention. of coherenceand fission a numberof For our we can think of as a temporal dependsupon purposes, melody factors, such as context, and seems to be under the influence connectedand orderedsuccession of tones It [28]. follows of attention. Dowling [17], for example, reported that if thenthat temporalcoherence is necessaryfor a sequenceof listeners knew beforehand melodies were inter- to be whi:h being tones perceivedas a whole. On the otherhand, a leaved, they could, with a bit of practice, extract the appro- . Stephen McAdams nindAlh"rt Rrtnmn- oRa,; ...A-- Page 41 priate melody even at very small separationsof the ranges musicor a musicalcomposition is to take the chanceof traversedby each melody. It is currently assumed that this relegatingone's work to the realmof whatmight be termed ability would degenerate at faster tempi. In addition, "papermusic." van Noorden found an effect at very fast tempi where it was virtuallyimpossible to hear sequenceswithin a rangeof two Acknowledgments or three semitones as other than temporallycoherent. At very fast tempi (about 12.5 tones/ sec.) the tones of such narrow This paperis basedon a workshopentitled "The patternsare not heard as separatemembers of a sequence, but PerceptualFactoring of AcousticSequences into Musical actually mergeinto a continuously ripplingtexture, as one can Streams"delivered at the 1978 InternationalComputer hear in Taped Illustration 17. There are thus attentional MusicConference, Evanston, Illinois, and publishedin the limits in the ability of the auditory system to tracka sequence Proceedingsof the Conference.The authorswould like to of events. When events occur too quickly in succession, the thankDr. Leonvan Noorden for his manyhelpful suggestions system uses the variousorganizational rules discussedin this and valuablecriticisms of the manuscript;Figures 5, 8, 9, article to reorganizethe events into smallergroups. It may 10 and 20 were redrawn,with permission,from then track events within a particulargroup if the listener is Dr. van Noorden'sthesis. The presentversion of this article paying attention to it, but this narrowingof focus necessarily was preparedwhile Mr. McAdams was a GraduateFellow causes a loss of information.One result is the inability to make of the NationalScience Foundation. Dr. Bregman'sresearch fine temporalorder judgments between streams.These organi- has been supportedby grantsfrom the NationalResearch zational mechanismsreflect the tendency of the auditorysys- Councilof Canada,the QuebecMinistry of Education,and tem to simplify things in the face of excessive complexity. In the McGillUniversity Faculty of GraduateStudies and the example where the fast sequence of tones mergesinto a Research.Much of the researchused the facilitiesof the continuous "ripple,"the auditory system is unable to success- Computer-BasedLaboratory of the McGillUniversity fully integrateall of the incominginformation into a temporal Departmentof Psychology.The editorsof ComputerMusic structureand simplifies the situation by interpretingit as Journalwould like to thankMr. McAdams for preparingthe texture (see also [31] ). Thus the auditory system, beyond illustrationsin thisarticle. certaintempi, may interpretthe sequence as a single event and assignto it the texture or timbre createdby its spectraland Appendix1 temporalcharacteristics. An understandingof (or intuition about) these organi- Descriptionof TapedIllustrations zational processescan lead to new dimensionsof control over musical structure.For example, one might construct contra- 1. A repeatingsequence of threehigh tones (1600, 2000, puntal sequences that play across various stream boundaries 2500 Hz.)is interspersedwith threelow tones(350, 430, 550 and through different borderlineregions between temporal Hz.)used by Bregmanand Campbell (1971). a) At a tempoof coherence and fission. (It may be that composerssuch as Bach five tones/sec.the sequenceis perceivedas one streamof were alreadyusing perceptualambiguity consciously in their alternatinghigh and low tones(cf. Figurela). b) At a tempo work.) Any or all of the relevantmusical parameters might be of ten tones/sec.the sequencesegregates perceptually into one used to accomplish this. Then, an appropriateuse of events streamof high tones and one streamof low tones (cf. that vie for or demandthe listener'sattention can be used by Figurelb). the composer to "sculpt" the attentional processes of the 2. Anotherrepeating six-tone sequenceis playedat a listener. Since some events seem more strikingto some persons tempoof ten tones/sec.,but the highertriplet is closerin than to others, this attentional sculpturein time would lead frequencyto the lower.Tone F maybe perceivedas belonging different listeners through different paths of auditory to eitherthe highstream or the low streamdepending on the experience. Further,perception is bound to vary from time to listener'sfocus. Note that tone F cannotbelong to both time within a single person, so the experience would be streamsat once (cf. Figure2). different with each listening. A composition of sufficient, 3. A repeatingsix-tone sequencestarts at a slow tempo. controlled complexity might thus be perceptuallyinfinite for As the tempois graduallyincreased, the sequenceis progres- a given listener. sivelydecomposed into smallerperceptual streams until it is no longerpossible to followthe tonalevents which merge into Conclusion the perceptof timbreor texture(cf Figure3). 4. Usingthe sameinitial sequence as in TapedIllustration An attempt has been made here to point out that 3, the frequencyseparation between temporally adjacent tones composersand music theorists should thoroughlyexamine the is graduallyincreased. The samesort of decompositioninto relationshipbetween the "musical"principles they use and smaller streams occurs. The limits in this example are espouse, and the principles of sensory, perceptual, and determinednot by temporalresolution but by the audible cognitive organizationthat operate in the human auditory frequencyrange (cf Figure3). system. Many of the principlesdiscussed in this article extend 5. The tones of two familiarmelodies are interleavedin the to higher-level perceptual analysis of musical context and same frequencyrange. As the frequencyrange of one melody structureand may well representa scientific counterpartto is shifted away from that of the other, the melodies become some extant music theoretical principles. In other cases, recognizable. though, this group of phenomenasuggests perceptual organi- 6. a) In the first part of this example (see Figure4), tones zations which have little relation to methods currently used A andB, alternatingin one frequencyrange, form a separate to construct or analyze musical structure. To ignore the streamfrom tones X andY, alternatingin anotherfrequency evidence from the real life system in developinga theory of range,b) In the secondpart, tones X andY aremoved into the proximityof the frequenciesof tones A and B with a sub- Page42 ComputerMusic Journal, Box E, MenloPark, CA 94025 Volume3 Number4 sequent perceptualchange. Now tones A and X stream,as do References tones B and Y. Note also that the rhythmsof the respective streamschange between the two examples. 1. Bismarck,G. von. (1974). "Sharpnessas an attribute 7. This example demonstratesthe streamingpotential of of the timbre of steady sounds." Acustica, Vol. 30, a sine wave used for frequency modulation of a sine tone. pp. 159-172. The progressionheard is from a very slow, trackablemodula- 2. Bregman,A. S. (1978a). "Askingthe 'what for' question tion, through a point where the high and low peaks of the in auditory perception," in Kubovy, M. and J. modulated sound segregatedand then to the point where a Pomerantz, eds., Perceptual Organization. Hillsdale, timbre is perceived. The example then returns to the initial N. J.: LawrenceEarlbaum Associates. modulation frequency. 3. Bregman,A. S. (1978b). "Auditory streaming:Competi- 8. Two sine waves are modulated exponentially in tion among alternativeorganizations." Perc. and Psycho- frequency. One ascends in frequency, the other descends in phys., Vol. 23, pp. 391 -398. frequencyand they cross at one point. Most listenersreport 4. Bregman,A. S. (1978c). "Auditory streamingis cumu- hearinghigh and low V-shaped contours as opposed to hearing lative." J. Exp. Psych./HumanPerc. Perf., Vol. 4, pp. the full glissandi. 380-387. 9. First, gated speech is heard. Note the difficulty in 5. Bregman,A. S. (1978d). "The formation of auditory discerningthe message. However,when the gaps are filled streams."in J. Requin,ed., Attention and Performance with white noise, the spoken message sounds continuous VII. Hillsdale, N. J.: LawrenceEarlbaum Associates. and a pulsingnoise is heardin the background. 6. Bregman,A. S. and J. Campbell(1971). "Primaryaudi- 10. The same demonstrationas in Taped Illustration9 is tory streamsegregation and the perceptionof orderin used on musicalmaterial. It is difficult to induce the informa- rapid sequences of tones." J. Exp. Psych, Vol. 89, tion lost in the gaps until those gaps are filled with white pp. 244-249. noise. 7. Bregman,A. S. and G. L. Dannenbring(1973). "The effect 11. The first part of this example is a repeatingfour -tone of continuity on auditory stream segregation."Perc. pattern, all of the tones of which have the same timbre. This andPsychophys., Vol. 13, pp. 308-312. representsa pattern usually perceived as being temporally 8. Bregman,A. S. and G. L. Dannenbring(1977). "Auditory coherent. However,when the third harmonicis added to the continuity and amplitude edges." Can. J. Psych./Rev. second and fourth tones of the pattern, thus changingtheir Can.Psych., Vol. 31, pp. 151-159. timbre, two streamsof differing timbre may be heard. The 9. Bregman,A. S. and G. L. Dannenbring(1978). "Auditory frequencies of the fundamentals are not changed (cf. streaming."Sci. Am., in press. Figure 14). 10. Bregman,A. S. and S. Pinker(1978). "Auditorystreaming 12. Two glissandiwith identical spectralstructures moving and the building of timbre." Can. J. Psych., Vol. 32, in opposite directionsare heard separatelysince none of their pp. 19-31. A. S. spectralcomponents cross in frequency(cf. Figure 16). 11. Bregman, and S. I. Rudnicky (1975). "Auditory 13. Two glissandiwith differingspectral structures moving segregation:stream or streams?"J. Exp. Psych./Human Vol. 263-267. in opposite directions are made to cross each other in Perc. Perf., 1, pp. frequency. Theirdiffering timbres still allow the listener to 12. Bregman,A. S. and H. Steiger(1978). "Auditorystream- segregatethem perceptually(cf Figure 17). ing and verticallocalization: interdependenceof 'what' 14. The same timbres used in Taped Illustration 13 are and 'where'decisions in audition." Unpub. manuscript, arrangedsuch that their pitches cross but none of the McGillUniv. Dept. of Psych., Montreal,Quebec, Canada. frequency components cross. They are still heard as separate 13. Dannenbring,G. L. and A. S. Bregman(1976). "Effect glissandi Figure 18). of silence between tones on auditory stream segrega- (cf. tion." Acous. Soc. Vol. 15. In the first part of this example (see Figure22) tones A J. Am., 59, pp. 987-989. 14. and B are close to each other in frequencywhile tones B and Dannenbring,G.L. and A.S. Bregman(1976)). "Stream C are asynchronous(tone C leads tone B by 29 msec.). Tones segregationand the illusion of overlap."J. Exp. Psych./ A and B form a sequentialstream and tone C is perceivedas HumanPerc. Perf., Vol. 2, pp. 544-555. 15. G. L. and A. S. being fairly pure. In the second part, tone A's frequency is Dannenbring, Bregman(1978). "Stream- fusion of sinusoidal of separatedfrom that of tone B and tones B and C are syn- ing vs. components complex chronousin attack and decay. Tone A forms a streamby itself tones." Perc. and Psychophys., Vol. 24, pp. 369-376. and tones B and C fuse into a single, richerpercept. 16. Deutsch, D. (1975) "Two-channellistening to musical 16. First, a repeatingfour-component tone is heard. Note scales."J. Acous. Soc. Arrt, Vol. 57, pp. 1156-1160. 17. the timbre of this tone (see Figure23). Then this tone A (at Dowling, W. J. (1973). "The perception of interleaved the same repetition rate) is precededby a tone B consisting melodies." Cog. Psych., Vol. 5, pp. 322-337. of the top and bottom components of tone A and is followed 18. Grey, J. M. (1977) "Multidimensionalperceptual scaling by a tone C consistingof the two inner components of tone A. of musicaltimbres." J. Acous. Soc. Am., Vol. 61, pp. 1270-1277. Two streamsare formed and the components of A are split 19. Grey, J. M. (1978), "Timbrediscrimination in musical into separatestreams with tones B and C. Accompanyingthis J. Acous. Soc. Am, Vol. 64, 467-472. streamorganization is a decompositionof tone A's timbreinto pattern." pp. the timbresof tones B and C. 20. Grey, J. M. and J. W. Gordon(1978). "Perceptualeffects of modifications 17. In this example a sequence of tones very close to each spectral on musicaltimbres." J. Acous. Soc. Vol. 1493-1500. other in frequency is slowly acceleratedin tempo until the Am., 63, pp. 21. J. M. and H. M. Nelson in the percept of one continuous, ripplingtone is achieved. Then Guilford, (1936). "Changes of J. the tempo is decelerateduntil the individualtones can be pitch tones when melodies are repeated." Exp. heard once again. Psych., Vol. 19, pp. 193-202. (continued on p. 60) Stephen McAdamsand AlbertBregman: Hearing Musical Streams Page43 (Buxton article continued from p. 25) in a Score Editor," Proceedingsof the 1978 Interna- tional Computer Music Conference, Roads (Comp.), Mathews, M. (1969) The Technology of Computer Music, Evanston: Northwestern University Press, Vol. 2: Cambridge:MIT Press. 392-409. Murray,D., Beauchamp,J., and Loitz, G. (1978) "Usingthe Ritchie, D. M. and ThompsonK. (1974) "The UNIX Time- PLATO/TI980A Music Synthesis System: The SharingSystem," Communicationsof the ACM, 17.7: PLACOMP Language," Proceedings of the 1978 Inter- 365-375. national Computer Music Conference, Roads (Comp.), Smith, L. (1978) MUSIO UserManual, Stanford: work in Evanston: Northwestern University Press, Vol. 1: progress,CCRMA, Dept. of Music, StanfordUniversity. 151-166. Smith, L. (1972) "SCORE-A Musician's Approach to New EnglandDigital Corporation(1978). SynclavierInstruc- ComputerMusic," Journal of the Audio Engineering tion Manual, Norwich, Vermont: New England Digital Society, 20.1: 7-14. Corp., PO Box 305. Tanner,P. (1972) "MUSICOMP,An ExperimentalAid for the Newman, W. and Sproull, R. (1979) Principles of Interactive Composition and Production of Music," ERB-869, Computer Graphics (Second Edition), New York: Ottawa: N.R.C. Radio and Electrical Engineering McGraw-Hill. Division. Patel, S. (1979) "Score Editor Design: The Foundations of Tucker, W. H., Bates, R. H. T., Frykberg,S. D., Howrath, Scriva," Toronto: unpublished manuscript, SSSP/ R. J., Kennedy, W. K., Lamb,M. R. and Vaughan,R. G. CSRG,University of Toronto. (1977) "An InteractiveAid for Musicians,"International Pulfer, J. (1972) MUSICOMP Manual, Ottawa: N.R.C. Radio Journalof Man-Machine Studies, 9: 635-651. and ElectricalEngineering Division. Vercoe, B. (1975) "Man-ComputerInteraction in Creative Reeves, W., Buxton, W., Pike, R., and Baecker, R. (1978) Applications," Cambridge:unpublished manuscript, "Ludwig:an Example of InteractiveComputer Graphics Studio for ExperimentalMusic, M.I.T. (McAdamsand Bregmanarticle continuedfrom p. 43)

22. Halpern, L. D. (1977). "The effect of harmonic ratio U.S. POSTAL SERVICE STATEMENT OF OWNERSHIP, MANAGEMENT AND CIRCULATION relationshipson auditory stream segregation."Unpub. (Required by 39 U.S.C 3685) 1.TITLE OF PUBLICATION A. PUBLICATIONi NO 2. DATE OF undergrad.research report, McGill University Dept. of COMPUTER MUSIC JOURNAL 3 51 6 . 9 0 28 Sep 19779 F.LI. .

Psych., Montreal,Quebec, Canada. quarterly four 4.LOCATION OFKNOWN OFFICe OF PUBLICATION (Street.City, County, State and ZIP Code) (Not printers) _ $. .20.00. 23. Heise, G. A. and G. A. Miller(1951). "An experimental 1263 El Camino Real, Menlo Park, CA 94025 (San Mateo County) 5.LOCATION OFTHE HeAD0UARTeRS ORGeNeRAL BUSINESS OFFICES OFTHE PUBLISHERS (.%ofprinter.; study of auditory patterns."Am. J. Psych., Vol. 64, same 68-77. 6. NJAMESAND COMPLETEADDRESSES OF PUBLISHER, EDITOR ANDMAOAGING EDITOR pp. PUBLISHER (Name and Address) 24. T. evidence for lateral Willard J. Holden, 1263 El Camino Real, Menlo Park, CA 94025 Houtgast, (1972). "Psychophysical EDITOR(N'ame and Address, inhibition in J. Acous. Soc. Vol. Curtis Roads (same) hearing." Am., 51, MANAGINGEDITOR (NOme and Address) pp. 1885-1894. Curtis Roads (same) 7. OWNER (if owned hy a corporation its name and address must be stated and also immediatel? thereunder the namesand addreses of tock 25. McAdamsS. E. effect of tone on holders owning or holding i percent or more of total amount of stokh If not owned bN a corporation, the names and addressesof the individual (1977). "The quality owners must be given If owned by a partnership or other unincorporated firm. its name and address, as well as that of each individual must auditory stream honours thesis, be given.) segregation."Unpub. NAME ADDRESS McGill University Dept. of Psych., Montreal,Quebec, People s Lomputer o. -I263 Ei Lamino Reai, Tenino yark, LAS94U025 Canada.

26. Miller, G.A. and G.A. Heise (1950). "The thrill thresh- . KNOWNBONDHOLDERS, MORTGAGEES AND OTHER SECURITY HOLDERS OWNINGOR HOLDING PERCENTOR MOREOF TOTAL AMOUNT OF BONDS, MORTGAGESOR OTHER SECURITIES(If there are none. so state) old." J. Acous. Soc. Am., Vol. 22, pp. 637-638. NAME ADDRESSS 27. Miller,G. A. and J. C. R. Licklider(1950). "Intelligibility none of interruptedspeech." J. Acous. Soc. Am., Vol. 22, 167-173. 9. FOR COMPLETIONBY NONPROFITORGANIZATIONS AUTHORIZED TO MAIL AT SPECIAL RATES (Section 132.122, PSM) pp. The purpos, function, and nonprofit natus of this organization and the exempt status for Federal income tax purpoeis (Check one) 28. van L. P. A. coherence in Noorden, S. (1975). Temporal AVE0TCHAG2 EDDURING []A CHANGEDhDURINt(It publisher mustsubmit explanation ofchange the tone PRECEDINGMONTHS PRECEDING12MONTHS with hanged,this perception of sequences. Ph.D. Diss., Tech. NO. COPIESOF SINGLEI tAVERAGE COPIES 10. EXTENT AND NATURE OF CIRCULATION ISSUE DURING PRECEDINGstatement.)EACH ACTUALISSUE PUBLISHED NEAREST TO Hogeschool, Eindhoven,The Netherlands,Pub. by The ,2 12 MONTHS NO.FILING DATE ______Institute of 4,0' 840, 3,300 Perception Research, Eindhoven, The B. PAID CIRCULATION Netherlands. NDOR CANDCOIERUNT RE 853 625 29. Ritsma, R. J. (1962). "Existence region of the tonal 2. MAILSUBSCRIPTOONS 1,984 1,800 residue.Part I."J. Acous. Soc. Am., Vol. 34, pp. 1224- . TOTALPAD CIRCULATION (SuM O10oand102) 2,837 2,425 O.FREE DISTRIBUTION BYMAIL. CARRIER OROTHER MEANS 1229. SA E-. COMPLIMEN._TARY,*AND OT4EFREE COPlE 797 100 neo .n 30. Schouten, J. F. (1962). "On the perception of sound and n.TOTAL OI (SumtoC aondDa 3,634 i 2,525 time Fourth Int'l. 1. LEFTOVER. SPOILED 1,051 775 speech: Subjective analysis." Cong. OFCEt USE. UNACCOUNTED, Acous., Copenhagen.Reports II, pp. 201-203. 2. CRETIN CFROM EI AGENTS 155 0 G. TOTAL(Sumn ofE. FI and2--oould equal net press run how n,300 ,840 31. Steck, L. and P. Machotka (1975). "Preferencefor in A)ue 84o 3,300 N musical complexity: Effects of context." J. Exp. Psych./ II.I certify that the statementsmade by met above are correctand complete. RT P

HumanPerc. Perf., Vol. 104, pp. 170-174. 12, FOR COMPLETIONBY PUBLISHERSMAILING AT THE REGULAR RATES (Section l ;.Postal Service Manual) 32. W. R. 39 U. S C. 3626 provides in pertinent part: "No person who wouldhive been enttledto mail matter under former sitlon 4359 of this title Thurlow, (1957). "An auditory figure-ground %hallmallsuch matter atthe rates provided underthissubsectlon unlesshefies annually withthePostal Servicea written requesRfor permlsion effect." Am. Vol. 70, 653-654. to mail matter at such rates," J. Psych., pp. In accordance with the provisions of this statute I hereby requst prm ilsion to mall the publlcatlon named in Item 1 at the phased posts" 33. Verschuure, J. (1978). Auditory excitation patterns: rates prently authorlzed by 39 U. S. C. 3626. / M-94?yl V 1sY =Ir (continued on p. 63)

Page 60 Computer Music Journal, Box E, Menlo Park, CA 94025 Volume 3 Number 4 (McAdamsand Bregman continuedfrom p. 60) The significanceof the pulsation thresholdmethod for the measurementof auditory nonlinearity.Ph.D. Diss., Erasmus University, Rotterdam, The Netherlands. 34. Warren,R. M., C. J. Obusek,and J. M. Ackroff (1972). "Auditory induction: Perceptual synthesis of absent MUSIC-11 sounds."Science, Vol. 176, pp. 1149-1151. Sound 35. Wessel,David L. (1979). "TimbreSpace as a Musical Synthesis System Control Structure."Computer Music Journal, Vol. 3, Runs Under Unix, RT-11 Or RSX-11M No. 2, pp. 45-52. On Any PDP-11 Or LSI-11 With EIS We are licensed to distribute this software system developed by the ComjmirMusic M.I.T. ExperimentalMusic Studio Features: -Very High ExecutionSpeed -Powerful And Intuitive OrchestraLanguage is available in -Dynamic MemoryAllocation -Variety Of Synthesis Methods, IncludingLinear Predictionand Fast Additive Synthesis microform For Detailed Information And Rental Rates Please send me additional information. Call Or Write University Microfilms International 300 North Zeeb Road 18 Bedford Row Digital Music Dept. P.R. Dept. P.R. Ann Arbor,MI 48106 London, WC1R4EJ Systems, Inc. U.S.A. England 306 Court Street Name New York 11231 Institution Brooklyn, Street (212) 643-1077 City State Zip

DMX-1000 SIGNAL PROCESSING COMPUTER 16-bit minicomputer specially For Real-Time All-Digital Sound Synthesis designed for audio signal processing Produces audio output with a self-contained D/A converter be controlled by any ...... Maygeneral-purpose computer Completely programmable fast i0,l:::iii Very Interfaces available for many popular computers

For More Information Call or Write Digital Music Inc. 306 Court Street, Brooklyn,Systems, New York 11231 (212) 643-1077