Grouping of vowel harmonics by frequency modulation - Springer

2524_3BF03203015.pdf

Perception&Psychophysics

1986. 40(3).183-187

Grouping ofvowelharmonicsbyfrequency

modulation:Absenceofeffectson phonemiccategorization

R.B.GARDNERandC.J.DARWIN

UniversityofSussex. Brighton. England

Amistunedharmonicmakesareducedcontributiontothephoneticqualityof a vowel. The twoexperimentsreportedhereinvestigated whethertherateoffrequencychangeovertime of aharmonicinfluenceswhetheritcontributesperceptuallyto a vowel'squality.Intheseexperi ments,frequency-modulatingoneharmonic atadifferentrateorwithadifferentphasefromthat used tomodulateremainingharmonicsofa vowel had no effect onthevowel'sperceivedcategory. Theseresultsareconsistentwiththoseofpreviousexperiments andwiththehypothesisthat coherenceoffrequencymodulationis not used togrouptogethersimultaneousfrequencycompo nentsintospeechcategories.

The harmonics

ofa periodic sound such as a voiced vowel have frequencies that are integer multiples ofthe fundamental. This fact can be exploited by perceptual mechanisms that grouptogethersounds fromdifferent sources. Forexample,frequencycomponentsthat devi ate by more than around 3 %from anintegermultipleof the fundamental make a reducedcontributionto the pitch ofacomplex tone (Moore,Peters,&Glasberg,1985) or to the phonetic quality ofa vowel(Darwin&Gardner,

1986). Adifferenceof 8% completely excludes a har

monic from the calculation of pitch.

Inaddition, formants

that have a commonharmonicspacing are more likely to begroupedtogetherindeterminingphonetic quality than are those that have different harmonic spacings (Dar win, 1981;Scheffers,1983).

Itis less clearwhetherdynamicpropertiesof pitch

movementexertan independent effect on perceptual grouping, over and above the static effectsjustdescribed. Will acomponentthat is acceptably in tune be excluded because the trajectory of its frequency movement is differ ent from that ofitscontemporaries?

Two simple types

ofpitchmovementthat are found in sungvowels, for example, are vibratoandjitter.In vibrato the frequencyofthe fundamental ismodulatedat a con stant rate, whereas injitterit fluctuates randomly.

Inboth

types ofmovement,harmonicrelationsbetween the fre quencycomponentsare maintained-theirmodulation is coherent. Itispossibleto synthesize signals in which the compo nentshave incoherent modulation(e.g.,McAdams, 1984). This research was supported by Grants GRIC 60099, GRID 65930, andGRIC 8522 fromthe UKScienceandEngineeringResearchCouncil. Early discussions with AI Bregman during his visit to the laboratory helped tocrystallizethe experiments. Theauthors'address is: Labora tory ofExperimentalPsychology, University of Sussex, Brighton BNI 9QG, England.Oneharmoniccan befrequency-modulatedat, say, a different rate from theothers. Provided the depthof modu lation is less than around 3 %,the instantaneous frequen cies oftheharmonicswould still be sufficiently close to integerratios for staticgroupingmechanisms totreat all theharmonicsas if they came from a commonsource, buttherewould be dynamicinformationavailable that could indicate that one oftheharmonicsdid not belong with the others.

Intheseexperiments,weinvestigated

whethersuchcoherence ofmodulation can be exploited by the auditory grouping mechanisms responsible for de terminingthe phonetic qualityofa vowel.

The influence

ofincoherentvibrato andjitteron source assignment was recently studied by McAdams (1984). He found that complex tones consisting of16equal-amplitude componentswereperceivedas being composed ofmulti plesourceswhen onepartialwas modulated with inco herentjitter,so that itsfrequencymodulation was incon sistent with that oftheremainingharmonics.However,

McAdams failed to find anyeffect

ofthecoherenceof modulationontheperceptual prominence of a target vowel whosefundamentalwas modulatedat a differentfrequency from that oftwo other simultaneous vowels, or which was modulated againstunmodulatedbackgroundvowels.

McAdams's(1984)experiments indicatethat coherence

of frequency modulationofharmonics influencesthe num ber ofsound sources but does not influenceperceptual prominence. A similar dissociation, this time between the number ofsound sources and their phonetic quality, has been found whendifferentformants of a vowel or sylla ble have different pitches (Cutting, 1976; Darwin, 1981). Listenerscan hear the phonetic quality given bytheir groupingtogether formants on different pitches while at the same time reporting that there is more than one sound source present.

Aratherdifferentapproachwas used by Bregman and

Doehring (1984) to investigate whether rate of linear fre-

183Copyright1986 Psychonomic Society, Inc.

184GARDNERANDDARWIN

quencychangeovertime(frequencyslope) is used to grouptogethersimultaneouslyoccurringfrequency sweeps.To test howstronglyaparticularcomponentwas boundtoothers,theytestedhow easily itcouldforma differentperceptualgroupwithothercomponents.They found that the middle one ofthreesimultaneousfrequency sweeps was more easilycapturedby adifferentpercep tualstreamwhen itsfrequencyslope was not the same as that oftheothersimultaneouslypresentsounds.Ifthe slope ofthe sweep wasparallelto theothertones,it was moreeasilycapturedwhenit did not form asimplehar monicrelationshipwiththem.This lastresultisconsis tent with the finding ofDarwinandGardner(1986) that mistuningaharmonicreducesitscontributionto the pho netic quality ofavowel.However,theeffectofthe fre quency modulation (PM) slope could also be based on mis tuning; a different slope on the centralcomponentmeans that simple harmonic relationships are not maintainedover time,introducing staticmistuning.The timeoverwhich componentswithdifferentslopes are sufficiently in tune (say ±8%)to be grouped together by purely static mecha nisms will vary with thedifferencein slope.Bregmanand Doehring'sresults arecompatiblewith thehypothesisthat simultaneouscomponentsare not grouped byvirtue ofa commonslope offrequencychange over time.

Insummary,it isclearthat amistunedcomponentis

less well integrated into a complex than one that is in tune.

Thelackofintegrationresultsinsubjects'hearingboth

multiplesoundsourcesand achangedphoneticquality.

But the onlyevidence that dynamic properties

offrequency movement contribute toperceptualintegration comes from

McAdams's(1984)experimentsonvibratoandjitter

of harmonics,inwhichsubjectsjudgedthenumberof sourcesthat theyheard.Therehas been noevidencethat thecoherence ofvibratoandjittercontributestogroup ing asmanifestedby theperceivedphoneticquality.In theexperimentsreportedhere,welookedfor such evidence.

Forvowels such as [I] and [e], which can bedistin

guishedby thefrequency oftheirlow firstformant(Fl), thelistener'scomputationofFlis based on therelative amplitudes of individual resolved components ofthe vowel spectrum(Darwin,1984b;Darwin &Gardner,1985).

Thecomputation

ofFlfollowsgroupingprocesseswhich assignthesecomponentsto the samesource(Darwin,

1984a, 1984b).

Ifcommonfrequencymodulationis used

togroupharmonicstogether,then aharmonicwhose FM characteristicsareinconsistentwith theremaininghar monicsof the vowelmightbeexpectedto beassignedto adifferentsoundsourceand tocontributeless to the as sessmentof vowelquality.Theperceptualintegration of the targetharmonicinto a vowel can bemeasuredby pho nemeboundaryshiftsproducedby changes inperceived vowelcolor. All ofthepresentexperimentsinvolvedmanipulation ofaharmonicclose to thefirstformantin aseriesof vowelsdifferinginfirst-formantfrequencybetween[I] and [e]. Perceptualexclusionof a harmonic close to a for

mant peak gives a shift in theperceivedfirst-formantfre-quency that can bedetectedin acategorizationexperi

ment as a change in the position of thephonemeboundary.

Ifthefrequencymodulationof aharmoniccauses it to

begroupedoutfromthe vowel, the[I]-[e]boundary should shift.

EXPERIMENT1

The aimofthisexperimentwas todeterminewhether

incoherentmodulation ofa singleharmonicofa vowel reducesthecontributionof thatharmonicto thevowel's phoneticquality.Inorderto do this, weusedan[I]-[e] continuum,differingin

Fl,and wetestedwhetherthe

phonemeboundaryshiftedwhen aharmonicnearto F1 wasmodulatedincoherentlyfrom the rest. As a control for simplemistuningeffects,weincludedconditionsin which the sameharmonicwasmistunedby aconstant amountequalto itsmaximummistuningunderincoher entmodulation.Tocalibratethe size ofanygroupingef fect, we alsoincludedaconditionin which the same har monic wasphysicallyremovedfrom thevowel.

Ifcoherenceofmodulationis used togroupharmonics

forphoneticcategorization,then weshouldfind a pho nemeboundaryshiftin theconditionsinwhichthe har monic close to

Flismodulatedincoherentlyrelativeto

the remaining harmonics. This shift should begreater than that obtained with simple static mistuning.Ifthe harmonic is beingcompletelygroupedout, thephonemeboundary shift should be aslargeas that found when theharmonic isphysicallyremovedfrom thevowel.

Method

Stimuli.Steady-state vowels were synthesizedusing additive sine wave synthesis based onKlatt's(1980) cascadesynthesizer.Klatt's publishedprogramwas modified toproducethetransferfunction (after the initialspectrallyflatpulse-traininput)appropriatefor a particularvowel. Thistransferfunction was thenevaluatedat har monicfrequencies ofa125-Hz fundamental, and sine-waves of the appropriateamplitudeand phase were addedtogetherto give the complete vowel.

Forharmonicsoffrequency-modulatedvowels,

thetransferfunction was evaluated at theinstantaneousfrequency for each sample point and the appropriate phase and amplitude values were derived. Vowel continua consistingof nine sounds of 500 rnsec duration with 16 rnsecriselfalltimes weresynthesizedon a fun damental of 125 Hz. The sounds varied in the value ofFlfrom

375 to 543 Hz in equal 21-Hzincrements,givingII/-likesounds

at low Flvalues andlei-likesounds at highFlvalues. The values ofthe second,third,fourth,and fifth formants were 2300, 2900,

3800, and 4800 Hz,respectively.The bandwidths

ofthe first three formants were 90, 110, and 170 Hz,respectively.The bandwidths ofthe fourth and fifth formants were set at 1000 Hz. Oneexperimentalcondition,the basiccontinuum,used no fre quencymodulation.

Forthe other vowelcontinua,some or all of

theharmonicsweresinusoidallyfrequencymodulated.The depth of modulation was always 2 %(that is, 34cents-withinthe range used byMcAdamsand well above detectionthresholdvalues) and themodulatingwaveformstarted in sine phase. In one of thecoherentmodulation conditions allharmonicswere modulated at a frequency of 6 Hz; in theother,allharmonicswere modulated at afrequencyof 10 Hz. In theincoherentmodulationconditions,the500-Hzharmonic was chosen toreceivedifferentmodulationfrequenciesfrom the otherharmonics,becauseour previousexperimentsshowed clear

VOWELHARMONICSANDFREQUENCYMODULATION185

435

12subjects

.--00Hzbackgroundmodulation •···..··..·..·06Hzbackgroundmodulation ...'-'-'-'710Hzbackgroundmodulation !.j . , I 460
455
-;; :I: 450
'""0445 c "0 .Q 440
u, o

Targetmodulationfrequency

large upwardshift in theboundaryto 484 Hz. An in dividual ttest showed this shift tobesignificant at

P<.01.

The effectsofmodulatingallcomponents coherently.

Modulatingallcomponentscoherentlyprovideda con

trol for any change in vowel quality that modulation it self might introduce. The filled symbols of Figure 1show theboundaryvalues for thecoherentmodulation condi tions, that is, conditions in which the target and the back groundmodulation frequencies were equal. Individual ttests showed that modulationofthe target at frequen cies of6 Hz and 10 Hz gave phonemeboundariesthat were not significantlydifferent (p>.05) from that for theunmodulated(0 Hztargetandbackground)condition or from each other.

The effects ofincoherentmodulation.Figure 1 also

shows the boundaries for thevariousincoherentmodula tionconditions,which arerepresentedby open symbols.

Individual

ttests showed that modulating the5OO-Hztar get at

10Hzagainstan unrnodulatedbackgroundproduced

nosignificantshift (p>.05) in theboundarycompared to that for the unmodulatedcondition.Neither was there anyeffect ofleaving thetargetunmodulatedagainst a 6

Hzbackground.

Modulating the target at a frequency different from that ofthebackgroundalsoproducedno significantboundary shifts. Thus the phonemeboundaryfor a 6-Hztarget against a 10-Hzbackgrounddid not differ significantly from that found when all theharmonicswerecoherently modulatedat 10 Hz.Conversely,theboundaryfor the lO-Hztarget and 6-Hzbackground conditiondid not differ from theboundaryfor thecoherent6-Hz condition. Figure1.Experiment1: Phonemeboundariesof vowelcontinua asa function of the modulation frequency of theSIlO-Hztargethar monic for different frequencies ofbackgroundmodulation. Filled symbols show theboundaryvalues forcoherentmodulation. Open symbols show theboundariesfor the variousincoherentmodula tion conditions in which thetargetmodulationfrequency differed from thatof the remainingbackgroundharmonics. Verticalbars arestandarderrorsacross 12 subjects. effects on thephonemeboundary of theperceptualremoval of this harmonicfrom the vowel(e.g.,Darwin,1984a, 1984b). In two conditions, the

5QO-Hzcomponentwasfrequency modulatedat 6 and

10Hz while theotherharmonics weremodulatedat10and 6 Hz,

respectively.Inanother condition, the

5QO-Hzcomponent was modu

lated at

10Hz against anunmodulatedbackground;in another, it

wasunmodulatedagainst a 6-Hzbackground. Two rnistunedconditionswere alsoincluded,in which the un modulated

5OO-Hzcomponentwas rnistuned by±10Hz against

anunmodulatedbackground.These valuescorrespondedto the greatest frequency deviation of the modulated

5OO-Hzcondition and

acted as a control for instantaneous rnistuning in the conditions with a modulated target against anunmodulatedbackground.

Inaddition,there was a continuum

ofunmodulatedvowels with the

5OO-Hzcomponentremovedcompletely.This acted as a com

parison for the size of the grouping effects. If an incoherently modu latedcomponentwas grouped outcompletely,its phoneme bound ary shouldapproachthat for this condition. Procedure.Todeterminethe phonemeboundaryin aparticular condition,only seven members ofthecontinuumwere used, to reduce the size oftheexperiment.Theparticularrange chosen for eachcontinuum was based onprevious experiments. Eachcontinuum member wasrepeated10times, giving a total of70 stimuli per con dition across the 10conditions-agrand total of 700 trials, presented inquasi-randomorder.The range ofaparticularcontinuum (cf. Brady &Darwin,1978) would not influence theresults,because all the conditionswererandomizedtogether.Twelvesubjects with normal hearing were used, eachcarryingout thecompleteexperi ment in one session. Stimuli werepresentedinrandomorder. The subjects were seated in asound-proofedcubicle. They listened to the sounds diotically over Sennheiser 414 headphones at

72dB SPL

on-line from a V

AX-ll/780computervia anLPA-lIKat a sam

pling frequency of10kHz; the sounds were low-pass filtered at 4.5 kHzand 48 dB/octave. Subjectsresponded on a VDU keyboard, pressingthe ikey forIIIsounds and theekey forlEIsounds. The subjects were free to repeat each sound as often as they liked. Trials followed keypresses afterIsec. Trial numbers were displayed and subjectscouldtake a rest atanytime. The subjectsreceived a practice sessionbeforetheexperimentalsession. Theentiresession lasted abouthalfan hour. Thecomputerscoredthe data on-line and fitted a probit function to each individualsubject'sdata for each continuum. The individual phoneme boundaries were taken as the 50% point ofthe probit func tion,expressedintermsof the

FIvalue used toprogramthe syn

thesizer.

Results

Mistunedconditions

andremovedcondition.The mistunedconditionsacted as a control against which any phonemeboundaryshift found for the modulated condi tions must becompared.On the basis ofpreviouswork (Darwin&Gardner,1986), we wouldexpectmistuning by ±10 Hz, as used here, to give nosignificantshift in theboundary.That expectation wasconfirmed.The boundariesfor the +10Hzand-10Hzmistuned con ditions were both 442 Hz. These did not differ signifi cantly from the original boundary value of446 Hz (in dividual ttest,P>.05).

Removing the

5OO-Hzcomponentcompletelygave an

estimateof the maximum shiftwe would expect ifthe 500

Hzcomponentwere beingperceptuallycompletely

grouped out ofthe vowel percept. We found the predicted

186 GARDNER AND DARWIN

Insummary,no evidence was found for grouping by

frequency modulation: Incoherently modulatedtargets did not produce thepredictedupward shifts in thephoneme boundary.

Discussion

The results suggest that coherence of frequency modu lation is not a necessary condition for the grouping togetherofharmonicallyrelated frequency components.

Harmoniccomponentswhose FMcharacteristicsdid not

match those of thebackgroundwere fullyintegratedinto the vowelspectrum.This was true not only for differ ences in modulation frequency between target and back ground, but also when the target wasunmodulatedagainst a modulatedbackgroundand when it was modulated against anunmodulatedbackground.Oneexplanation of these results is that the differences in modulation fre quency between the target and background harmonics were too small to be effective.

EXPERIMENT2

This experiment waspartiallya replication of Experi ment 1, but using adifferentrange of target modulation frequencies against a 6-Hz background. In addition, a number of conditions were introduced in which the start ing phase of the target modulation waveform wasvaried while its frequency was held constant at 6 Hz. This in troduced a phase-based incoherence into the target modu lationcharacteristics.

Method

Stimuli.Thesynthesisproceduresandsteady-statecharacteris tics of the vowels were identical to those of Experiment 1. The origi

nalunrnodulatedvowelconditionwas again included in theexperi-ment. Thebackgroundmodulationfrequencywas set at 6 Hz and

conditionswithtargetfrequenciesof 3, 6(coherent),12, 18, and

24 Hz werecreated.The depth ofmodulationwas again 2

%and the phase of themodulationwaveformwas 0

0•

Threefurthercon

ditions wereincluded,in which thetargetfrequencywas equal to thebackgroundfrequencyof 6 Hz but thestartingphase ofthe tar get wasvaried.Values of 90 0, 180
0, and 270 0 were used. Procedure.Theexperimentalprocedurewasidenticalto that of

Experiment1 except that there were only nine

conditions-agrand total of630trials.Eightsubjects with normalhearing,7ofwhom hadparticipatedinExperiment1, were used, eachcarryingout the experimentin onesession.

Results

Theeffectofcoherentmodulation.The solid sym

bols inFigure2 show the boundaries for the two coher ent modulationconditions. AsinExperiment

I,there was

no significantdifference betweenthe boundary for the un modulated condition (solid circle) and that for the condi tion in which all the harmonics were modulated coher ently with a frequency of 6 Hz (solid triangle).

Theeffectofincoherentmodulation.The open sym

bols inFigure2 show the boundary values for the inco herent modulation conditions, in which the target fre quency differed from that of thebackground.No differences were found between theboundariesfor the coherent andincoherentmodulation conditions. Figure 3 shows that there was no effect of varying the phase of the target modulatingwaveformwhen its fre quency was equal to that of thebackground(6 Hz).

Discussion

These resultsconfirmedthe findings ofExperiment1,

thatincoherentmodulation of a harmonic had no effect on theintegration ofthat harmonic into the vowel spec trum. This was true for differences between the modula-

8subjects"\1-.6Hzbackground.

modulatton 455
• OHzbackgroundmodulation N 450
+ t-t--t- :I: -t ", t :; 445
" c " 0 .c 440
L;: 435
o36121824

TargetmodulationFrequency(Hz)

Figure2.Experiment2:Filledcircle shows theboundaryfor theunmodulated vowelcontinuum(target andbackgroundmodulationfrequencyequal to 0 Hz); filledtriangleshows theboundaryfor vowelscoherentlymodulatedat a frequency of 6 Hz. Open symbols show theboundariesfor thevarioustargetmodulation frequencies of theincoherentmodulationconditions, againstabackgroundmodu lation frequency of 6 Hz.Vertical barsarestandarderrorsacross8subjects.

VOWELHARMONICSANDFREQUENCYMODULATION187

GENERALDISCUSSIONtion frequency

ofthe harmonic and thatofthe background ofup to 2 octaves. There was also no effectofincoher enceintroducedby phase-shifting the modulating wave form ofthetargetharmonic relative to that of the back groundharmonics,so that at a phase shift of180 0 the frequency ofthetargetrose while thatofthe others fell.

Startingphaseoftargetmodulation(deg.)

Figure3.Experiment2:Phonemeboundariesofvowelcontinua as afunctionofthestartingphaseofthemodulationwaveformof the500-Hztargetharmonic.Filledsymbolsbowstheboundaryvalue forcoherentmodulation.Opensymbolsshowtheboundariesfor thevariousincoherentmodulationconditionsinwhichthestarting phaseofthetargetharmonic'smodulatingwaveformdifferedfrom thatofthebackgroundharmonics.Verticalbarsarestandarder rorsacross8subjects. 455

8subjects

450
t---r-?---i = '" 445
"tl c: :l 0 440
D 435I-
the effects on phonemeboundariesofsmall amountsof addedenergyto a singleharmonic(Darwin&Gardner,

1985), to the effects

ofmistuningaharmonic(Darwin& Gardner, 1986),and to the effects of makingone harmonic start at a different time fromanother (Darwin,1984a,

1984b). The failure isconsistentwith theproposalthat

incoherence in FM influencesthe number ofsources heard but not thecategoryperceived. It is possible thatothertasks may be able to reveal simultaneousgrouping by frequency slope, sinceitis clear that we are able to detect thedifferencebetween coher ent and incoherentmodulation.Speech may not be the bestparadigmto use inattemptingto show such effects, since a substantial part ofspeech (allofvoiceless speech and at least part of voiced speech) hasexcitationthat is incoherentacrossdifferentfrequencyregions.

Itmight

bemoreappropriateto usejudgmentsofthe timbreof a melodicinstrumentwhose excitationis more consistently coherentthan that ofthe voice.

REFERENCES

(ManuscriptreceivedJanuary23, 1986; revisionacceptedforpublicationJune 17, 1986.) BRADY,S.A.,&DARWIN,C.J.(1978). A range effect in thepercep tion ofvoicing.JournaloftheAcousticalSocietyofAmerica,63,

1556-1558.

BREGMAN,A. S.,&DoEHRING,P. (1984). Fusionofsimultaneoustonal glides: The roleofparallelnessand simplefrequencyrelations.Per ception &Psychophysics,36,251-256.

CUTIlNG,

J.E. (1976). Auditory and linguisticprocessesin speech per ception:Inferencesfrom six fusions in dichoticlistening.

Psycholog

ical Review,

83,114-140.

DARWIN,C. J. (1981).Perceptualgrouping of speech components differ ing infundamentalfrequencyandonset-time.

Quarterly Journalof

ExperimentalPsychology,33A, 185-207.

DARWIN,C. J. (1984a).Auditoryprocessingand speechperception. InH. Bouma & D. G.Bouwhuis(Eds.),Attentionandperformance X:Controloflanguageprocesses(pp. 197-210).Hillsdale,NJ:

Erlbaum.

DARWIN,C. J.(l984b).Perceivingvowels in thepresenceofanother sound:Constraintsonformantperception.

JournaloftheAcoustical

Society

ofAmerica,76,1636-1647. DARWIN,C.J.,&GARDNER,R. B.(1985).Whichharmonicscontrib ute to theestimation ofthe firstformant?Speech Communication, 4,

231-235.

DARWIN,C.J.,&GARDNER,R. B. (1986).Mistuningaharmonicof a vowel:Groupingand phaseeffectson vowel quality.

Journalofthe

AcousticalSociety

ofAmerica,79, 838-844. KLATT,D. H. (1980).Softwarefor acascade/parallelformantsyn thesizer. JournaloftheAcousticalSocietyofAmerica,67,971-995. McADAMS,S. (1984).Spectralfusion, spectralparsingandtheforma tionofauditory images.Doctoralthesis,StanfordUniversity. MOORE,B. C.J.,GLASBERG,B.R.,&PETERS,R. W. (1985). Relative dominanceof individualpartialsindeterminingthe pitch ofcomplex tones. JournaloftheAcousticalSocietyofAmerica,77,1853-1860. ScHEFFERS,M. T. (1983).Sifting vowels:Auditorypitch analysisand sound segregation.Doctoralthesis,GroningenUniversity,The

Netherlands.

270180

I 90o
I The two experiments reported here have shown the fol lowing: (1)Modulation of a singleharmonicof a vowel at a frequency different from that of the other harmonics does not affect theintegrationof that harmonic into the vowelspectrum(Experiments1 and 2). (2) A change in the phase ofthe modulatingwaveformofa single har monic of a vowel relative to that ofthe other harmonics does not influence theintegration ofthat harmonic into the vowelspectrum(Experiment2). (3) Modulation of a single harmonic against anunmodulatedbackgroundof the remainingharmonicsdoes not affect the integration ofthe harmonic into the vowel spectrum (Experiment 1). (4) No difference exists between the phoneme boundaries for unmodulated vowels and those for vowels in which all the harmonics are modulated coherently (Experiments

1 and 2).

In theseexperiments,we failed to find any evidence for theauditorysystem'suse ofcoherenceofmodulation to grouptogetherresolved spectralcomponents.This failure is unlikely to bedue to insensitivityofthe method employed; this method has provedextremelysensitive to