Scribd Logo
Upload

Welcome to Scribd!

  • Sethares Consonance PDF

    Uploaded by

    Giancarlo Liano
    12 views11 pages

    Original Title

    Sethares consonance.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    12 views11 pages

    Sethares Consonance PDF

    Uploaded by

    Giancarlo Liano

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 11
    Local consonance and the relationship between timbre and scale William A. Sethares Blecrical and Computer Engineering, Unversity of Wisconsin, 1415 Johnson Drive, Madison, Wesconsin $3706 (Received 30 November 1992; accepted for publication 25 May 1993) ‘The principle of local consonance is based on an explicit parametrization of Plomp and Level’s [. Acoust. Soc. Am. 38, 548-560 (1965)] consonance curves. It explains the sclationship between the spectrum of a sound (its timbre) and a tuning (or scale) in which the timbre will appear most consonant. This relationship is defined in terms of the local minima of a family of dissonance curves. For certain timbres with simple spectral configurations, dissonance curves ‘can be completely characterized, and bounds are provided on the number and location of points ‘of local consonance. Computational techniques are presented which answer two complementary questions: Given a timbre, what scale should it be played in? Given a desired scale, how can appropriate timbres be chosen? Several concrete examples are given, including finding scales for snonharmani “compressed” timbres, FM timbres with no timbres for arbitrary sons. PACS numbers: 43.75.Be, 43.66.h INTRODUCTION Carlos observed in Ref. 1 that “the timbre of an in- strument strongly affects what tuning and scale sound best, cn that instrument.” Interpreting “timbre” to mean spec- tral decomposition and “sounds best” wo refer to the tonal consonance theory of Plmp and Levelt,” this paper estab- lishes a conerete relationship between the timbre of a sound and a family of intervals (a scale) in which the sound will appear most consonant, ‘A musical interval is generally considered to be con- sonant if t sounds pleasant or restful a consonant inerval has Tite or no musical tension or tendency to change Dissonance i the degree to which an interval sounds un- pleasant or rough; dissonant intervals generally fe tense and unresolved. The auditory system typically perceives certain intervals as consonant (such as the octave and fifth) while other intervals are. perceived as. dissonant (such asthe minor second or major seventh). Noting that the consonant intervals tend tobe associated with simple ratios of frequency suggests that people might naturally be asad toward auch "sip interval ‘In the previous century, Heimbolt2 suggested thatthe perception of consonance and dissonance could be under- Stood in terms ofthe presence or absence of rapid beating between the sinusoidal components of a complex toe. Plomp and Levelt? coneretized this in a sri of experi ‘ments in which subjects reported the relative consonance of intervals of pure sine waves. The resulting curves exhibit ‘one of the completes expected from studies conducted with musical timbres. Figure 1 shows a family of such curves (which are averaged curves obtained from Ref. 2), cach curve covering a diferent frequency range. All have the same qualitative properties: They begin a unison with high consonance (which appears in Fig. I as zero disso- nance), and rise rapidly to @ maximum dissonance, then slowly crease. This argues against a natural preference timbres (the natural resonances of a uniform beam, teger carrer-to-modulation ratios), and fi “stretched” and for simple rating, since the thirds, fourths, 6fths, and oc taves are undistinguished in these plots. lomp and Levelt explained perception of musical in- tervals by noting that most “musical tones” have a spee- ‘rum consisting of a root or fundamental frequency, and a series of partials that occur at integer multiples of the fun- dlamvental asin Fig, 2. Assuming that the total dissonance between two complex tones i the sum of the dissonances of all the sinusoidal components leads to the curve in Fig. 3 (see Ref. 4 for an alternate method of combining the Sinutaidal dissonance curves). Observe that this curve, which we call the dissonance curve generated by the timbre of Fig. 2, contains major dips at intervals ofa fifth and an ‘octave. It also has local minima at most of the 12-tone ‘equal-tempered scale degrees: The major and minor thirds, the fourth, the major sixth, and the seventh (scale steps 4, 3,9, and 10, respectively). Thus intervals which oocur at Points of local minima in the dissonance curve are per- ceived as relatively consonant. ‘This correlation between consonant intervals and the points of local minima of the dissonance curve suggests two interesting avenues of investigation, Given an arbitrary timbre 7’ (perhaps one whose spectrum des not consist of 2 standard harmonie series), it i straightforward to draw the dissonance curve generated by T. The local minima of this curve occur at values which are good candidates for notes ofa seal, since they are (locally) points of minimum dissonance (or maximum consonance). Alternatively, given a desired scale (perhaps one which divides the octave Into equal pieces, or one Which is not based on the oc- fave), one can find timbres which will generate a disso- nance curve that contains local minima at precisely these scale degrees The relationship between timbre and scale was first explored in Pierce's brief note,’ which reported synthesi2- ing a timbre designed specifically to be played in an 8-tone 1218 J Acoust. Soc. Am. 94 (3), PL 1, Sept 1983 0001-4956/90/04(3/1218/11/S600 © 1999 Acoustical Sonety of Amara 1218 I rd @ oo - Pa ot aa equal-tempered scale. Pieroe concludes, music with tones having accurately specified but nonhar- ‘monic partials, the digital computer can release music from the tyranny of 12 tones without throwing consonance over- board.” Slaymaker* investigated timbres with stretched (and compressed) partials, and Mathews and Pierce? ex- plored their potential musical uses. Geary’s study” em- ployed quantitative methods to demonstrate the existence ‘of consonance and dissonance with nonharmonic timbres. Recently, Mathews and Pierce” examined a scale with steps ‘based on 13 (rather than the standard "2) which is de- signed to be played with tumbres containing only odd par- tials. Carlos! investigated scales for noaharmonic timbres by overlaying their spectra and searching for intervals in which partials coincide, thus minimizing the beats (or roughness) of the sound. This is similar to the present approach, but we provide a systematic techuique tat cat ‘be used to find scales for a given timbre, or to find timbres {or a given scale arene FIG. 2. The standard harmonic timbre wid to generate the disonace carve of Fig 3 Amplities fal a rate of 08 The fropency ast ‘ormalna oat the rot Fegan’ funy 1219. J. Acoust So. Am. Vol 94, No.3, PL 1, Sept. 1999 FIG, 3. Dissonance cure fr the timbre of Fig, 2. The rina i epresentsfequcey dierene. Dts mask he lean of nots ithe Standard [one equattenpered scale. The vrai eatery ard [Ml Sieonance curves te norman! sta the lrgen vale cect at oy To make the discussion more concrete, Sec. I param- etrizes the Plomp-Levelt curves with a model that is used ‘throughout the paper. This model is studied in detail inthe second section, which explores the principle of local con- sonance and derives results about the existence and place- ‘ment of local minima. This lead, in Sec. IIT, to a way of caloulating the best seale to use with an arbitrary timbre. Section IV examines the complementary question: How to construct appropriate timbres given an arbitrary scale. The final section summarizes, discusses the implications of the principle of local consonance, and makes suggestions for further experimentation. A simpler (nonmathematical) ‘view of the timbre scale relationship can be found in Ref. 10. | A QUANTITATIVE MEASURE OF DISSONANCE ‘The Plomp-Levelt curve of Fig. | an be conveniently parametrized by a model of the form dace, o) where x represents the diference in frequency between two sinusoids, and a and b determine the rates at which the function rises and falls. Using a gradient minimization of the squared error between the (averaged) data and the curve d(x) gives values of a=3.5 and b=5.75. ‘The dissonance function d(x) can also be sealed so that the curves for diferent base frequencies and with dit ferent amplitudes are represented conveniently. Ifthe point betvensinosois at frequency / with amplitade 2 and at Frequency f, with ampitde v (or ff) 8 Af fasnry) Hl MP eos @ where sad*/(sfite) @ Witla A, Setnares: Local consonance 1219 op=om, a) ‘The form of Eq, (4) ensures that softer components con- tribute fess tothe total dissonance measure than those with larger amplitudes. For instance, if either o oF vy ap- ‘roaches zero, then vg decreases and the dissonance in Eq. (2) vanishes. Conversely, ifthe amplitudes ofthe partials increase, the dissonance increases. The point of maximum. dissonance d® =0.24 is derived dveotly from the model (1) above. The s parameters in Eq. (3) allow a single func- tional form to interpolate between the various curves of Fig. | by sliding the dissonance curve along the frequency axis so that it begins at f;, and by stretching (or comprest- ing) it 20 that the maximum dissonance oscurs at the ap- propriate frequency. A least-square fit was made to deter- ‘mine the values s,=0.021 and 5,=19. ‘More generally, a (complex) timbre F with base fre- quency fis a collection of n sine waves with frequencies, Lic fye'- , EZ orymep o 2 Consider a fixed timbre F with base frequency J When {v0 notes of this timbre are played simultaneously at an interval with rato , te resulting sound has a dissonance that isthe same as that ofa timbre with frequencies fand af, (wth amplitodes v). Ths Eq. (5) can be used i rectly to caleulate the dissonance between intervals (and chords) as wel asthe dissonance of isolated timbres. De- fining the timbre aF to contain the frequencies (chav avotef) (oth armplitades 0), the dissonance of the timbre Fat an interval cis ret Dart BE dSuaf vin) (0) Finally, the dissonance curve generate by the timbre F defined asthe function Dp(a) overall intervals of interest Dela) Figure 3 isthe dissonance curve generated by the ti bre of Fig. 2 over range slightly larger than an octave, itey for 1$as22. The model predicts that the most eon sonant interval isthe unison, followed closely by the oc- tave. Next isthe fifth followed by the fourth, the major third, the major sixth, and the minor third. These agree with standard musical usage. Looking at the data closely (Gee Table 1) shows that the minima do not occur at ex- actly the scale steps of the 12-tone equa-tempered scale. Rather, they oscur atthe nearby simple ratios 1:1, 2:1, 32, 43, 5, and 5:3, respectively. These are exactly the loca: tions of note inthe “justly intoned” scales." Thus qual tatively. dissonance calculations using harmonic timbres, agree with musical experience and expectations. Te would be naive to suggest that truly musical prop- erties can be measured in terms of a simple consonance. Even in the realm of harmony (and ignoring musically ‘essential aspects such as melody and rhythm), consonance 1220, J. Acoust. Soc. Am, Vol, No 3, PL 1, Sept. 1998, isnot the whole story. Indeed, a harmonic progression that was uniformly consonant would likely be boring. Har- monic interest arses from a complex interplay of diso- nance (restlessness) and consonance (rest). Pethaps the most striking aspect of Fig. 3is that many of the scale steps are coincident with local minima of the dissonance curve. Intervals which occur at points of local ‘minima ofthe dissonance curve are perceived as relatively ‘consonant. This observation forms the basis of the princi- ple of local consonance. A timbre and a scale are said to be related if the timbre generates a dissonance curve whose local minima occur at scale positions. [Note that the shape of the dissonance curve is highly dependent on the frequencies (and amplitudes) of the components of the timbre. Changing these frequencies (and amplitudes) changes the location and depth of the local minima. By the principle of local consonance, this will change the optimum scale on which the timbre should be played. Thus we investigate scales and timbres which are related in the sense that local minima ofthe dissonance ceurve occur at (or very near) scale steps. The next section provides a detailed examination of the dissonance model. I PROPERTIES OF DISSONANCE CURVES For certain simple timbres, dissonance curves ean be ‘completely characterized. This section derives bounds on the number and location of points of local consonance, and reveals some general properties. Two simplifications are sade to streamline the discussion in ths section. A single dissonance function is assumed for all frequencies, and all partials ate presumed to have unit snyplitades. Thus s=1 in Eq, (3) and vy=1 in Eq, (4), simplifying Eq. (5) to Eq, (1) with x=|f2—/|- Taking the derivative of Eq. (1) and setting it to zero shows that the point of maximum dissonance occurs when d*=(1/(a—B)]in(a/6). ‘Theorem 1: For any timbre F with partials at fvfavoafy): (@) a1 is the global minimum of the dissonance Curve Dp(a); (b) lim, .-.Dp(@)=Dp+ Da Proof: All proofs are relegated to the Appendix. art (b) shows that dissonance decreases as the inter- val a grows larger, approaching a value that is no more than the dissonance ofthe timbres Dpand Dag themselves. Part (a) affirms that the unison is the most consonant interval. Since a* is the global minimum of Dp(a), all ‘other minima must be local. The next several results Seek to characterize these local minima. Waar A. Sothares: Local consonance 1220 Theorem 2: Let timbre F have partials at (ff) with Sy~fx>d*. Then the dissonance curve Dy( ct) has local minimum at a*= f4/f, This shows that the dissonance curve generated by a timbre with partials at (/./;) has a point of local conso- nance when a* ,= fz. For example, for the timbre with partials at (500, 750), a®=1.5. The result asserts that the timbre aF, with ftequencies (750, 1128) is a point of local consonance. In symbols, Dy(a*—e)> Dy(a*) and Dp(a®-+e)> Dp(a*) for small e. Thus both (748, 1122) and (752, 1128) are less consonant than (750, 1125). This result is intuitively reasonable since when aff, the dissonance between the partials st af, and fi large, but when af, fa, this term disappears from the consonance ‘measure, Interestingly, the result fails when f, and fare too close. Theorem 3: Let timbre F have partials (ff). Then there is a €>0 such that for [ff] 4d". Then there isa point of (loc) maxi mum dissonance near afi=/y+d", and another near f= f,—d*, Consequently, there must be a minimum for Some a between a= (fi-+dh)/f and ay=(Sy—d")/fy ‘The full rauge of possible dissonance curves for {Wo partial timbres is shown in Fig. 4. Figure 4(b) demon- strates the point of local consonance when a*= f,/, for moderate values of fa—/;. Figure 4(a) illustrates theorem 3, for which ff, 's (00 small to allow a point of local ‘consonance. Figure 4(c) shows the point of loal conso- nance at a*=/,/fy, and also the minimum sandwiched between a, and ay. Such minima are typically broad, and tend to disappear for timbres with more than a few par- tals. The dissonance curve of Fig. 3, for instance, consists sxclusvely of local minima caused by coinciding partials; the broad, in-between minima have vanished ‘The nest result describes points of local consonance for timbres with three partials. ‘Theorem 4: Let timbre F have partials (/;.fs./) Then there are c>0 and ¢>0 such that whenever Fifi ae, and fy— f2>a +e, local minima of the Aissonance curve oveur at a= fy/fy, @)= f/f» and aff Thus timbres with three partials have three points of local consonance at aifi=Sa, arfi=S, and afr=fs, which are represented schematically in Fig. 5. Essentially, ‘4 minimum occurs whenever two of the partials coincide. ‘Of course, other local minima may exist as well. Figure ‘6(a) shows the dissonance curve for the timbre (/BLEAf ). where B = "2. Note thatthe three minima Predicted by theorem 4 are at exactly the fist and fourth scale degres of the ten-ione equal-tempered scale, and also at the “difference” frequency B?/. Figure 6(b) places the partials at (/.B/#°F ), generating the expected scale steps at Land 6, and the diference Bf at step 5. There is also a 1221. J, Acoust Soe. Am. Vol. 94, No.3, Pt. 1, Sep. 1990, :” oo “a o> ie pS i e” * ome FIG. 4, Dissonance cares demonstrating he eal mina of teorems 2 and 3 (a) Fx the timbee (1.017). the pst a to lot allow {point of lea consonance. (b) Point af eel consonance for timbre (E1137) oxeus ata? =15. (e) Poa of ae! comanance or ime (F867 Jat a* = 1.86 and second “oro” minimums da 0 sera of paras. “broad” equilibrium between the third and fourth steps, ‘which isa result of the sparsity of the partials. ‘The final result of this section specifies the maximum ‘number of points of local consonance of timbre ‘Thoorem 5: Let timbre F have partials (f.fyyonf)- ‘Then the dissonance curve generated by F has at most 2n(n—1) local minima. ‘The local minima are symmetrically located (on a log rithmic scale) so that half occur for a between O and 1, and half occur for a between 1 and ao. Half of the local ‘minima are the broad type of Fig. 6(b), while half occur when fi=a/, for some i and j, which’ tend to be steep. Since the musically useful information is located for values Wilam A. Sethares: Local consonance 1221 a oat a5 te oat FIG. 5. Schematic epresetatin of tesla minima of thee parti of «near 1 oF 2, since the broad minima tend to vanish (except for sparse timbres), and since many minima are annihilated when partials are densely packed, (as in Fig. 3), typical dissonance curves exhibit far fewer than 2n( =1) local minima. In Fig. 3, for instance, there are only rine local minima within the octave of interest, consider- ably fewer than the theoretical maximum of 2+7%6 = 84. It's possible to achieve the bound. For instance, the timbre (£2 FF ) over the range 0 Dela) Mya) and De(a—€)> Dela), (11) otherwise, for some small €>0. Thus Ma(q,) is 0 if a, is a point of local consonance and is 1 otherwise. Clearly, J alone ‘would not be an appropriate criterion to minimize since it only reacts to the existence of minima and not to their ‘actual value. The weighted sum of J, and J Wiliam A, ethers: Local contenance 1224, J= min a, D Undaotal | will tend to place the sale steps at local minime aswell as to minimize the value of the dissonance curve. Through trial and error, we have found weightings of Jaf 1000/1 tobe reasonable Minimizing J of Fa, (12) i 4 nimensiona optim zation problem with a highly complex error surface. Gra- dient descent methods are unlikely o prove useful since they search only a small subset of the possible eombina- tions of f,. Random search methods such as simulated annealing or the genetic algorithm are slow, but they are slobal optimizers, atleast asymptotically. We have found that the genetic algorithm (GA) often gives sensible an- swers. See Goldberg" for a complete discussion ofthe al- sorithm, Applications of the GA require thatthe problem be coded: in a finite string called the “gene” and that “ft ness” function be defined. Genes forthe timbre selection problem are formed by concatenating binary represente- tions of the /;, and the fitness function of the gene (PipFvonf) is measuted asthe value J of Eg (12). Thus timbres are judged “more At” Hf the cost J 1 lower. The GA searches n-dimensional space measuring the ness of timbres. The most ft are combined (Via a “mating” pro- cedure) into “child timbres” for the next generation. As generations pass, the algorithm tends o converge, and the most Bt timbre i a good vandidate forthe mize of J Indeed, the GA tends to return timbres which are well satched to the desied scale in the sense that scale steps tend to occur at points of local consonance and the total dissonance at scale steps is low. For example, when the 12 tone equal tempered seule is specified, the GA. con ‘ergs to harmonic timbres about | of the time. This is 200d indication that the algorithm is functioning and that, the fee parameters (the 2's) have been chosen sensibly Drlay+ds SZ Mela) (12) B. Timbres for an arbitrary scale ‘As an example of the application of the genetic algo- rithm to the timbre selection problem, a desired scale was chosen with sale steps at 1, 1.1875, 1.3125, 1.5, 1.8125, and 2. A set of amplitudes were chosen a8 10. 8.8, 7.7, 6. FIG. 10, Dissonance curve for timbe speisd in Se. 1B 1225. J, Acoust Soc. Am. Vol 84, No.3, PL. 1, Sept. 1988 59, $2, 4.6, 40, and the GA was allowed to search forthe ‘most fit timbre. The frequencies were coded as 8-bit binary fmumbers with & bits forthe integer part and 4 bits For the fractional part. The best three timbres out often teil runs (of the algorithm were (/ 1.8125 /,4.9395,14.0725/,9.875f, 14.8125 /,6.425/,12.96/ ), (/.Sf,3.325 10.255 f,7.8125 /,7.0975 3.52 3.875), (f2.39f,9.9295f,7.5625f,114025/;3.99/, 637/,10.6025/ ) ‘The dissonance curve of the best timbre is shown in Fig. 10. Clearly, this timbre is matched to the specified scale, since points of local consonance occur close to the scale steps . Timbres for equal temperaments, For certain scales, such as the m-tone equal-tempered scales, properties of the dissonance curve can be exploited to quickly and easily design timbres, thus bypassing the need for an iterative Solution to the constrained optimiza- tion problem (12). Let = "(2 be the ratio between suocessive scale steps, and conser timbres for which successive partials are 1a. tios of BY for integers fj, j=1,2s... Since B"=2, there is fan integer ky so that 11/2" — Bi for some ky between 0 and m—1. Thus, each partial, when transposed into the same octave asthe fundamental, lies on a note ofthe scale Such a timbre is suid to be induced by the m-tone equal- tempered scale. For example, harmonic timbres are in duced timbres forthe jas intoned scale: Induced timbres are good candidate solutions to the ‘optimization problem. Recall from Sec. that points of local consonance tend tobe located a intervals for which Smaf, where f,and f, are partials ofthe timbre F. Since dts rao betweon any pair of partials in an induced timbre B* for some integer k, the dissonance curve will tend t0 have points of local consonance at such ratios these ratios ‘cur precisely at steps of the scale. Such timbres tend to minimize J of Eq, (12) since the term M(B) is 2er0 whenever scale steps and points of local consonance coin- cide This insight canbe exploited in two ways. First, it can bbe used to reduce the search space of the optimization routine. Instead of searching over all frequencies in a ‘bounded region, the search need only be done over induced timbres. More straightforwardly, the timbre selection problem for equal-tempered scales can be solved by careful choice of induced timbres ‘Asan example, consider the problem of designing tin- ‘bres tobe played in ten-tone equal temperament, Ten-tone ss ollen considered one of the worst temperaments for har monic music, snee the steps ofthe ten-fone sale are dis- tinct from the (small) integer ratios, implying that har- ‘monic timbres are very dissonant when played at intervals (of Bor B = 2. The principle of local consonance asserts ita A, Sotros: Local consonance 1225, FIG. 11. Timbre designed to be played in ten-toneequaltempered sae Note that ait ofl consonance coin with the te-one scl (0p st), bu ot withthe 2one equal enpered sel sep (bottm a) that these intervals will become consonant if played with correctly designed timbres. Here are three timbres induced by the ten-tone equal-empered scale: SELES PLE LOLI ), SEIDEL OLED). SELB LBP IBS) ‘The dissonance curve of the fist timbre is shown in Fig. n V. DISCUSSIONS AND CONCLUSIONS, ‘This paper developed the principle of local consonance and proposed two complementary computational tech- niques: A way to find consonant scales given a specified ibe, aid a way to fad consonant timbres given a spec~ ifled scale. Local consonance is based on an explicit param- etrization of Plomp and Level’s consonance curves. By exploiting this parametrization, several concrete results were derived, which describe in generic terms the number ‘and location of local minima of the consonance curves ‘These results explain certain consonance related phenom- cena and simplify the timbre selection problem. One impli- cation is thatthe musical notion of consonance of intervals such as the octave and fith can be viewed as a result ofthe timbre of the tones we listen to. The justly intoned scales ‘can similarly be viewed as a consequence of the harmonic timbees of musical instruments. ‘The consonance theory of Plomp and Levelt is proba- bly the most important current consonance theory, but its ‘not uncontroversial, In the long run. the importance and. usefulness of the timbre-scale relationship presented here depends on the validity and power of the Plomp-Levelt theory. Conversely, the computational techniques of this paper can be viewed as a way t0 apply the Plomp-Levelt results to nonharmonie music and instrument design. Mu sic composed using related timbres and scales provides in- direct support for the Plomp-Levelt consonance theory. The results of Sec. If show some of the generic prop- erties of dissonance curves. The first theorem says that the ‘unison is always the most consonant interval for any tim- 1226. 4, Acoust. Soc. Am. Vol. 94, No. 8, Pt, Sop. 1888, bre, and that dissonance always decreases as the interval increases, at least for very large intervals. A tuba and a piccolo may play together without fear of dissonance. The ‘ofems 2 and 4 show that local minima often occur at in- tervals which are ratios of the frequencies of the partials. ‘The local minima corresponding to such coinciding par- lias are good candidates for intervals of a sale while the ‘broad minima due to wide separation of partials tend to vanish, at least for complex timbres. The final theorem gives a bound on the number of points of local consonance in terms of the complexity of the timbre, ‘Theorem 3 suggests that local minima of the disso- nance curve are unlikely for intervals larger than about half the interval d* at which maximum dissonance occurs. Plomp and Levelt estimate that d¥ corresponds to a little less than } ofthe eritical bandwidth. Thus theorem 3 pre- dicts that scale steps closer together than about $ of the critical bandwidth should be rare. Plomp and Levelt show ‘that adjacent partials in a Bach sonata fall between critical bandwidth and about | critical bandwidth. Tt would be interesting to conduct studies on the average distance be- tween partials in nonwestern music. ‘The principle of local consonance sufers from the im- itations of the model used vo calculate the dissonance curves. I is only valid for “reasonable” frequencies and amplitudes. Frequencies outside the range of auibility, for instance, eannot contribute to dissonance. Even within the audible range, there is considerable variation in the fre- quency response of the ar, To compensate, the amplitudes of the partials could be weighted by the inverse of the Fletcher-Munson curves," but this has not been pursued here in order to retain the (relative) simplicity of the ‘model. Similarly, che amplitudes must not be oud enough to introduce significant nonlinear distortions in the ea."” ‘The problem of timbre selection for a specified set of scale tones is posed as a multidimensional optimization problem, and can be solved by “random search” methods. Because the optimization must be constrained to avoid trivial solutions, the amplitudes are specified a prion, lim- iting the selection of timbres t0 a choice of frequencies. Different timbres with different mixes of frequencies and ‘amplitudes may solve the optimization problem equally ‘well, implying that they are equivalent in terms of conso- ‘The principle of local consonance makes concrete pre- dictions about the consonance of related timbres and scales, and these must be verified by controlled listener tests. Experiments which would validate (or falsify) the principle might involve fixing some nonharmonic timbre, and finding the most consonant intervals (or scales) from the dissonance curve. Listeners could then judge the rela tive consonance of these intervals (or scales) when intes spersed among a number of other intervals (or scales) Another type of test would be to create a large number of nonharmonic timbres. Listeners could then be asked 10 ‘order the timbres in terms of consonance, and the ordering ‘compared to the predictions of the theory. All of the timbres mentioned in this paper were gen- crated on a personal computer and downloaded to an En- Wilam A Setrares: Local consonance 1228 sonig EPS sampler, where the appropriate scale could be easily defined in a “tuning table." Informal listening tests (using colleagues and friends) upheld the essential predic- tions of the principle of local consonance. One of the mast convincing tests involved the stretched timbres of Sec. IIL B. Two tones were generated which were identical in all parameters except that one used A=2.0 (a normal har- monic timbre) and the other was stretched to A=2.1 [as in Fig. 1(c)]. Two scales were used, a standard 12-tone equal- tempered scale, and a scale stretched so that the pseudo- ‘octave occurred at 2.1. The four possibilities were com- pared: (1) 4=20 timbre with standard seale, (2) A=2.1 timbre with standard scale, (3) A=20 timbre with stretched scale, (4) A=2.1 timbre with stretched scale. AA ist of adjectives was placed before the listeners (who were not told the purpose ofthe test) and they were asked to check those that applied. Overwhelmingly, cases (1) and (4) were described as pleasing, consonant, and musical, while (2) and (3) were described as dissonant, ugly, unmusical, and out-of-tune. Some of the more sophis- ticated listeners (i.e, those with significant musical train- ing) also described (4) as out-of-tune. A similar test was conducted with the timbres of Sec. IV C which were spe- clally designed for ten-tone equal temperament. Again, the ‘elated scales and timbres were judged consonant while the ‘unrelated scales and timbres (such as the ten-tone timbre played in the 12-tone scale and the 12-tone timbre played jn the ten-tone scale) were judged dissonant. ‘Such informal tests cannot be taken too seriously, but they do indicate the need for more rigorous experiments. (One subtlety that arises when dealing with nonharmonic partials is that itis important for the partials to fuse, to be perceived holistically as a single entity rather than as a collection of oddly placed sine waves. Fusing is grestly helped by commonalities in the partials: Similar pitch vi- brato, similar amplitude fluctuations, similar onset time, and similar envelopes. Whatever the experiment, care must be taken that the timbres fuse propery. ‘Another way ta test the predictions of the principle af local consonance is to investigate the relationship between timbre and scale in other musical cultures. For instance, the gamelon orchestras of Indonesia contain ensembles of ‘metalophone instruments of various sizes, materials, and timbres. There are as many different eamelon tunings as there are gamelon orchestras because instruments in the Indonesian musical tradition are not tuned toa single stan dard reference. Rather, each instrument is tuned and tim- brally adjusted to work ina particular orchestral context!" {would be an important test forthe theory to measure the spectra of the instruments of various gamelons and to com- pare the scale predicted by the dissonance curves to the actual scales used. ‘The advent of inexpensive musical synthesizers capa- ble of realizing arbitrary sounds allows exploration of non- hharmonic acoustic spaces, and the principle of local con- sonance provides guidelines on how to sensibly relate scale and timbre. More ambitious, its easy to imagine a whole orchestra of nonharmonic instruments capable of playing consonant music. The computational techniques of this pa 1227 J, Acoust Soe, Am, Vol 94, No.2, PL. , Sept. 1998 per allow specification of timbres and scales for such in- struments. An important area for further investigation is how to engincer acoustic instruments so as to play in spec ified (nonharmonic) timbres and in related tunings, Ult ‘mately, the test of the theory is whether it will be used to make music ACKNOWLEDGMENTS ‘The author would like to thank Tom Staley for exten- sive discussions on tunings, tonality, and timbre. D. Hall, B. Hopkin, W. Strong, and two anonymous reviewers were helpful in commenting on various versions of this paper. APPENDIX Proofs of the theorems follow. Throughout this Ap. pendix, the dissonance curve (6) is writen Dpla) ay where the last two (amplitude) arguments of Eq (2) have been abandoned since v,=1 is assumed for all and s=1 is also assumed. When needed, the values. are used, though generally b> a>0 is enough. Theorem 1: For any timbre F with partials at SPvhevorfa): (a) a*=1 is the global minimum of the dissonance curve D-(a); (b) lim, .. Dp(a)=Dy-+ Da. Proof of (a): For all a, Dp= Dap. The function (1) has its global minimum at x=0 since 6>a>0. Thus BP d(fnaf;)>2Dp whenever aél, and BB nas) =2D- for aml. Consequently, Dg(1)=4Dp < g(a) for all a. Proof of (b): Cleacly, d(x)-+0 as x22, Thus ad(fiaf;)—0 for all i,j a8 Goo, Consequently, Dp(a)~+Dp+ Dap 88 a a ‘Theorem 2: Let timbre F have partials at (ff) with ‘Sy—fy>d*. Then the dissonance curve D(a) has Leal ‘ninimum at a= f/f, Proof: Let inibre G have partials (¢,,g:) =(af,a/2). ‘Then Dp=Do= Dap, and any change in Dj(a) mist arise from the double sum in Eq, (A), which contains the terms d(fig)) for i=12 and j=12. For a*=fy/fy, (21.83)=(Suarfs). As ais perturbed from a, the contr- bation from the term d(fs.g;)=dlSave/:)_ increases, since at at, atf,—f, and 20 d(foay) —d(fasfs ‘Thus the result can be demonstrated by showing that the increase in d(Jy.g) is greater than the decrease in the other three terms combined. The inezease in d( 3) is proportional to 2 (0)=— ae" be (a2) Since fr—fi>d", the decrease in d(Jygi) can be ‘bounded’ by the slope at the point x** on d(x) where (a) as nepal ad of largent absolute vale. Ths curs when d"(a*#) =0, Since endo, x**=2/(a—b)In(a/b). After some simplification, (a3) Willa A. Sethares: Local consanence 1227 aut a/b) 0044-246) =Ka—b). cay ‘Thus the slope at x** is & times the slope at 0. For a=3, b=6, k=L Hence the increase in d(f,g,) is eight times larger than the decrease in d(f,g,)- But |d(f,.gi)| is approximately equal to |d(f2:)|, and is greater than |d(Fus:)|. This proves the desired result a ‘Theorem 3: Let timbre F have partials (/),/3). Then there is a €>0 such that for |/;—fi| <6 the point at=/,/f, is not a local minimum of Dea), Proof: Define G as in theorem 2. Again, any change in De(a) is a result of the four terms in the sum of (Al). For small ¢>0, note that: difygite)> (fig) > dlfigi~ er dfrsrt©) > dfs) > dfi8:~€)s A fpgt©) > d(frgs) >a So8s—€), and dl fogi+e) > di Soa). On the other hand, Ad faygi—€) >d U2) =d(f,f2) =0. For small , the change in all four toms is approximately e(4—a) in magnitude. Thus the dissonance Vale is dered 6 i moved € cher to F and a= f/f; isnot anima, “Theorem & Ler Unbre F have pests fof. ‘Then there are ¢\>0 and ¢,>0 such that whenever fa-fi> dhe) and fy—f,> d¥-+¢, local minima ofthe dissonance curve occur at a@j=fy/fy, m=S/fy anfife Proofi Let Gave partials (¢: 628) =(afafef,). Suppose fst that f2—f,>f.— fiber. ‘Consider the candidate minimum a,. For small, the most significant terms in Dp(a+€)—Dg(a) are d(f)) and d(f;.82), since all others are separated by at least d* +c) For €>0, d(fxgit@>dfrei, d(fybr+e) >d(fy8s), and d(f2.g\—€)>d( 2,81). On the other hand, “difsg)—©) |d(fus—0)|._ Consequently, Dp(ay+€)> Dp(ay) and Dp(a—€) > Dp(cr), showing that a, isa local minimum. The ease fy—f,>fa—fr-+ey, follows identically. The profs for a; and ay are sian ‘Theorem 5: Let timbre F have partials (Fafarenfs) ‘Then the dissonance curve generated by F has at most In(n—1) local minima 1228 J. Acoust S06. Am, VoL 96, No.2, PL. 4, Sept 1999. Proof: Consider the portion of Dye) due to the par- tial f interacting with a fixed partial f,. For small a, af

    You might also like