‘THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA Circularity in Judgments of Relative Pitch VOLUME 36, NUMER 12 Rocex N. SHEPARD Bell Tdephone Leboroories, nc, Maroy Will, New Jersey (Received 23 July 1961) ‘A special at of computr-geaes rated compler tones is shown to lead to a complete breakdown of transitivity ih Jadguentsofratve pitch, Indeed, the tones ean be represented as equally spaced points around a circle Inducks way thatthe clockwise neighbor ofeach toe is judged higher in pitch while the counterclockwise teighbor fs edged lower n piteh, Diametscally opposed tones—though clearly dierent in pitch—are quite Mibiquous as fo the direction of the diflerence, The results demonstrate the operation of a “proximity nce" forthe eatinaum of frequency abd suggest that perceived pitch cannot be adequately repre- ented by a purely reetlinear scale INTRODUCTION ‘HE construction of one-dimensional psychological scales of pitch corresponding to the one-dimen- sional physical scale of frequency for tones has been accomplished by Stevens, Volkman, and Newman' and others. Still others, however, have suggested that the perception of pitch may be too complex to be adequately represented by a single, linear scale. They have pointed out, for example, that the resemblance in pitch between ‘two notes may actually be greater if they are exactly tan octave apart than if they are somewhat less than an. ‘octave apart. Indeed, there is some evidence that even the rat hears an increase in similarity at the interval of an octave.‘ Moreover, I have observed that, at least for harmonically rich tones, a sequence that increases in intervals of a major seventh (ie. intervals that are just one half-tone short of an octave) has a peculiarly #8, 8, Stevess, J. Volkan, and E. B. Newnan, the Measurement” of the Pyychologcal Magnitude of Pitch J Acoust, Soe, Am. 8, 185-190 (1937); see also, SS. Stevens and | Metin the ln of ch rene; & Revd ls” Am J. Psychol. $8, 329-359 (1940) Teg, J, Beck and W. A. Shaw, “The Scaling of Pitch by the Method Magnitude Estimation,” Am. Jo Psychol. 74, 202-251 (19617. "See E. G. Boring, Sensation and Perception i the History of ae Ne 6, 50 Epica 2s particularly pp. fof the Anditory Stimuls; ‘logy, 8-8, Stevens, Ed {981 pp. 988-1039, part ‘CH, Re, Blackwell and H Schlosberg, “Octave Generalization, Pitch’ Discrimination, and Loudness ‘Thresholds in the White Rat," J. Bspil. Prychol. 33, 47-419 (198), “Appleton Century, New York, 1942), CoR. Let Base Correlate: ambiguous character. Although it tends, naturally, tc bbe heard as increasing in pitch, it ean in Some sense be heard, also, a8 a chromatically decreasing. sequence Clearly, such effects are difficult to reconcile with < rectilinear, unidirectional scale of pitch “Asearly.as 1846, Drobisch attempted to accommodate some of these effects by distorting the continuum o: pitch into a helical curve in such a way that tones jus Zn octave apart would be represented by corresponding points on successive turns of the helix* Such a repre entation has the advantage of bringing tones an octave apart into closer spatial proximity. At the same time it permits an explanation of other seemingly anomalou: phenomena by providing for the analysis of pitch inte two distinct dimensions: namely, “height” (or overal pitch level) represented by the vertical axis ofthe helix End “tonality” (“tonal quality” or “tone chroma”) represented by the circular scale atthe base ofthe helix ‘The ambiguous character of the sequence of, note: increasing in major sevenths, for example, can then be explained by noting that, while these notes are indeee {increasing in height, they are simultaneously moving it a contrary direction with respect to tonality (as reflected inthe inverse alphabetical order ofthe letters by which these notes are designated). “The fact that one of the two components of pitch it cixculacin the helical model raises the remarkable poss bility that, by appropriately exaggerating that com ponent (viz., tonality), one might be able to bring about of Tonal Qualities’ * See, A Ruckmick, “A New Classiiat ayebok Rev. 36, 172-180 (1929) 2346CIRCULARITY IN PITCH JUDGMENTS B patown of transitivity in judgments of relative ich ‘In the extreme case, if the dimension of height: Somehow be suppressed altogether, all tones an fr pel Bae oN ronal helix would be collapsed into atonal cle. ig its of relative pitch should then become com- eular in the sense that there would be no 609, orlonest tne in the set but aly an oto Hh ich every tne bas both a docks neighbor gis edged higher in pitch and a counterclockwise 2 pot tha is judged lower i pitch. i experiments desebed her, this cxrows site (ie, wealized by means ofa specially contrived set sie ote tones. The generation of hese tons wae "Paible by a computer program developed by i Rtathews for the synthesis of musical sounds in 8 Nieemely flexible and precisely controlled manne. 1. GENERATION OF THE TONES sch tone consisted of many sinusoidal components ped at sucessive intervals of an octave and sounded taneously. Thus the frequency of each component Me the lowest was exactly twice the frequency of the seveat below. The amplitudes were large for the com freots of intermediate frequency only, however, and erred of gradually to subthreshold levels for the friponents at the highest and lowest extremes of figuency. The sound-pressure level (in decibels) con- feted by each component of one of these, multi- Minponent tones is represented, graphically, by the fGght of the heavy vertical fine corresponding to that Mnponent in Fig. 1. At the beginning of each tone all Guponents were started in the same phase relation yward from a 2er0 crossing). ‘The most important aspect of the scheme adopted for the generation of these tones was that the spectral envelope” of the sound levels, as represented by the tight curve in the Figure, was identical for all tones in the set. Consider, for example, a second tone in which i components are shifted up (in log frequency from the corfesponding components of the first tone) the same fraction of the way toward the next higher octave. The spectral composition of this second tone would difer from that for the original tone in the manner in stich the dashed vertical lines in Fig. 1 differ from the teiginal solid vertical lines. The essential point to notice tere is that the upward shift in frequency has been sffset, in some measure at least, by increasing the con- tributions of the lower components while decreasing the tontributions of the higher components. Indeed, if the second tone is shifted up one whole octave, it becomes identical to the original tone. For, at this point, the highest component (Which has already faded below threshold) is dropped out and a new component (which rll also be below threshold) is introduced one octave below the previously lowest component. The tones generated according to this cyclic scheme } Fo, 1, Smmdpnre rls Go) of 10 saneny sored aasldal components spaced at octave intervals. (The panded seseerrexpond Coan upward sift in the frequencies of ‘i'compaveats) were confined to a discrete set of positions equally paced (in log frequency) between the original tone and the return (one octave “above”) to that same tone. Tn Ah set consisting of fax different tones, the frequency of the cth component of the #th tone is Ue) foine He teats, aw where fain is the frequency of the lowest, component of the frst tone. The sound-pressure level of any one com- ponent (in decibels) is Lniat Lee Lain) [1 eost) 2, (2) where Emin and Lense fix the range of sound levels for the individual components and where the function @ is, defined by 8b) =2xL(c— 1) banat TV Caer coe] 3) Here, caus denotes the total number of components in ‘each tone. ‘The trigonometric “envelope” defined by Eq. 2 (and itlustrated in Fig. 1) was chosen as the most convenient function of the general shape desired that is “smooth” and, at the same time, satisfies the relation FE L(,)=const (over all). o ‘This relation ensured that the different tones in the set did not vary appreciably in over-all loudness. For, if the components of a complex tone are widely sepa- rated (as here), the over-all loudness of the tone is approsimated by the sum of the loudnesses of its separate components.*-? i. Fetcher and W. A. Munsoo, 1 et and Calculation,” J. Acoust. Soe. Am. 8, 82-1 Mea lowes, wie Ludazss of Malicomponent Tones, ip, Poychal 6,190 (950 “towdness Ts Definition, ident tones.‘The waveforms of these specially tailored multi- ‘component tones were readily constructed on an TBM- 7004 computer by means of Mathews’ program for synthesizing quite arbitrarily specified musical sounds.* ‘The only addition that had to be made to Mathews’ program was a short FORTRAN subroutine for the cora- Putation of the parameters of frequency and sound Ievel as prescribed, above, in Eqs. (1)-(3). For each suc- cessive tone, the output consisted of a long sequence of digitally ended numbers on magnetic tape. These num- bers specified displacement as a function of time at a rate of 10.000 samples/sec of subsequent playing time. (Chis sampling rate corresponds to a bandwidth of S000 cps) An analog tape suitable for playback (at 74 ips) ona standard audio tape player was then obtained from the digital tape by off-line equipment for digital- to-analog conversion. “Although each of the complex tones generated in this way was realy made up of many sinusoidal tones, ost Isteners did not hear these sinusoids es subjec- tively separate components—I.., as in a chord. Rather, they tended to describe the total complex as a single tone with sonorous, rather organlike timbre. This is perhaps not surprising, since all higher components 0c- Gurred at harmonic intervals of the lowest audible t. Moreover, the spectral-amplitude envelope odified these components much as a broad-band reso nance would give rise to a formant in the spectrum of & ‘musical instrument. Tn addition to these complex tones, pure sinusoidal tones and bursts of wide-band noise were also generated by means of Mathews’ computer program. The pure tones were used in some initial tests of each subject's ability to discriminate differences in pitch, and the noise bursts were used for certain signaling purposes in some of the experiments described here. The sequences of tones, noise bursts delays, as well as other parameters of the sounds themselves, were communicated to the Computer program by a deck of punched cards (called Y, Mathews, “The Digital Compateras a Musial Instr SRE sr iss) seat, \ OR, N. SHEPARD © the “‘score”). In. ander to reduce. attack and decay “envelopes (lasting abs each) were imposed'at the begining andi OF CIRCULARITY B eee Soa. ‘Owing to the cyclic nature of the complex tong sidered here, they can conveniently be representa ac regulaely spaced points around a circle. The conveggion fallowed i that a clockwise displacement betweer neighboring tones represents an upward shift.imjthe equi ofall coresponding compones Fog the purposes of preliminary tests, the tones were genemted Er the order in which they are encountered in Pasting faround this circle in one direction. Since the; aepe between adjacent tones were produced by halétonc shifts in all corresponding components, the sequencr Fesembled a chromatic scale (of equal temperament) anc 412 steps were required to complete exactly one:-revalu tion about the circle. mt ‘Subjective circularity was clearly demonstrated in. sequence of this kind that made many of these revolu tions, Each tone in this sequence was sounded for 0.1: fer and successive tones were always separated by + ‘O.84 sec period of silence. The components of each ton spanned 10 octaves and the speciic values of the ps Tameters were a8 floWS: taax=12, Cmax 10, fui L863 cps, Lnin=22-4B, and [mas 56 4B. (The over ‘il sound: pressure level of each of the complex tone was about 66 dB, but this measure is meaningful onl, jn a relative sense; the audio gain of the tape playe twas adjusted only to achieve & comfortable listenin; level)* ‘The resulting audio tape, or minor modifications of it have now been played to over 60 listeners and, so far all have described the sequence as progressing mono tonically in pitch (upward or downward, dependin upon whether the tape is played forward or backward) With the forward order, that is, each tone evidently wa ‘The ain Sateen here isin the < dene mein eelanty ot judgments. rather, than iat bien of eaten exposing the ial ren lg tee ya 1 eae af ceelaity an be sccomplshed Ra ment lone, dg nat really depend ope yt the nie my orey ine leet thre ot more components We rae ata with the paral vals choca eben eters (cope since the Howes components TaN the parame ab cea. Thus, the syrametry ofthe distibutien do ey tire of it components with respect, to K Bee eUe hdd at ental srt symmely of the sib eguency Cfgielloudneses However, astrct ammeryEuat eo rc ear for tne phenome and, probably, emo £5 Sa ition that ager of to uber eve Try Sra eens would have dove as well ws the cour curve actly eaployed.