Journal of Experimental Psychology:Human Learning and Memory1980, Vol. 6, No. I, 77-90

The Role of Tone Height, Melodic Contour,and Tone Chroma in Melody Recognition

Dominic W. Massaro, Howard J. Kallman, and Janet L. KellyUniversity of Wisconsin—Madison

The present experiments assessed the contribution of tone height, melodiccontour, and tone chroma to melody recognition. Rather than using highlyfamiliar folk songs as in earlier studies, subjects were taught new melodies.Novel melodies were used to (a) more precisely control potential cues (e.g.,rhythm) that are not of present interest, (b) eliminate unison intervals thatcannot be transformed appropriately, (c) provide a direct analysis of thenature of confusion errors, (d) test whether recently learned melodies arerecognized differently than highly overlearned melodies, and (e) evaluatethe extent to which practice in the experimental task alters the process ofrecognition. The results replicate previous studies using familiar folk songs.Transformations of the original melodies were accurately recognized whentone height was violated, but both melodic contour and tone chroma weremaintained. Violating both tone height and contour while maintaining chromaproduced extremely poor recognition. Performance was intermediate whenjust melodic contour was preserved. There is now good evidence to supportthe idea that melodic contour and tone chroma, in addition to tone height, con-tribute to recognition of both highly familiar and recently learned melodies.

This article addresses the question of thedegree to which three auditory character-istics are functional in melody recognition.The characteristics are tone height, tonechroma, and melodic contour. Tone heightcorresponds to a tone's frequency; forexample, the note A4 has a frequency of440 Hz. Tone chroma corresponds to theposition of a note within the musicaloctave. Notes of the same name coulddiffer in terms of tone height (e.g., 440and 880 Hz) but would be consideredequivalent with regard to chroma (repre-sented by the note name "A"). Finally,contour refers to the up-and-down patternof the successive notes of a melody and,

This research was supported in part by NationalInstitute of Health Grant MH19399. Thanks aredue to Wendy Idson and the reviewers for theirhelpful comments.

Requests for reprints may be sent to Dominic W.Massaro or Howard J. Kallman, Department ofPsychology, University of Wisconsin, Madison,Wisconsin 53706.

consequently, is defined by the directionof change of the tone heights of temporallyadjacent notes.

One method to study this question is tomodify well-known melodies by distortingone or more of these characteristics. Anyinduced changes in recognition performanceshould index the contribution of the cor-responding auditory characteristic. In arecent study, Idson and Massaro (1978)asked subjects to identify highly familiartunes either in their original, undistortedforms or after they had been subjectedto various transformations. In Transforma-tion PC (preserve contour), the melodiccontour was preserved, but the relativesize of each successive interval was notmaintained. Therefore, both tone heightand tone chroma were violated. Trans-formation OPC (octave, preserve contour)resulted in displacement of each note of theoriginal melody by one or more octaves in adirection consistent with the original con-tour. In this transformation, only tone

Copyright 1980 by the American Psychological Association, Inc. 0096-1S1S/80/0601-0077$00.75

77

78 D. MASSARO, H. KALLMAN, AND J. KELLY

height was violated. Transformation OVC(octave, violate contour) caused octave dis-placement of the notes of the originalmelody without regard to contour; thedirection of pitch change for approximatelyhalf of the intervals in transformationOVC was opposite to the contour of theoriginal melody. This transformation pre-served only tone chroma.

If tone height were necessary for ac-curate melody recognition, then perform-ance should be significantly poorer ontransformation OPC, since tone height isviolated. The results showed only a slightdecrement in recognition relative to theoriginal undistorted melody. This resultmeans that contour and chroma are usuallysufficient for correct melody recognition.If both of these characteristics are necessaryfor recognition, then performance on trans-formation OVC should show a decrementbecause, although chroma information ispresent, the melodic contour is violated.Idson and Massaro's (1978) results sup-ported this hypothesis. Recognition per-formance was consistently better on trans-formation OPC than on OVC. In addition,performance was also better on OPC thanon PC, indicating that contour pluschroma rather than contour alone con-tributed to melody recognition. Similarresults for some of these distortions havebeen reported by other investigators(Deutsch, 1972; Dowling, 1972; Dowling& Fujitani, 1971; Dowling & Hollombe,1977; Kallman & Massaro, 1979). Thepurpose of the present study was toreplicate the experimental conditions ofIdson and Massaro using unfamiliar tonesequences as test stimuli to test thegenerality of their results.

There are many good reasons to replicatethe experimental conditions using un-familiar tone sequences. In addition to themore precise experimental control, theuse of artificially designed melodies allowsus to assess whether highly overlearnedand familiar melodies are recognized dif-ferently than are relatively novel tonesequences. One might argue that previousresults are unique to well-known melodiesand, therefore, do not illuminate the

processing of most tone sequences. Forexample, Attneave and Olson (1971) foundthat subjects could transpose across octavesonly very well-known sequences such as theNBC chimes. Extending this finding to thepresent task of recognizing melodies, itmight be predicted that tone chroma willnot be as functional in the recognition ofunfamiliar melodies as it is in the recogni-tion of well-known melodies. By testingobservers for a number of consecutivedays, it becomes possible to evaluate theextent to which practice in the experimentaltask alters the process of melody recog-nition.

Most familiar melodies contain unisonintervals that are impossible to transformappropriately. For example, Transforma-tion OPC (octave displacement, preservecontour) cannot be applied to the unisoninterval because displacing the notes anoctave apart will also change the contourfrom level to ascending or descending. Inprevious studies, unison intervals wereleft intact in Transformation OPC, and thiscould have artificially inflated performanceon this transformation. The unfamiliarmelodies used in the present experimentsdid not contain any unison intervals.

Another potential problem in the useof familiar melodies is that it is not clearwhether to eliminate rhythm informationentirely, as in the Idson and Massaro(1978) studies, or to include it, as in theDowling and Hollombe (1977) and Kall-man and Massaro (1979) studies. The useof unfamiliar melodies that do not differrhythmically provides a straightforwardsolution to this problem.

The present study extends the body ofknowledge on melody recognition by ana-lyzing the nature of confusion errors be-tween melodies. If a distortion of MelodyA is usually called Melody B, then it ispossible to determine what propertiesMelody B has in common with the dis-tortion of Melody A. As an example, thedistortion of Melody A may produce acontour that is similar to Melody B, andit can be expected to be identified as suchto the extent that contour is functionalin recognition.

MELODY RECOGNITION 79

To study the recognition of recentlylearned melodies, it is necessary to use aforced-choice procedure with a limited setof stimulus-response alternatives. Thisprocedure is identical to that used byIdson and Massaro (1978) but differsfrom the procedure of Deutsch (1972) andKallman and Massaro (1979). The lattertwo studies presented a given melody ortransformed melody only once, and subjectswere not given any specific response alter-natives. Both procedures have providedconverging results: Performance is verypoor if chroma is preserved and contouris violated (OVC), and preserving bothchroma and contour (OPC) is significantlybetter than simply preserving contour(PC). Given that the results of interestare the same with both procedures, itcannot be argued that the results foundusing the forced-choice procedure are theproduct of recognition strategies uniqueto this procedure as Deutsch (1978) hasargued.

Experiment 1

Method

Subjects. The subjects were 11 University ofWisconsin students who volunteered to participatein the study to earn extra credit in an introductorypsychology course. Subjects were not aware of thenature of the experiment at the time they volun-teered, and musical experience was not a factorin their selection. All subjects reported havingnormal hearing.

Stimuli. Four six-tone sequences were chosenfor use as original untransformed stimuli. Threesequences (1-3) were selected from among exercisesin an introductory sight-singing text; the fourthwas constructed by an experimenter. Each six-tonesequence began and ended on €4 (262 Hz); alloriginal melodies (we use the terms tone sequencesand melodies interchangeably) were in the key of Cmajor. The original untransformed tone sequencesare referred to as Transformation O.

The notes comprising the four original tonesequences as well as the transformations on them(described later) appear in Table 1. The rangespanned by each untransformed sequence, that is,the distance between the lowest and highest notesin the sequence, was less than one octave.

The transformations used were similar to thoseused by Idson and Massaro (1978). Like the originalmelodies, each transformed sequence began andended on C4 (262 Hz).

1. Preserving contour (PC): Transformation PCpreserved the contour of the original sequencewithout regard to maintaining either absolute pitchinformation or the relative size of successive inter-vals from the original sequence. Each note of theoriginal was displaced within the octave either upor down a small number of semitones. The totalrange in semitones of each PC transformation wascomparable to the range for the original sequence.

2. Linear transformation (LT): TransformationLT was constructed by dividing the number of semi-tones between each directionally denned intervalin the original sequence by two. For example, thefirst interval in Melody 1 is a two-semitone jumpfrom C4 (262 Hz) to D4 (294 Hz); in transformationLT this became a one-semitone rise from €4 (262Hz) to C*4 (277 Hz). In some cases, use of thisalgorithm resulted in inclusion of tones not foundin the chromatic scale. For example, the originalSequence 1 contains a seven-semitone fall in pitchfrom D4 (294 Hz) to G3 (196 Hz) between Notes4 and 5 of the sequence. In transformation LTthis becomes a 3.5-semitone decrease from C*4(277 Hz) to the tone between A3 and A*3 (227 Hz).In such cases the frequencies of the tones wereplaced halfway between the frequencies of the twoadjacent semitones.

Transformation LT had the following effectswith regard to each original sequence: (a) It pre-served contour; (b) it preserved the relative sizeof the original intervals; that is, the ratios of thedistances between successive tone pairs remainedthe same as in the original; (c) it destroyed absolutepitch information; and (d) it reduced the totalrange of the original sequence by half.

3. Octave-preserving contour (OPC): Trans-formation OPC preserved the chroma informationand the contour of each original sequence. Eachnote of the original melody was displaced up ordown, depending on the direction of pitch change,by one or more octaves. For example, the first threenotes of original Sequence 1 (C4, 262 Hz; D4, 294Hz; Ei, 330 Hz) became (C4, 262 Hz; D5, 587 Hz;and E8, 1319 Hz). An effort was made not to dis-place two successive notes into the same octave;however, because it had been decided not to includetones below 110 Hz or above 4000 Hz in any se-quence, this was not always possible.

Transformation OPC preserved both the contourand chroma information of the original melodies.The sizes of successive intervals in the originalmelodies were changed. In addition, the absolutepitch information from the original was partiallydestroyed. However, each OPC sequence containedat least one tone (in addition to Tones 1 and 6) thatwas identical in frequency to the tone in the sameposition in the original sequence. In Sequences 1and 3, the identical note in the original and trans-formation OPC was in the fifth position; thus thesetwo sequences contained not only an extra identicaltone but also one identical interval. Comparing per-formance on these two sequences to performanceon the other two sequences will allow a direct assess-ment of this property.

80 D. MASSARO, H. KALLMAN, AND J. KELLY

Table 1Notes and Note Frequencies of Tone Sequences Used in Experiment 1

Transformation Frequencies Notes0 Contour1"

Sequence 1

O 262, 294, 330, 294, 196, 262PC 262, 349, 392, 330, 220, 262LT 262,277, 294, 277, 227,262OPC 262, 587, 1319, 587, 196, 262OVC 262, 147, 659, 1175, 392, 262

C4 D4 E4 D4 Gs C4C4 F4 G4 E4 A3 C4C 4 C* 4 D 4 C* 4 * C4C4 D6 E6 D6 G3 C4C4 D3 E6 D6 G4 C4

Sequence 2

OPCLTOPCOVC

262,262,262,262,262,

294,330,277,587,587,

196,247,227,196,784,

294,349,277,587,294,

330,392,294,

1319,165,

262262262262262

C 4D 4C 4 E 4C4C*4C 4 D 6C 4D 6

G3Ba*G3G6

D4 E4F4 G4C*4D4D6 E6D4 E3

C4C4C4C4C4

-) 1- -1 —-) 1- -1 —H h H —+ — 1 — 1 —+ i iT^ T

Sequence 3

OPCLTOPCOVC

262,262,262,262,262

247,220,254,123,494

220,196,240,110,880

330,349,294,659,

2637,

294,330,277,294,S87

262262262262262

C 4B 3C4A3C 4 *C 4 B 2C 4 B 4

A3G3

*A2A6

E4F4D4E6E7

D4 C4

E4 C4

C*4C4D4 C4D6 C4

1 __ 1 _

1_ 1_i_ _i_ _i_ —

Sequence 4

OPCLTOPCOVC

262,262,262,262,262,

440,494,339,880,220,

392,349,320,392,784,

494,440,359,

1976,494,

392,330,320,784,196,

262262262262262

C 4 A 4C 4 B 4C 4 »C 4A 6C4A3

G4 B4F4 A4* *G 4 B 6G 6 B 4

G4E4*G6Ga

C4C4C4C4C4

H 1 — —-1 j_j- _ _j_ _ _

H 1_i_ i~r ~

Note. 0 = octave; PC = preserving contour; LT = linear transformation; OPC = octave-preserving con-tour; OVC = octave-violating contour.a Asterisks denote tones with frequencies that do not correspond to notes in the chromatic scale.b A minus sign denotes a descending interval; a plus sign denotes an ascending interval.

The range in semitones of each sequence undertransformation OPC was much larger than in theoriginal sequence, on the average of 33 semitones.

4. Octave-violating contour (OVC): Transforma-tion OVC was effected by displacing each note of anoriginal sequence by one or more octaves, dis-regarding the direction of the interval change in theoriginal melody. Transformation OVC preserved thechroma information but violated the contour of theoriginal melody. The number of contour violationspossible in a six-tone sequence is five; Transforma-tion OVC of Sequences 1 and 3 each had threeviolations of original contour, whereas the OVCversions of Sequences 2 and 4 each had four contourviolations.

Transformation OVC distorted the absolute pitchand successive interval size information of theoriginal sequences. Sequences 2 and 4 each had onetone in Transformation OVC that was identicalto the corresponding original tone. None of themelodies had any identical intervals in the originaland OVC versions. The range of notes used in theOVC transformations averaged 32 semitones.

Since the range of tone frequencies used in trans-formations OPC and OVC was much greater thanfor the original, PC, and LT versions of each se-quence, the latter three tone patterns were displacedas a whole into each of five octaves (one octavebelow and three above the octave beginning at C4).This was done to assure that any differences foundbetween the transformation conditions was not dueto differential discrimination at different frequencies.The octave in which the original sequence, Trans-formation PC, and Transformation LT were pre-sented was chosen randomly on each trial.

Apparatus. All experimental events and datacollection were controlled by a PDP-8L computer.The sine-wave tones were generated by a Wavetek(Model 155) digitally-controlled oscillator. Theoutput of the oscillator was gated by an Iconic(Model 0137) audio switch and then amplifiedthrough a Mclntosh (Model MC-50) amplifier.Rise and fall times for the tones were 10 msec. Thetones were diotically presented to subjects overGrason-Stadler TDH-49 headphones at an am-plitude level of approximately 80 dB (A). Since the

MELODY RECOGNITION 81

Wavetek oscillator could be adjusted to only threedecimal places, the last digit of any frequency over1,000 was truncated. The visual displays werepresented over an array of light-emitting diodes(Monsanto Model MDA-III). Four or fewer sub-jects could be tested simultaneously in separatesound-attenuated rooms.

Design, There were three major independentvariables: melody, transformation, and octave ofpresentation. In addition to the original melody,there was a total of four transformations on eachmelody. The original melody and TransformationsPC and LT were subject to presentation in any offive octaves; the first note could range from Cs(131 Hz) to C7 (2,093 Hz), and the frequencies ofthe following notes were adjusted accordingly. ForTransformations OPC and OVC, octave was adummy variable; the first note of these transformedmelodies was always C4.

Procedure. During the experiment, on each triala six-tone sequence was presented and then fol-lowed by a response period. An asterisk presentedfor 250 msec over a visual display signaled thebeginning of each trial. The first tone of the sequencewas presented 750 msec after termination of thevisual display. Including the rise and fall times,each tone had a duration of 150 msec. There was a150-msec silent interval between tones. Immediatelyfollowing the offset of the sixth tone, subjects weregiven 5 sec to indicate by pressing one of fourbuttons whether the tone sequence sounded mostsimilar to Sequence 1, 2, 3, or 4. On those trialsin which one of the four original (untransformed)sequences had been presented, visual feedbackwas provided indicating whether the sequence wasSequence 1, 2, 3, or 4. This feedback, which lasted250 msec and immediately followed the responseperiod, was designed to aid learning and retentionof the four untransformed sequences. There was a1-sec interval between trials.

The experiment was conducted on 4 consecutivedays; each session lasted approximately 45 min.On the 1st day of the experiment (practice day),subjects were first introduced to the four untrans-formed sequences. These were presented in the orderSequence 4, 3, 2, 1. The four sequences were pre-sented in this order six times. The temporal con-straints governing presentations of the sequenceswere the same as on experimental trials exceptthat 3 sec separated the presentation of eachsequence. The numerical label assigned to eachsequence was presented over the visual displayduring each tone sequence presentation. The sub-ject's task during this phase was simply to listenand learn to recognize the different tone sequences.Following the initial presentation of the tone se-quences, subjects were presented 112 practice trials.These trials were exactly like experimental trialsexcept that only the four original sequences in theiroriginal octave (starting on C4) were presented.The subjects' task was to indicate which sequencewas presented, and visual feedback was given oneach trial.

After a 5-min break, a block of 112 experimentaltrials was presented. The first 12 trials were practiceand were not recorded (although the subject didnot know this); random sampling without replace-ment began with Trial 13.

On experimental Days 1, 2, and 3, subjects'memories were refreshed by playing the untrans-formed tone sequences six times each (as on thepractice day). This was followed by two blocks of112 experimental trials separated by a short break.Given that the first 12 trials of each block were notrecorded, one block involved a complete replicationof the 4 melodies X 5 transformations X 5 octaves= 100 conditions of interest. Within each experi-mental block each experimental condition occurredrandomly without replacement. Two blocks per daygave two observations per condition per day foreach of the 3 experimental days.

Results and Discussion

Only data from subjects who correctlyidentified the original untransformedmelodies on at least 50% of the trials overthe 3 experimental days were includedin data analysis. Of the 11 subjects, datafrom 8 met this criterion. An analysis ofvariance (ANOVA) was performed on thepercentages of correct identifications. Tonesequence (four levels), transformation (fivelevels), octave of presentation (five levels),and day (three levels) were all within-subjects variables.1

The percentages of correct identificationsof each transformation of each sequenceon the 3 experimental days appear in Table2 along with the data averaged over the3 days. The main effect of transformationwas significant, F(4, 28) = 44.36, MS,= 1.041, and thus orthogonal comparisonswere conducted to test differences betweenthe means. The absence of a significantdifference between the O and OPC condi-tions (F < 1) replicates the findings ofIdson and Massaro (1978). One factorthat might bias the results on the OPCtrials toward good performance is thefact that two of these sequences (Se-quences 1 and 3) included one melodicinterval identical to the correspondinginterval in the original sequences. However,examination of the data shows that per-

1 Unless otherwise noted, the claim of a significanteffect will denote a p value of less than .05.

82 D. MASSARO, H. KALLMAN, AND J. KELLY

^%}

s•-s>fc$•S

i3a

~s

"3S

8

^-«

t

|

'-S

11OG•fc.t^tv,

"KM

aj

ll[> 2® IH

QJ *fflbfi g2 ^J) £"

^O

UOHO

J

OH

0

c.9

Zgrt Q

gn)|H

U

, J

OHO

_J

uOH

O

C

•Bcs 2>, 6a g

1

U

uOHO

H

uOH

O

G

•a^H g

ts S

C

•ti

O

O

f J

O

H

(jft^

O

"O

1

w> cs uo es oCS f-t rH

00 CS "0 O -*t~ 00 00 OS 00

Ov ^O 00 -^ VO

2 K^ vO^O

ON CO CO CO CO

T-H

CO CO NO -^ ONr* CO ON ON CO

T-H ON O CO CSON t^- CO co t .

IONOIOONOj>. 10 \O *O O

ON lO ON iO <NON ON CO CO ON

CS *-* +4

00 CO CO ON CO

lO O i-i CSON 10 cO 10 t^

*^J fO O^ CO *—*cO 10 o t . t .

o c o c o S g

r-i cs *-*

vO O^ CO ,-H SOt » *O t » 00 l""*

ON PO CO CO ^£

CO ON CO •«* ONl>» -^ lO lO "

ON O -* -^ CSO co r- r- co

Ig• H CS CO ••*

sOcOO

a

1£I%IIuO•-

1IO

c*s

i4>a

1cti

11uOHO

B.0

K

1

1

fe

C

IIH

u

18bfl.s

aIIUOH

(L)

ayII

O

^

^

formance on the OPC transformationrelative to the original sequence was nobetter for Sequences 1 and 3 than for Se-quences 2 and 4. Thus the good perform-ance on Transformation OPC cannot beexplained by the fact that one interval intwo of the melodies was identical underthis transformation to an interval in theoriginal melodies.

No difference was found between the twoconditions that maintained the contourof the original sequence but violated tonechroma (i.e., PC and LT, F < 1). However,although performance was quite good underthese contour-preserving conditions, thethird orthogonal comparison indicates thatwhen correct chroma information wasavailable as well as contour information(O and OPC), performance was approxi-mately 20% better than when only con-tour information was correct (PC and LT),F(l, 28) = 15.87. Finally, when contourinformation was incorrect but accuratechroma information was maintained(OVC), performance was much worsethan on the contour-preserving transfor-mations, ^(1, 28) = 161.05. This latterfinding is in agreement with the resultsof Idson and Massaro (1978) and suggeststhat accurate tone chroma is only usefulfor melody recognition when it is accom-panied by accurate contour information.

Two of the OVC transformations (Se-quences 2 and 4) contained one note (inaddition to the first and last note of thesequence) that was identical to the cor-responding note in the original sequence.It is noteworthy that performance on thesetwo OVC sequences was higher than onthe other two sequences (22% and 12%correct vs. 3% and 5% correct). Thisfinding could be interpreted as suggestingthat the presence of an identical note in theoriginal and transformed sequences re-sulted in a performance increment. Ex-tending this line of reasoning, good per-formance on OPC trials might be due tothe fact that all OPC transformations had anote (other than the first or last) identicalto the corresponding note in the originalsequence. However, such a conclusion isnot warranted by the data. Both the

MELODY RECOGNITION 83

OPC and OVC transformations of Se-quences 2 and 4 contained a note identicalto a corresponding note in the originaltone sequence, and yet performance on theOPC transformations of these sequenceswas substantially better than on the OVCversions of the same sequences. The higherperformance found on the OPC trials ofSequences 2 and 4 clearly cannot be ex-plained by the presence of notes identicalto corresponding notes in the original se-quences, since the OVC transformationshad an equal number of identical cor-responding notes.

There was an overall improvement ofapproximately 7% in performance fromexperimental Day 1 to Day 2, as is reflectedby a significant day effect, F(2, 14) = 9.43,MS,, = .118. Furthermore, there was asignificant interaction between day andtransformation, F(S, 56) = 2.49, MSe =.118. An examination of Table 2 reveals thatperformance on the four contour-pre-serving transformations improved overdays, but performance on TransformationOVC actually deteriorated. It will beargued later that to the degree that subjectsused melodic contour information in mak-ing their decisions about OVC transforma-tions, performance should have been pooron these transformations. Given this inter-pretation, the form of the interactionbetween days and transformations suggeststhat the subjects continued to learn thecontours of the melodies and used thisinformation to identify the melodies.

Although melody was not a significantfactor, F(3, 21) = 2.22, MS, = .690, therewas a significant interaction between trans-formation and melody F(12, 84) = 3.73,MSe — .493. This interaction may beexplained, at least in part, by noting thatfor sequences that yielded relatively poorperformance on the OVC transformations(Sequences 1 and 3), performance on thecontour-preserving, chroma-violating trans-formations (Transformations PC and LT)was relatively good compared to thosemelodies yielding better performance onthe OVC transformations. It is interestingto note that the melodic contours of theuntransformed Sequences 1 and 3 appear

more salient than do the contours of theuntransformed Sequences 2 and 4 (seeTable 1). Sequence 1 could be coded as"two ups, two downs, and one up," whereasSequence 3 consists of "two downs, one up,and two downs." The contours of originalSequences 2 and 4 appear more difficultto code. These possible coding differencescould have accounted for differences in thedegree to which contour information wasused to identify the different sequences.If contour information was weighted moreheavily in identifying Sequences 1 and 3than in judging the other sequences, wewould have expected both the relativelylow performance on the OVC transforma-tions of Sequences 1 and 3 and the elevatedperformance on the PC and LT transforma-tions of these sequences.

Octave of presentation was not a signifi-cant variable (F < 1, MS, = .051), al-though octave did interact with transforma-tion, F(16, 112) = 1.81, MSe = .045. How-ever, no systematic trend underlying thisinteraction was apparent, and the statis-tical significance probably reflects primarilythe high power of the test given the largenumber of degrees of freedom.

Performance on Transformation OVCwas not only much worse than for any ofthe other transformations, it was wellbelow the 25% correct expected by chance.To understand this result, consider thatthe relatively good performance withTransformations PC and LT indicates thatcontour information played a substantialrole in subjects' identifications. To theextent that the OVC transformation of atone sequence results in a sequence whosecontour resembles that of an untransformedsequence other than that from which theOVC sequence was derived, confusionsbetween the OVC sequence and the Osequence with similar contour would beexpected. The response confusion matrix forthe OVC transformations presented inTable 3 supports this analysis (for refer-ence refer also to the contours of the tonesequences presented in Table 1). Forexample, the OVC transformation of Se-quence 2 was identified as Sequence 1 69%of the time. This high percentage of mis-

84 D. MASSARO, H. KALLMAN, AND J. KELLY

Table 3Percentage of Each of the Four TransformedSequences Identified as Each of the FourOriginal Sequences for the Five Transformationsin Experiment 1

Response*

Melody

Transformation OVC

1234

3 (2)69 (S)6(3)

54(4)

40 (2)22 (1)24(3)22 (0)

11 (4)5 (1)5 (3)

12 (2)

46 (3)3 (2)

65 (4)12 (1)

Transformations OPC, PC, O, and LT

1 87 (5) 7 (1) 5 (1) 1 (2)9(1)4(1)6(2)

71 (3)11 (3)19 (4)

4 (3) 17 (4)78 (5) 8 (4)4 (4) 70 (S)

Note. O «= octave; OPC = octave-preserving con-tour; OVC = octave-violating contour; PC = pre-serving contour; LT = linear transformation." Scores in parentheses are the number of successiveintervals shared by the transformation and the origi-nal tone sequence.

identifications occurred because the con-tour of the OVC transformation of Se-quence 2 was identical to that of theuntransformed (0) Sequence 1. Similarly,the reason why the OVC transformationof Sequence 2 was identified as Sequence2 on only 22% of the trials was probablybecause its contour differs from the un-transformed (O) Sequence 2 at four of thefive intervals.

One way to test this interpretation moredirectly is to calculate, for all combinationsof OVC transformations and original se-quences, the number of correspondingsuccessive intervals that have the samedirection of pitch change. We wouldexpect that the probability of confusingthe OVC transformation of Tone SequenceX for the untransformed Sequence Yshould increase as the number of cor-responding intervals between the sequenceshaving the same direction of pitch changeincreases.

For each combination of O and OVCsequences used in Experiment 1, thenumber of successive intervals having the

same direction of pitch change appearsin Table 3. Although the fit is not perfect,the data generally are in accord with ourinterpretation. Disregarding correct re-sponses (i.e., a response of X to OVCSequence X), a test of correlation was per-formed between the number of common(in terms of ups and downs) successiveintervals between the OVC and O stimuliand the number of times the OVC stimuliwere identified as the incorrect O sequences.This correlation was significant, r(10)= .60, supporting our analysis.

It is also possible to examine the degreeto which the response confusions on trialsother than those in which OVC transforma-tions were presented depended on thedegree of contour similarity between thepresented melody and the response melody.To compute a correlation coefficient, theconfusion data were averaged over Trans-formations O, OPC, LT, and PC; theseaveraged data are given in Table 3. Thecorrelation coefficient was computed usingthe same procedure as for the OVC con-fusion data; correct responses were dis-regarded. Although there was a relativelyhigh correlation between the number ofcommon directions of the successive inter-vals and the number of response confusionsfor each tone sequence pair, the correlationwas only marginally significant on a one-tailed test, r(10) = .49, .05 < p < .10.Nonetheless, the presence of a relativelyhigh correlation is consistent with the viewthat contour information did play animportant role in subjects' judgments.

Two general points can be made aboutExperiment 1. First, melodic contourplayed a substantial role in subjects'judgments. Second, the elevated perform-ance on Transformation OPC relative toTransformations LT and PC suggeststhat in addition to melodic contour, tonechroma played a role in the judgments.These findings are consistent with thefindings of Idson and Massaro (1978).

Experiment 2

In Experiment 1, performance on OPCtrials was about as good as when the

MELODY RECOGNITION 85

original melodies were presented. Wetentatively interpreted this result as indi-cating that tone chroma was a functionaldimension in melody recognition. Thisconclusion followed because when onlycontour information was available (PCand LT transformations), performancesuffered relative to the OPC condition.Presumably, the use of chroma informationin addition to contour information resultedin the increment in performance found onOPC trials.

An alternative explanation of the goodperformance found on OPC trials is pos-sible. Since the intervals were on theaverage much larger under the OPCtransformations than under the othertransformations, the contour of the OPCmelodies may have been more salient (cf.Idson & Massaro, 1978). Since the resultsof Experiment 1 indicated that contourwas an important cue used by subjectsin making their decisions, the addedsaliency of the contour on OPC trials mayhave been the sole cause of the good per-formance found on these trials. If this werethe case, it would follow that chromainformation played no role in the OPCdecisions.

To test this idea we used a transforma-tion, PCS (preserve contour, stretched),in Experiment 2 that preserved the contourof the original melody and had intervalsapproximately as large as those used underTransformation OPC; however, correctchroma information was not availableunder Transformation PCS. Using a similartransformation, Idson and Massaro (1978)found that in their experiments, the possibleadded saliency of the OPC contour couldnot account for the good performance onOPC trials. However, given the prominentrole of contour found in Experiment 1of the present study, it is important todetermine whether the stretched contourhypothesis could account for the goodperformance on OPC trials. If the goodperformance on transformation OPC wasdue solely to the added saliency of thecontour under this transformation, thenequally good performance should be foundon Transformation PCS. If, instead, the

good performance found on OPC trialsresulted from accurate chroma informationrather than the enlarged intervals, thenperformance on Transformation PCSshould be worse than on TransformationOPC and about equal to that on Trans-formation PC.

Method

Subjects, Ten students at the University ofWisconsin participated in Experiment 2. Musicalexperience was not a factor in their selection, butall reported having normal hearing. In return fortheir services, 6 were awarded extra credit in anintroductory psychology course and 4 were paid$2.SO an hour.

StimttU and procedure. Four six-tone sequencesdifferent from those used in Experiment 1 (butsimilar in nature) were chosen from exercises in asight-singing book for use as original untransformedstimuli. Each six-tone sequence began and ended onC.^ (262 Hz); all original melodies were in the keyof C major. The original tone sequences and theirtransformations are given in Table 4.

The same transformations used in Experiment 1were used in Experiment 2 with the exception that anew transformation, PCS, was tested in placeof Transformation LT. Transformation PCS wassimilar to Transformation OPC; however, ratherthan displacing tones by an octave (12 semitones)relative to the original melody, the tones in Trans-formation PCS were changed an additional ±2semitones. This transformation thus preservedcontour without preserving chroma but yieldedsequences whose ranges were approximately thesame as those of the sequences in TransformationOPC, which preserved both contour and chroma.

Unlike the OPC transformations in Experiment1, none of the OPC transformations shared tonesor intervals with its corresponding original sequence.Also, only one OVC transformation (Melody 2)contained a note identical to the correspondingnote in the original sequence.

With the exception of the use of different melodiesand the use of the PCS transformation instead ofthe LT transformation, the procedure of Experiment2 was identical to that of Experiment 1. Since thePCS transformation spanned a wide range offrequencies, the PCS sequences always began onC4, as did the OPC and OVC sequences. ThePC and O transformations, however, were sub-ject to presentation in any of five octaves as inExperiment 1.

Results and Discussion

Only data from those subjects whocorrectly identified the original untrans-formed tone sequences on at least 50%

86 D. MASSARO, H. KALLMAN, AND J. KELLY

Table 4Notes and Note Frequencies of Tone Sequences used in Experiment 2

Transformation Frequencies Notes Contour*

Sequence

OPCPCSOPCOVC

262,262,262,262,262,

392,330,698,784,196,

294,247,165,147,587,

440, 349,392, 294,

1568, 784,1760, 698,220, 698,

262262262262262

1

C4G4C 4 E 4

C 4F 6

C 4 G 6C4G3

D4B3E,D3D6

A4G4G6

AsA3

F4D4

G6F6F6

C4C4C4C4C4

-j |-I 1-| 1H 1- + - + -

Sequence 2

OPCPCSOPCOVC

262,262,262,262,262

247,220,147,123,404

262,294,466,698,131

349,494,

1175,1397,349

330,349,740,659,659

262262262262262

C4B3C4AsC4D3C 4 B 2C4B4

C4 F4D4 B4A*4D6

F6 F6

C3 F4

E4 C4

F4 C4

F*6C4

E6 C4

E6 C4

h H— -j — |

|- -1

h HH (- +

_

———

Sequence 3

OPCPCSOPCOVC

262,262,262262,262,

330,392,587659,165,

294,330,165147,587,

392,494,

13071568,

196,

349, 262440, 262784 262698, 262698, 262

C4C4C4C4C4

E 4 D 4G 4 E 4D6E3E 6 D 3

E S D 6

G4B4

F6

G6G3

F4

A4G6F6F6

C4C4C4C4C4

H 1 —H 1 — —-1 j-

- + - + -

Sequence 4

OPCPCSOPCOVC

262,262,262,262,262,

294,349,523S87147,

330,392,

1175,1318,659,

247, 330,294, 440,139, 587,123, 659,494, 165,

262262262262262

C4C4r,c,C4

D 4 E 4F4 G4C6 D6

D B E 6

D3E6

B3 E4D4 A4C*3D6

B4 E3

C4C4c,C,C4

+ H h+ H h+ H h

I I i n i 1

- + - -

—_

_ T.

+

No/e. O = octave; PC = preserving contour; PCS = preserving contour, stretched; OPC = octave-pre-serving contour; OVC = octave-violating contour.a A minus sign denotes a descending interval; a plus sign denotes an ascending interval.

of the trials over the 3 experimental dayswere included in data analysis. Of thesubjects who participated in Experiment2, data from seven met this criterion.

The percentages correct for each trans-formation of each sequence on the 3experimental days appear in Table 5 alongwith the data averaged over the 3 days.

The main effect of transformation wassignificant, F(4, 24) = 14.61, MS, = .734,and thus orthogonal comparisons betweenthe means were carried out. Althoughperformance on the original melodies wasslightly better than performance on theOPC melodies, this difference was not sig-nificant, F(l, 24) = 3.46, .05 < p < .10.This finding agrees with the results ofExperiment 1 and with those of Idson andMassaro (1978) in suggesting that the dif-

ference between the O and OPC conditionsis relatively small and usually nonsignifi-cant. In agreement with the results ofExperiment 1 was the finding that per-formance on the contour-preserving trans-formations that preserved chroma informa-tion (O and OPC) was better than on thecontour preserving transformations thatdid not (PC and PCS), F(l, 24) = 17.93.Of particular interest in Experiment 2 wasthe absence of a significant difference be-tween Transformations PC and PCS,^(1, 24) = 1.40. This finding suggeststhat the good performance on Trans-formation OPC cannot be explained byreferring to the stretched contour presentunder this transformation. As expected,performance on Transformation OVC wasworse than the average performance on the

MELODY RECOGNITION 87

IS

IHft]

•*.

"^3§

fl3Ǥ,^11t?S

"a

|

I

1

£Cj

^So

*2 "«3 S£-H 0^

&

3$> B0 jjj

||

|"

U

O

otf)

kJCJ0,

O

C_o'+3

°12"*"*

)j*

0

y0

8BH

Oa,

O

fl_g

IN »j

>. IQ to

S

u

uj^o

Scu

o

C.9

«|

*

S

S§

PH

O

•g.21

O fO *— ' *O O

SK^SS

co t— O O CN^ " r^ 00 10

00 Oj l/) o 10

S5 oo S Si S

o 'i fo r-> co

\O ^* rH C* 00*o ^* o^ o

IO " (N GO 10

0 ^0 fO o\ CM ° -^ t*~ *O

\§ W 10 S ^»

S3SSS

vO O T-I CO 00IO *^" CO J>« IO

CO -* CS OO »O

S S 5 5

\O O *-i O f^10 00 so O> S

co w> ro CS

ON O O\ 00IO OO CO iO IO

%s^^

S 3 S S 3

^SKg;s

la-1 " "> •*

,_;3Oa8

1jSo

I%uooIH3Q+jCH

8a

S.

18uuo"S

!touT9OB8.S

i2Jo,II

C/5Ucu

-*-)

ba.S

II

kJ

S

u0.$!§

transformations that preserved the originalsequences' melodic contour, F(l, 24) =35.51.

Overall performance remained constantfrom Experimental Day 1 to Day 2 butimproved by 6% on Day 3, F(2, 12)= 10.44, MSe = .082. Unlike Experiment1, day did not interact with transformation(F < 1, MS* = .124).

Tone Sequences 2 and 4 were identifiedcorrectly more often than were Sequences1 and 3, as indicated by a significant effectof tone sequence, F(3, 18) = 13.08, MSe

= .990. Sequence did not interact withtransformation, F(U, 72) = 1.70, MSe

= .556.Both the effect of octave of presentation

and the interaction of octave and trans-formation were not significant (F < 1) ineach case (MSes were .088 and .104,respectively). The absence of an interactionbetween octave and transformation inExperiment 2 supports our view that thesignificant interaction between these fac-tors in Experiment 1 was due primarilyto the high power of the test.

Response confusion matrices for Experi-ment 2 are presented in Table 6 for boththe OVC transformation and averagedover the other transformations. To testthe degree to which performance on Trans-formation OVC depended on contourinformation, we calculated the correlationbetween the number of common directionsof the successive intervals between theOVC and O stimuli and the number oftimes that the OVC stimuli were confusedfor incorrect O sequences. This correlationbased on the data presented in Table 6was only marginally significant on a one-tailed test, r(10) = .42, .05 < p < .10.Although this degree of correlation issomewhat less than what was found inExperiment 1, performance on the othertransformations indicated that contourinformation did play an important role inthe judgments made in Experiment 2.Table 6 presents the averaged confusionmatrix averaged over Transformations O,OPC, PCS, and PC. The correlation (dis-regarding correct responses) between thenumber of common directions of the sue-

88 D. MASSARO, H. KALLMAN, AND J. KELLY

Table 6Percentage of Each of the Four TransformedSequences Identified as Each of the FourOriginal Sequences for the FiveTransformations in Experiment 2

Response"

Transformation OVC1234

1234

11 (D16(2)5 (1)7 (1)

Transformations48(5)10 (3)41 (5)

6 ( 2 )

47 (3)43 (2)42 (3)40 (3)

OPC, PC,

11 (3)65 (5)53 (3)10 (2)

25 (1)30(4)32 (1)20 (1)

O, and34 (5)15 (3)40 (5)

6 (2)

17 (4)11 (3)21 (4)34(2)

PCS8(2)

10(2)6 (2)

79 (5)

Note. O = octave; OPC = octave-preserving con-tour; OVC = octave-violating contour; PC = pre-serving contour; PCS = preserving contour,stretched.0 Scores in parentheses are the number of successiveintervals shared by the transformation and the ori-ginal tone sequence.

cessive intervals and the number ofconfusions between tone sequences wasvery high for the data from the contour-preserving transformations, r(W) = .71,and suggests that contour played animportant role in the judgments.

The major results of Experiment 2 are inaccord with those of Experiment 1 and,moreover, suggest that the salience of thecontours of the OPC tone sequences wasnot the sole contributing factor to the goodperformance on the OPC sequences.

General Discussion

The main purpose of the present studywas to evaluate the relative contributionsof tone chroma, melodic contour, and toneheight to melody recognition when un-familiar tone sequences were used asstimuli. The results of the present experi-ments implicate both tone chroma andmelodic contour as important. The smalldecrement in performance on Transforma-tion OPC relative to performance on the

original melodies replicates the Idson andMassaro (1978) experiments but contrastssomewhat with the results of a recentstudy by Kallman and Massaro (1979).In that study, subjects heard familiarmelodies (e.g., "Yankee Doodle") andwere asked to try to identify the melodies;however, each melody was presented to agiven subject only once. One of the threemelodies ("London Bridge") gave a rela-tively large decrement in identificationperformance on the OPC relative to theuntransformed version. It may be thatdistortions of tone height can be easilycompensated for when they are expected(as would be the case when subjects hearnumerous repetitions of distorted melodies),but distortions of the relative tone heightof successive intervals might result inpoorer recognition when the distortionsare unexpected. However, because therelatively large decrements on Transforma-tion OPC occurred for only one of themelodies in the Kallman and Massarostudy, this result must be interpreted withcaution.

Similar to the case with tone height,it is necessary to evaluate how the im-portance of tone chroma and melodiccontour varies depending on the experi-mental task. For example, both the resultsof Idson and Massaro (1978) and thoseof the present study showed relatively goodperformance when melodies with correctcontour but incorrect tone chroma werepresented. But Kallman and Massaro(1979) found that when subjects heardeach of three melodies only once and werenot told the names of the possible melodiesin advance, PCS transformations of thefamiliar melodies were almost never recog-nized. This is not to say that contourinformation was not used as effectivelyby subjects in the Kallman and Massarostudy. Rather, good recognition of themelodies depended on the presence of bothcontour and chroma information.

To effectively use contour informationin the absence of tone chroma information,it may be necessary that the subject knowthe possible melodies in advance. In thiscase, the subject could find the best match

MELODY RECOGNITION 89

between the contour of the presentedmelody and the contours of the possiblemelodies. To the extent that this strategyseems effective, the subject might use con-tour as the primary cue. In fact, in a post-experimental interview, one of the subjectsin the present study indicated that heremembered the untransformed tone se-quences by drawing lines in his head corre-sponding to the direction of pitch changes.This representation of melodic contour wasthen used as a basis for categorizing thetest stimuli.

Despite some differences in results duepossibly to experimental procedure, thereare important consistencies among theresults of Idson and Massaro (1978),Kallman and Massaro (1979), and thepresent experiments. Specifically, perform-ance on the OPC transformations wasconsistently much better than on the OVCand the PCS transformations. This resultstrongly suggests that tone chroma isfunctional in recognition of both highlyfamiliar and recently learned melodies.Furthermore, all of the studies demon-strate that tone chroma information aloneis not sufficient for accurate melody recog-nition. For chroma to be effective it mustbe accompanied by accurate contour in-formation.

We have argued that tone height,melodic contour, and tone chroma mayall contribute to the perception of amelody. How these characteristics areprocessed remains an important concern.For example, to say that tone height isimportant is not to say that the heightsof the individual tones in a melody providethe cues necessary for melody recognition.Whether a given melody is sung by a bassor by a soprano voice does not usuallyinfluence a listener's ability to recognizethe melody. The important feature formelody recognition is that the relativeheights of successive tones are maintainedwhen a melody is transposed from bassto soprano.

Exact tone height is not necessary,however, for melody recognition as indi-cated by the good performance on Trans-formation OPC. When an ascending major

ninth is substituted for an ascendingmajor second, the relative height of the suc-cessive tones (interval) is violated. Wepropose that octave generalization is re-sponsible for the good performance onTransformation OPC. Octave generaliza-tion refers to the perceived similarity ofnotes standing in an octave relationship(Blackwell & Schlosberg, 1943; Hum-phreys, 1939). In the melody recognitiontask, octave generalization suggests thatthe ascending major ninth should beperceived as similar to an ascending majorsecond. Although octave generalizationfacilitates melody recognition, substitutingan ascending major ninth in a melody for amajor second might increase the difficultyof recognizing the melody (cf. Kallman &Massaro, 1979). This is because octavegeneralization does not fully compensatefor the violation of the relative heightof the successive interval.

The relatively good performance ontransformation PC suggests that the direc-tion of pitch change (melodic contour)provides an important cue to the listener.Given an ascending ninth, octave gen-eralization would suggest a number ofpossible interpretations of the interval,among them an ascending major ninth,an ascending major second, and a descend-ing minor seventh. However, given thatthe direction of pitch change was up, adescending minor seventh would representan inappropriate interval; the intervalwould thus not be interpreted as a de-scending major seventh. It is for this reasonthat we found that tone chroma informa-tion was not helpful when the melodiccontour was violated. An explanation forwhy a major ninth might be readily inter-preted as a major second rather than aninth is that melodies rarely containvery large successive intervals, and, con-sequently, an interpretation of the intervalas a second would be appropriate.

In summary, melodic contour, tonechroma, and relative tone height allcontribute to melody recognition. Anytheory of melody recognition will haveto describe how these characteristics are

90 D. MASSARO, H. KALLMAN, AND J. KELLY

evaluated and integrated together in per-ception and recognition.

References

Attneave, F., & Olson, R. K. Pitch as a medium:A new approach to psychological scaling. Ameri-can Journal of Psychology, 1971, 84, 147-166.

Blackwell, H. R., & Schlosberg, H. Octave generali-zation, pitch discrimination, and loudness thresh-olds in the white rat. Journal of ExperimentalPsychology, 1943, 33, 407-419.

Deutsch, D. Octave generalization and tune recogni-tion. Perception & Psychophysics, 1972, 11, 411-412.

Deutsch, D. Octave generalization and melodyidentification. Perception & Psychophysics, 1978,23, 91-92.

Dowling, W. J. Recognition of melodic transforma-tions: Inversion, retrograde, and retrogradeinversion. Perception & Psychophysics, 1972, 12,417-421.

Dowling, W. J., & Fujitani, D. A. Contour, interval,and pitch recognition in memory for melodies.Journal of the Acoustical Society of America, 1971,49, 524-531.

Dowling, W. J., & Hollombe, A. W. The perceptionof melodies distorted by splitting into severaloctaves: Effects of increasing proximity andmelodic contour. Perception & Psychophysics, 1977,21, 60-64.

Humphreys, L. G. Generalization as a function ofmethod of reinforcement. Journal of ExperimentalPsychology, 1939, 25, 361-372.

Idson, W. L., & Massaro, D. W. A bidimensionalmodel of pitch in the recognition of melodies.Perception 6- Psychophysics, 1978, 24, 551-565.

Kallman, H. J., & Massaro, D. W. Tone chroma isfunctional in melody recognition. Perception &Psychophysics, 1979, 26, 32-36.

Received April 19, 1979Revision received August 7, 1979 •