Detection of across-frequencydifferencesin fundamental frequency RobertP. Carlyon Laboratory ofExperimental Psychology, Sussex University, Brighton BN19Q.G, Sussex, England

LaurentDemany a)andCatherineSemal Laboratoire dePsychoacoustique, Universitb deBordeauxII, 146rueLbo-Saignat, F-33076Bordeaux,France

(Received1November1990;accepted for publication27 August199'1) Listenersdiscriminated betweenpairsof complexsounds,eachconsisting of two groupsof components. Two harmoniccomplexes wereplayedout throughseparatechannels,andeach filteredto obtaina "lower"anda "higher"group.The "carrierfundamental frequencies (F0s)" of bothgroupswereusually125Hz; onlythosecomponents in thelowergroupwere resolvable by the peripheralauditorysystem.For the standardstimulus,the F0s of the two groupswerefrequencymodulatedcoherentlywith eachother,sothat they werealwaysequal. For the signal,the F0s of the two groupsweremodulatedincoherently(•r modulatordelay), sothat theydifferedby an amountthat variedsinusoidally betweenvaluesproportionalto the depthof FM (the dependent variable).Stimuliwereusuallypresented in continuous pink noise.The resultsshowedthat (i) whenthecomponents wereaddedin sineor cosinephase, the meanthresholdacrosslistenerscorresponded to a zero-peakmodulationdepthof 6%-7% (rms mistuning= 8.5%-10% ); (ii) performancedroppedto chancewhenthe upper comp6nents wereaddedin alternatingsine-cosine phase,but wasonlymoderatelyaffectedby thephaseof thelowercomponents; (iii) thresholdfor sine-phase stimuliimprovedby a factor of 1.6whennoisein the frequencyregionof the two component groupswasremoved;(iv) thresholdincreased moderatelywith increases in thefrequencyseparation betweenthe two componentgroups;(v) thresholddroppedmarkedlywhen the F0s of both groupsof componentswereincreasedsoasto be resolvableby the peripheralauditorysystem;and (vi) performance droppedto chancewhenthenominalcarrierF0s of thetwo groupsof components differedfrom eachother.It is concludedthat listenerscan performsimultaneous comparisons of F0s derivedfrom resolvedand unresolved harmonics,and that their performanceon this taskis fairly robust.Implicationsfor the perceptualsegregation of concurrentcomplexsounds, andfor modelsof pitchperception,are discussed. PACS numbers:43.66.Hg,43.66.Mk, 43.66.Nm [WAY]

INTRODUCTION

The basisfor manyof thesemodelscomesfrom modern theories of pitch perception(e.g., Goldstein,1973;Moore, There is a growingbodyof evidencethat listenerscan 1989), which in turn are basedmainly on experimentsin usedifferences in fundamental frequency(F 0) betweentwo which listeners are requiredto identifythe pitchof a single simultaneous periodicsounds to perceptually separate those sound source. Of particularrelevanceto the presentstudy two sounds(Broadbentand Ladefoged,1957; Scheffers, are experiments investigating therelativerolesof two groups 1983; McAdams, 1984; U. T. Zwicker, 1984; Brokx and of harmonics, which differ in the degreeto which they are Nooteboom,1982;Stubbsand Summerfield,1988;Chalikia resolved by the peripheral auditory system(Plomp, 1967; and Bregman,1989;Assmannand Summerfield,1989, 1990; Ritsma, 196'7; Moore et al., 1985). Such experimentsare Culling, 1990;Summerfieldand Assman,1991). For examinteresting because the two groups of componentsare asple,Scheffers (1983) reportedthat identification of simultasumed to convey F0 information in different ways:The lowneouspairsof vowelsimprovedwith increasing differences er, resolved, iharmonics form the input to a central "pattern between theF0s ofthetwovowels.F0 usuallycorresponds to recognizer" which derives pitch from the relationship bepitch,and its rolein the perceptualsegregation of simultatween the individual frequency components (Goldstein, neous sounds is one of the reasons for the continued interest 1973; Terhardt, 1974; Piszczalskiand Galler, 1979; Terin pitchperception,andhasbeenthe subjectof severalcomhardtet al., 1982), whereasthe upper,unresolved, harmonputationalmodels(e.g., Parsons,1976;AssmannandSumics convey pitch information in the repetition rate of their merfield,1989, 1990;Meddisand Hewitt, 1991a,b;Slaney composite waveform (Schouten, 1940, 1970). The data and Lyon, 1990). showthat the low-frequency,resolved,harmonicsare more important('"dominant")for pitch perceptionthan are the high-frequency, unresolved, harmonics (Plomp, 1967; Alsoat:Laboratoire d'Audiologie Exp6rimentale, INSERMU.229,H6pital Pellegrin,F-33076Bordeaux,France. Ritsma, 1967; Moore et al., 1985). However, there is evi279

J. Acoust.Soc.Am.91 (1), January1992

0001-4966/92/010279-14500.80

•c•1992 AcousticalSocietyof America

279

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.

dencethatlistenerscanalsojudgepitchfromthepurelytemporal informationconveyedby unresolved components or by amplitudemodulatednoise(BurnsandViemeister,1981; Warren and Wrightson, 1981; Houtsma and Smurzynski, 1990). Becauseof thesefindings,a numberof authorshave proposedschemesinvolvinga commonmechanismfor the

betweenresolvedand unresolved harmonics,and it is possible that identificationwas basedon the upper three formants, which all containedcomponentsthat were unresolvedby the peripheralauditorysystem.It is alsopossible that components of differentformantsinteractedwith each

extraction ofF0

influencedby within-channel,rather than across-channel

information

from both resolved and unre-

other on the basilar membrane, so that identification was

solvedcomponents.Such schemesinclude "autocorrelogram"models(Licklider, 1951;AssmannandSummerfield, 1990; Meddis and Hewitt, 1991a,b,e;Slaney and Lyon, 1990), the qualitativelysimilar "crude sketch"outlinedby Moore (1989), and the "pulse-ribbon"modelproposedby Patterson (198Yo; Patterson et al., 1991). Recently, HoutsmaandSmurzynski(1990) havepointedout that the modelproposedby SruloviczandGoldstein(1983) canalso

mechanisms.

perceivedsoundsources(see also Gardner et aL, 1989). However,his studywas not concernedwith the interaction

In the initial procedurelistenerswere requiredto discriminate betweena steady harmonic complex tone (e.g.,.

Broadbentand Ladefoged(1957) overcamethe problem of peripheralinteractionsby presentinglistenerswith synthesizedspeechsoundsmade up of two formants,with eachformant playedto a differentear. They reportedthat evenwith this dichoticmodeof presentation,listenersreported hearingone "voice" when the two formantswere playedon the sameF0. When the formantswereplayedon derivepitchfrombothresolved andunresolved harmonics.• differentF0s, but to thesameear,listenersreportedhearing Many of the newermodelsfall into the "autocorrela- more than one "voice." However, like Darwin, Broadbent tion" category.They proposethat the listenerperformsan andLadefogedwerenot interestedin theextentto whichthe autocorrelation ontheoutputsof eachof severaloverlapping two groupsof components differedin theirresolutionby the linearbandpass filters,with centerfrequencies (CFs) cover- peripheralauditorysystem.The two formantfrequencies ing the audiblerange.Recentversions(Assmannand Sumvariedcontinuously, and it is likely that therewerepartsof merfield, 1990;Meddis and Hewitt, 1991a,b,e) statethat the the speechsignalduringwhichbothformantscontainedreindividualautocorrelations are then summedto producea solvedcomponents. They did performa secondexperiment "summaryautocorrelogram." The channelspassingthe rewith steadysoundsconsistingof pairs of steady-stateforsolvedcomponentseachproducea seriesof peaksin their mants.However,astheypresentedtheir stimuliin quiet,and individualautocorrelograms at multiplesof the period of useda speechsynthesizer that producedformantswith shalthat component.When the autocorrelograms are summed low slopes(Lawrence,1953),2 it is possible that partsof theyproducea maximumat a periodequalto lIFO. Chantheirupperformantcontainedresolvedcomponents, or that nelsthat passgroupsof unresolved components havepeaks parts of the lower formant containedunresolvedcomponents. in their individualantocorrelograms corresponding to multiplesof the repetitionrate of their outpots--alsoequalto The experimentsreportedhere measuredthresholds (TAFOs) for the detectionof simultaneousdifferencesin F0 lIFO. Thus a multicomponent harmonicsoundproducesa peakin the summaryautocorrelogram at lIFO whetherthe betweentwogroupsof harmonics. The harmonics werechocomplexcontainsresolvedharmonics,unresolved harmon- sensothat onegroupwouldbewell resolved,andthe other its, or both. unresolved, by the peripheralauditorysystem.In mostof Regardless of the accuracyof specificm'odels,the idea our experiments,the stimuli were filteredso that the harthat listeners can combine information from unresolved and monicnumbersof the mostintensecomponents of thelower resolved harmonics isintuitivelyandecologically appealing: groupwerebetweenI and 5, and thoseof the.uppergroup were between11 and 15. Our choiceof filter settingswas it providesa parsimonious explanationfor a widerangeof influencedby evidencethat the resolvabilityof individual pitch phenomena,and wouldsupplya basisfor listenersto harmonicsdecreasesmarkedly as harmonicnumber is inusea commonF0 to grouptogethercomponents coveringa wide frequencyrange (Assmann and Summerfield,1990; creasedfrom 5 to 11 (Plomp, 1964;Moore and Glasberg, MeddisandHewitt, 1991a).However,thereisno unequivo- 1983;Houtsmaand Smurzynski,1990); this differencein cal evidence for a simultaneous interaction between the FOS resolvabilitywasexplicitlytestedin experiment1. Stimuli conveyedby resolvedand unresolvedharmonics. were presentedin a backgroundof pink noisein order to Darwin ( 1981) askedlistenersto identifyan ambiguous reducethe detectabilityof within-channelinteractionsbefour-formant sound,whosefirst three formants (F 1,F 2,F 3) tweenthetwogroupsof harmonics, andto limit therangeof formed the syllable/ru/, and of which F1, F3, and F4 harmonics in eachgroupthatwereaudible.Theresultsshow formedthesyllable/li/. PlayingF2 on a differentF0 to that that listenerscandetectacross-frequency differences in F0, of the other formants led to an increasein "/li/" ( F 1,F 3,F 4) andthat theirperformance isquiterobust.TAF0s weremearesponses, whereasplayingF4 onadifferentF0increasedthe suredas a functionof componentphase,the differencein numberof "/ru/" (FI,F2,F3) responses. Thus Darwin frequencyregionoccupiedby thetwogroupsof components, showedthatplayinga formantona differentF0 to thatof the signal-to-noise ratio (SNR), andbaselineF0. othersmadeit lesslikely to contributeto the perceptionof the compositesoundthan if all formantshad the sameF0. I. GENERAL METHOD He alsoshowed,usinga relatedtask,that playinga formant A. Rationale and preliminary experiment on an anomalous F0 led to an increase in the number of

280

J. Acoust.Sec. Am.,Vol. 91, No. 1, January1992

Carlyonet al.: DetectingF0 differences

280

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

harmonics1-5 and 11-19 of 125Hz) and onein whichthe

STANDARD

(b) SIGNAL

lowerandhighergroupsof components weremistuned in oppositedirections.This procedurewasabandoned, but is mentionedheresoasto clarifythe rationalefor the revised procedure,andbecausethe newprocedureand the way the dataareplottedwereinfluenced by theresultsobtained.The problemswith the originalprocedurewerethat (i) mistuningwasaccompanied by changes in the overallspectralextent of the complex;(ii) the F0 had to be randomizedfrom presentationto presentation,to preventlistenersfrom detectingmistuningby listeningto the F0 of an individual group(evenso,listenersmighthaveerroneously latchedon to the F0 of onegroupof harmonics);and (iii) with steady toneslistenersdo not haveto performa simultaneous comparisonof thetwogroupsof harmonics, evenwhentheoverall F0 is randomized:rather, they can simplyswitchtheir attentionfromonegroupto anotherandperformsuccessive comparisons (Demany and Semal, 1990). Despitethese problems,two findingswere obtainedthat were reliable acrosslisteners, andwhicharerelevantto thepresentexperiments.Oneof thesewasthat d' wasroughlyproportionalto thepercentage mistuningbetweenthetwo groupsof components;the other was that the adaptivethresholdwas about 10% for all listeners.

B. Main experiments: Signal generation and trial structure

The methodof signalgenerationand the trial structure areshownin Figs.1and2, respectively. In boththestandard and signalintervalsthe firstN (usually34) harmonicsof a 125-HzF0 werefrequencymodulatedat a rateof 2.5 Hz and

z

TIME .... > FIG. 2. Schematicspectrogram of (a) the standardand (b) the signalintervalsof each2I, 2AFC trial. The components drawn in dottedlineswere attenuatedby thefiltersshownin Fig. 1.Only thefirst 19harmonics of each complexare shown.Not shownis the fact that the startingphaseof the modulationimposed oneachcomplexwasrandomized fromintervalto interval.

played out through a 12-bit DAC, which we call DAC 1 (CED 1401laboratoryinterface;samplingrate 20 000 Hz).

All components wereaddedin sinephase,unlessotherwise stated,andweresinusoidallymodulatedby the samepercentageof their startingfrequencies, with the percentagemodulation (zero peak) beingthe dependentvariable.The output

harms of 125

Hz Filter 1 (125-625 Hz) DAC 1

of DAC 1 was filtered between 125 and 625 Hz (3-dB-down points), attenuated(WilsonicsPATT, not shown), and fed

to one input of a headphoneamplifier.The bandpassfilter consisted of two KemoVBF25.03filtersin series,(onehigh

pass, onelowpass,48dB/octave each3), andwillbereferred Filter2 (1375-1875 Hz) DAC 2

to henceforthas "filter 1." The secondinput to the headphoneamplifiercamefrom the output of a secondDAC ("DAC 2"), passedthrougha secondbandpass filter ("filter 2"), and attenuated (Wilsonics PATT). Filter 2 was similar to filter 1, but had cutoffs of 1375 and 1875 Hz. In the

standardinterval, the output of DAC 2 wasidenticalto that

of DAC 1,sothatthelistenerheardtwogroupsof frequencymodulatedcomponentsthat remainedin tune throughout the 400-msstimulus[Fig. 2(a) ]. The signalwasidenticalto the standard,exceptthat the modulationimposedon the high-frequencycomponents(DAC 2) was out of phase(•r modulatordelay) with that imposedon the lower compo-

PINK NOISE

nents (DAC 1). Hence the signal [Fig. 2(b)] consistedof

FIG. 1.Schematic representation of themethodof stimulusgeneration. 281

J. Acoust.Soc. Am., Vol. 91, No. 1, January1992

two groupsof frequencycomponentsthat movedin and out of tune with eachother throughoutthe stimulus:the ratio betweentheir F0s variedsinusoidallythroughoutthe stimuCarlyoneta/.: DetectingF0 differences

281

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.

lus,and the extentof thisvariationwasproportionalto that of the modulationimposedon the two groupsof components (Demany and Semal,1988). In bothsignalandstandardintervals,thestartingphase of the modulationwasrandomizedfrom presentationto presentation.This wasachievedby synthesizing an 800-mssample of eachstimulusand choosinga startingpoint for each 400-mssampleat randomfromthefirst400 ms.The levelsof all components withfrequencies in thefilterpassbands were 45 dB SPL, andthecomplextonesweregatedonandoffwith 5-msraised-cosine ramps.A 5-kHz-widepinknoisewaspresentedcontinuously. Its spectrumlevelin dB SPL was 17.8 at 500 Hz, 15.2 at 1000 Hz, 12.2 at 2000 Hz, and 8.8 at 4000 Hz.

All stimuli were presentedthrough one earpieceof a SennheiserHD414 headset,and were monitoredusingan HP3561A spectrumanalyzer.To convertthe voltageacross the headphones at eachfrequencyto the soundpressurelevelsquotedabove,thesoundpressure levelin a B&K artificial ear (type 4153, B&K condenser microphonecartridgetype 4134, externaldiameter0.5 in.) in responseto a 1-kHz sinu-

soidwasmeasured. The spectrumwasfurthershapedby the frequencyresponse of the headphones, which,measuredin the artificial ear, was flat between 125 Hz-1 kHz, and had

gainsof + 3, + 8, + 6, 0, and -- 1 dB re: the outputat 1 kHz at 2, 3, 4, 5, and 6 kHz, respectively. One advantageof usinga noisebackgroundis that both signaland noiseare shapedby theheadphone response, sothat thesensation level of the signalis not substantially affectedby it. The new paradigmis better than the original one describedin Sec.I A for threereasons:(i) The spectralextent of signaland standardare alwaysidentical;(ii) there is no needto randomizetheoverallFO,astherangeofF0s covered by eachgroupof components is the samefor standardand signal;and (iii) listenersare forcedto makea simultaneous comparisonof the continuallychangingFOsin eachgroup.

is proportionalto d' andto plot that variableon a logscale (e.g.,oneusuallyaverages and plotssignalenergyin dB). Both of these conventions were followed.

In someexperiments psychometric functions weremeasure
ageof 100trims.Eachof tenblocksconsisted of tentrialsat eachmodulationdepth,preceededby six practicetrials at the largestmodulationdepth used (20%). Listenerswere testedindividuallyin an IAC.single-walled sound-attenuating booth within a large single-walledsound-attenuating room.

II. EXPERIMENT

1: EFFECT

OF COMPONENT

PHASE

A. Rationale

The purposeof experimentI wasto providean initial measure of the smallest detectable difference in F0 between a

groupof resolvedcomponents and a groupof unresolved components, while controllingfor a numberof alternative cues.We wereparticularlyconcernedby two potentialcues. First, despitethe pink noise,listenersmightdetectbeating betweenthe (attenuated)components playedout from the two DACs, for example those with harmonic numbers aroundsevenandeight.Second,it is possible that the componentsin the uppergroupwould be partially resolvedby theperipheralauditorysystem.Below,wearguethatbothof theseconcernscan be addressedby independentlymanipulatingthe phaseof theupperandlowercomponents. Figure 3(a) showsthe outputof a simulatedauditory filter centered on 1675 Hz to the first 128 ms of an unmodu-

latedcomplex,in quiet,with anF0 of 125Hz andwith componentsaddedin sinephase.The F0 caneasilybeidentified from the regular repetition rate of the waveform.In Fig. 3 (b) the input is the same,exceptthat componentsare add-

edin alternating sine-cosine phase. 4 Theoutputwaveform has an ambiguousperiodand a much smallerpeakfactor, with the resultthat the F0 is hard to identify. In contrast,

C. Listeners and procedure

A total of five listenerstook part in differentexperiments.Their absolutethresholdsat octavefrequencies between 250 and 8000 Hz were within 15 dB of the 1969 ANSI standard. Listener RC was the first author. Stimuli were

presentedusing a 21, 2Arc procedurewith feedback.In most experimentsthresholdswere obtainedusingLevitt's

( 1971) two-downone-upadaptiveprocedure,whichconvergedonthe71%-correct(d' = 0.78) pointonthepsychometric function.The modulationdepth was multipliedby 1.07after everyincorrectresponse and dividedby 1.07after everytwo consecutive correctresponses, exceptfor the trials beforethe first four turnpointswhen a factor of 1.15 was used.Each run endedafter 16 turnpointsand the threshold for each run was obtainedfrom the geometricmean of the modulationdepthsat the last 12 turnpoints.Eachthreshold reportedhereis basedon the geometricmeanof six such

b)

runs. Geometric, rather than arithmetic, means were calcu-

lated becausethe preliminary experimentshad indicated that d' wasroughlyproportionalto % mistuning,whichin thepresentparadigmisproportionalto modulationdepth.It isconventional to averagethe logarithmof the variablethat 282

J. Acoust.Soc. Am.,Vol.91, No. 1, January1992

FIG. 3. Simulatedoutputof auditoryfiltersto thefirst 128msof thestimuli of experimentI. (a) filter CF= 1675 Hz, sine-phasestimuli; (b) CF = 1675Hz, ALT phase;(c) CF = 375 Hz, sinephase;(d) CF = 375 Hz, ALT phase.

Carlyonot at: DetectingFOdifferences

282

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

phasedoesnot markedlyaffectthe outputof filterswith low CFs, which resolvethe individuallower harmonics(Fig. 3(c) and (d); CF = 375 Hz). Given that listenersare less

sensitive to across-thanto within-channel phasedifferences (E. Zwicker, 1952;Patterson,1987a,1988), one would expectcodingof F0 (and hencethe detectionof across-frequencydifferences in F0) to beaffectedlessby the phaseof the lower harmonicsthan by that of the upperharmonics. Consistentwith this prediction,musicalintervalidentification basedon high harmonicsdeteriorateswhen the peak factor of the stimulus is reduced, whereas that based on low

harmonicsis unaffected(Houtsma and Smurzynski,1990).

TABLE I. Thresholds obtainedin thedifferentphaseconditions for eachof

thefivelisteners ofexperirnent 1.Figuresin parentheses arethenumbers by whichthethresholds shouldbemultipliedanddividedtoobtain+ / -- one standarderror. "*" indicatesthat an adaptivethresholdcould not be tracked.

Phase

RC

JD

6.0(1.2) 5.6(1.1) 8.4(1.1)

SINE

COS ALT-

HC

6.7(1.2) 5.0(I.1} 3.5(I.2) 5.8(I.2} 10.8(1.1) 5.2(1.2}

JC

AB

10.9(1.4) 7.7(I.1) 9.5(1.1} 9.5(1.3) 19.4(1.1} 24.4(1.2}

Mean

7.0 6.4 11.7

COS

COS ALT

ALT

B. Method

Five conditions were used. In two conditions, termed

"SINE" and "COS", all components werein sineor cosine phase, respectively. In a thirdcondition("ALT"), theoddnumbered components werein sine(0 deg)phase,andthe even-numbered components werein cosine(90 deg)phase. In conditionCOS-ALT, harmonics1-7 (playedout from both DACs) wereaddedin cosinephase,and harmonics834 wereaddedin alternatingsine-cosine phase,startingwith theeighthharmonicin sinephase.In conditionALT-COS harmonics 1-7 werein alternatingphaseandharmonics 834werein cosine phase. Thresholds weremeasured forthese fiveconditionsusingthe adaptiveproceduredescribed in Sec.I C, exceptthat in someconditions listeners couldnot reliablytracka thresholdsmallerthana 30% initialmodulationdepth.Therefore3-pointpsychemetric functions (modulationdepths---5%, 10%, 20%) weremeasured in those

obtainedby multiplyingour thresholds by V• and by 2, respectively. Thresholdsarequitesimilarfor thesineandcosinestimuli, with reasonable agreementacrosslisteners,andmeansof 7.0% and6.4%, respectively. Thisisconsistent with thefact that the outputsof auditoryfiltersto the two typesof stimu~ lusaresimilar(checkedusingsimulations similarto thosein Fig. 3). For the ALT-COS stimulussomelisteners'thresholds were elevated relative to those for the sine and cosine

stimuli, but all listenerscould perform the task and it was possibleto tracka thresholdin eachcase.This wasnot true when the higher componentswere added in alternating phase(conditionsALT and COS-ALT), whenit provedimpossibleto track a threshold.The reasonfor thisis apparent in Fig. 4, which showspsychemetricfunctionsfor conditions ALT, COS-ALT, and ALT-COS. Whereas the func-

conditions. The distinction between conditions in which a

tionsin conditionALT-COS (triangles)are monotonicand

thresholdcouldbe tracked, and thosein which it could not, was clear-cut:for the six trials of any condition,listeners

reacha d' of at least 1.5 for eachlistener,thosefor conditions

couldalwaystrackeithergreaterthanfive,or fewerthan two, thresholds.

In conditionCOS-ALT, psychemetric functionswere additionallymeasuredwith the pink noisereplacedby a different noise,termed "barrier noise,"which had energyattenuatedin frequencyregionscorresponding to the passbands of filters 1 and 2. It was derived from two sources, which were summed. One was a 400-Hz-wide

band of noise

centeredon927Hz andwith a spectrumlevelof 16.2dB SPL (equal to the level of the original pink noiseat 927 Hz), generated by multiplyinga 200-Hz low-pass noiseby a 927Hz sinusold.The other sourcewas the originalpink noise high-pass filtered(48 dB/octave)at 2362 Hz, to maskthe

high-frequency slopeof theexcitation pattern. • C. Results

Thresholdsestimatedfrom the adaptiveprocedureare shownin Table I. Here and throughout,thresholdsare ex-

pressedaspercentage zero-to-peakmodulationdepths;the figuresin parentheses arethe valuesby whichthethresholds shouldbe multipliedand dividedto obtainplus and minus onestandarderror.Whencomparingour resultsto thoseof studiesthat usedstaticmistunings(e.g., Mooreet aL, 1986; Houtsmaand Smurzynski,1990), it is appropriateto consider measuressuchas the rms and maximum mistunings producedby our out-of-phaseFM. Thesemeasurescan be 283

J. Acoust.Sec. Am., Vol. 91, No. 1, January1992

COS-ALT (squares)and ALT (circles) are fairly flat and usuallyhovercloseto chance(d' -- 0}. This is strongevidencethat phaseof the highercomponents is criticalfor the performanceof the presenttask, but that the phaseof the lowercomponents is lessso.This in turn suggests that listeners were detectingdifferencesin F0 conveyedby resolved andunresolvedcomponents. Note that themagnitudeof any beatingbetweenthe two groupsof components, suchasbetweenharmonicssevenandeight,wouldhavebeensimilarin the COS-ALT

and ALT-COS

conditions.

An alternativeexplanationfor the poor performance observedwhen the uppercomponentswere addedin alter-

natingphaseis thattheresultinglow-peakfactorat theoutput of high-frequency auditoryfiltersmight have reduced the detectability of the highercomponents: in contrast,a very peakyoutput may havebeenmoredetectableby "poking up" abovethe noise.This wasthe reasonfor measuring psychemetricfunctionsfor conditionCOS-ALT in barrier noise.The invertedtrianglesin Fig. 4 showthat evenin this reducednoise,performance wasmuchworsethanfor condition ALT-COS in pink noise. III. EXPERIMENT SEPARATION

2: EFFECT

OF FREQUENCY

A. Rationale

Many of thestimuliwhichlistenershaveto groupon the basisof F0, suchas the vowel soundsof speech,are broadCarlyonota/.: DetectingF0 differences

283

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

4

listener: RC

--

listener: HC

listener: AB

3

2

1

--

0

FIG. 4. Psychometricfunctions

-t

4

listener': JD

_. listener: JC

I

l0

40

d'=

0.78, which is the value to

which the adaptiveproceduresof

pink noise

3

for the differentphaseconditions of experiment1. The horizontal dashed line correspondsto experiments 1, 2, 3, and 4 con-

verged.

phose 2

••. '•

................................ 0

-t

1

10

40

- Alow air cos ___._•. high

I

[]

cos

al'J'

0

air

air

bar-r'ier'

•7

10

noise

cos

alt

40

modulation depth (%)

band.Experiment2 investigated thefrequencyrangeacross whichlisteners cancompareF0s, andshiftedthefrequency regionoccupied by the uppercomponents to progressively higherfrequencies. In anattemptto checkfor anydeterioration in the codingof F0 that mightoccurwith increases in frequency region, frequency modulation thresholds (FMTs) for theuppergroupof components weremeasured at eachfilter setting.

oftheupperfilter,andhence withincreases inthefrequency separation between thetwogroupsof components. Thelargest deteriorationis for listenerRC, whosethresholdin-

creases by a factorof 2.4overthefrequency regionstudied, whichrepresents an increase in frequency separation, betweenthe 3-dB downpointsof filters1 and 2, from 750 to 2400 Hz.

50

B. Method

I{stenee:RC

listener-:

HC

Thestimuliwereasdescribed for thesine-phase stimuli in Sec.I B, exceptthat the numberof components in each group(beforefiltering)wasincreased to 50,andthesettings

offilter2 werevaried. Thesettings, chosen tobeequal in

bandwidth on a logarithmicscale,were 1375-1875 Hz,

1788-2438 Hz,2325-3172 Hz,and3025-4125 Hz.Thepink noisebandwidth wasincreased to 10kHz. FMTs werealso 5O

measured foreachof these filtersettings. Theoutputs of

listener-:

JD

DAC 1 andfilter I weredisconnected from the headphone amplifier,andeachtrial consisted of a modulated(signal)

andanumnodulated (standard) version of thehigherhar-

monics. Thestarting phase ofmodulation wasrandomizod from presentation to presentation, asit wasfor the TAFOs, and the stepsizeand adaptiveprocedurewerethe samefor the two typesof measurement.

lower

C. Results

TAFOsareshownforthreelisteners in Fig. 5.Theyshow a moderate increase in threshold with increases in the cutoff

284

I

J. Acoust.Soc.Am.,Vol.91, No.1, January1992

cutoff

of upper filter

(kHz)

FIG. 5. Thresholds(TAFOs) for experiment2 as a functionof the lower cutoffof filter 2. Error barsrepresentplusand minusonestandarderror.

Carlyonet aL:Detecting F0 differences

284

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j

TheFMTs for thehigh-frequency groupof components areshownin Fig. 6, andincrease astheCF of thatgroupis raised.Thismightsuggest thatpartof theincreasein TAFOs wasdueto a degradation in thecodingofF0 at highfrequencies.However,in the next sectionwe presentevidencethat FMTs for two groupsof components do not providea good predictionof the thresholdfor detectinga differencebetweentheir F0s. Hence, this conclusionmay be unjustified. The main result of experiment2 is that TAF0s are fairly robustto increases in frequencyseparationbetweenthe two groupsof components.This finding providesfurther evidencethat listenerswere not performingthe task by monitoringthe outputsof auditoryfilters,tunedbetweenthe two groupsof components, that wereresponding to interactions betweenthe two groups. IV. EXPERIMENT

3: EFFECTS OF SIGNAL-TO-NOISE

RATIO A. Rationale

B. Method

For the main part of the experiment,the sine-phase stimuliof experimentI wereused;whereFMTs weremeasuredthe methodwasasdescribed for experiment2. Four typesof noisewere used,includingthe pink and barrier noisesof experiment 1. In addition, a "high-frequency (HF)" noisewasgeneratedby high-passfilteringthe pink noiseat 927 Hz, anda "low-frequency( LF)" noisewasgeneratedby low-passfilteringthe pink noiseat 927 Hz and addingit to a high-pass-filtered (2362 Hz) versionof itself. In all casesfilteringwasaccomplished usingtwo sections era Kemo VBF/8 filter in series,with a combined attenuation rate of 48 dB/octave. TAF0s were measuredusingall four

typesof noise,FMT•ts weremeasuredin the pink and LF noises,and FMTLs were measuredin the pink and HF noises.The TAF0s in pink noisewere measuredafreshfor this experiment,to control for any effectof the practicelistenershad had sinceexperiment1. In an auxiliaryexperiment,FMTs weremeasuredusing theALT-phasestimuliof experiment1,in a pinknoisebackground.In otherrespects, themethodwasthesameasforthe main experiment.

Experiment1providedaninitialmeasureof thesmallest detectable difference in F0 betweentwo groupsof resolved and unresolvedcomponents.In order to eliminatewithinchannelcuesthestimuliwerepresented in a pink noiseat a C. Results fairly low signal-to-noise ratio (SNR). It is known that the detectionof FM in a singlegroupof components isimpaired The resultsof the main part of experiment3 are shown by noise(Horst, 1989;Carlyonand Stubbs,1989). Experi- in Table II. Comparisonwith the sine-phase datain Table I ment 3 examinedthe extentto which the pink noiseelevated confirmsthat TAFOsin pinknoisehadnot decreased signifiTAFOs by degradingthe codingofF0 changesin eachfrecantlysinceexperiment1. For all fivelistenersTAF0s were quencyregion.The generalapproachwasto removediscrete lower for stimulipresented in barriernoisethan for those frequencyregionsof the pink noisecorresponding to the presented in pinknoise,indicatingthatthepresence of noise passbands of filters1, 2, or both.Thresholdswerethenmea- in thefrequency regionsoccupied bythetwogroupsof comsuredbothfor the detectionof across-frequency F0 differ- ponents didindeedincrease thresholds. However,thedifferences(TAFOs) and for the detectionof frequencymodula- ence in thresholdsbetween the two conditionswas fairly tion of the lower (FMTLs) and higher (FMTns) small,corresponding to a meanfactorof 1.6.It isalsoworth components. notingthat FMTs for the low-frequency resolved groupof harmonics weregenerallylowerthanthosefor thehigh-frequency unresolvedgroup. (Compare the FMTLs and FMTt•s in pinknoise.)Thisisconsistent withearlierdatafor 5O listener: RC the detectionbothof FM and of staticfrequencydifferences (Hoekstra and Ritsma, 1977;Hoekstra, 1979;Horst et al., r

1984; Horst, 1989). TABLE II. FMTs and TAFOsfor the differentnoiseconditionsof experiment 3. Standarderrorsare indicatedin the sameway asin Table I. o

TAFO

I

RC

JD

HC

JC

AB

Mean

Pink LF HF Barrier

5.2(1.1) 4.6(1.2) 6.4(1.1) 3.0(1.1)

6.4(1.2) 5.7(I.I) 4.4(1.3) 3.3(1.1)

6.2(1.2) 10.3(1.1)7.4(I.I) 2.3(I.2)12.2(I.I)4.7(I.2) 4.0(I.2) 9.3(!.1) 5.2(1.1) 3.1(1.2) 7.8(1.3) 6.2(1.1)

6.9 5.1 5.6 4.3

FMT. Pink

9.0(!.2)

7.5(I.I)

1.4(I.!)

4.8(I.I)

3.7(1.1)

4.4

LF

5.4(1.1)

6.3(1.2)

2.2(1.3)

2.5(1.1)

1.6(1.1)

3.1

Pink HF

1.2(1.1) 0.9(1.l)

1.6(1.1) l.l(l.l)

2.3(1.2) 1.4(i.1)

0.9(1.1) 0.8(1.1)

1.4(1.1) 1.1(1.1)

1.4 1.0

41o

E 50 listener:

JO

D_ 10

2.0 3.0 4L.o lower' cutoff of upper filler' (kHz) FIG. 6. Thresholds(FMTs) for experiment2 as a functionof the lower cutoffoffilter2. Error barsrepresent plusandminus1 standarderror.

285

J. Acoust.Sec. Am.,Vol.91, No. 1, January1992

Carlyoneta/.: DetectingF0 differences

285

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j

TableII showsthat therewasno clearrelationship betweenthe effectsof removinga specificnoiseregionon the performance in theFMT andTAF0 tasks.For example,the TAFOfor listenerHC waslowerin thepresence of LF noise thanin HF noise,indicating thatremoving noisein thehighfrequencyregionimprovedperformance morethan did removingthe low-frequency noise.Conversely, her FMT, wasnot reducedby the removalof high-frequency noise, whereasherFMT•: droppedwhenthe low-frequency noise was filtered out. This shows that the relative influence of

noisein the low- and high-frequency regionsis not always

Hz. Thisensured that,evenat a modulation depthof 20%, the highestcomponent wouldalwaysfall at leasthalf an octaveabovethe uppercutoffof filter 2 (27 dB downfrom passband).We did this becausewe wantedthe width of the

excitation patternof the two groupsof components to remainroughlyconstant throughout themodulation. TheF0s used,in Hz, withthenumber ofcomponents in parentheses, were as follows:62.5 (52), 77.5 (42), 100.0 (33), 125.0 (50), 157.5 (21), 197.5(17), and 250.0 (13). In an additional condition,thresholdswere measuredwith a 26-component250-Hz complexandwith thecutoffsof filters1and2

the same for both tasks. Another relevant observation comes

doubled to 250-1250and2750-3750Hz, respectively. Four from the auxiliaryexperiment,whichmeasuredFMTs for listeners (RC, HC, JD, AB) tookpartin all conditions. the uppergroupof components with ALT-phasestimuli. B. Results For thetwo listeners whotookpart (RC andJC), FMT,s for ALT-phasestimuliweresimilarto thosefor sine-phase TAF0sforfourlisteners areplottedasa functionofF0 stimuli,eventhoughtheycouldnotdetectacross-frequency in Fig.7. Thresholds decrease markedly at highF0s:the F0 differenceswhen the upper componentswere in ALT biggestdecrease occursasF0 increases from either 157.5to phase.FMTs for the ALT-phaseand sine-phase stimuli, 197.5Hz (listeners RC, JD ) orfrom 125to 157.5Hz (listenwith standarderrorsin parentheses, were9.5% (1.05) and ersHC and AB). The reductionin thresholdbetweenF0s of 9.0% (1.2) respectivelyfor listenerRC, and 5.0% (1.29) 125and 250 Hz corresponds to a factorof 3.4 for listener and4.8% ( 1.1) for listenerJC. Thusthephaseof theupper RC, 6.1forJD, 4.0forHC, and2.1forAB. Thepatternof components markedlyaffectedthresholdsin the TAFO task, results isconsistent across listeners withminorexceptions. but did not affectFMTs at all. It is alsoworthnotingthat For threelisteners, threshold isroughlyconstant forF0s at listeners' percepts of theALT-phasestimuliin theFMT task and below 125 Hz, but RC's thresholdsdecreaseas F0 is werequalitativelydifferentfrom thosefor the sine-phase reduced from125to62.5Hz. Listener AB'sdatashowlarger stimuli:for the ALT-phasestimuli listenersheard "somestandarderrorsthan thoseof the other listeners,and the thingchanging" duringthesignal(modulated)interval,but smallest threshold reduction at highF0s. the perceptwasnot of a modulationin pitch.For the sineThereduction in threshold at highF0s maybeattributphasestimuli,the perceptwasof a clearlychangingpitch. edto theuppergroupof components becoming resolved by Boththedataandlisteners'introspections fromexperiment theperipheral auditorysystem. F0 estimates basedonresol3 indicate that there are cues available in the detection of FM

that are not useful in tasks, such as that used to measure

TAFOs,that requiretheextractionofF0 information.

V. EXPERIMENT A. Rationale

4: EFFECT OF OVERALL

vable componentsare more accuratethan thosebasedon

unresolvedcomponents(Hoekstra and Ritsma, 1977;

F0

listener: RC

listener': JO

and method 10

Experiments 1-3measured TAF0sforstimuli withan F0 of 125Hz. The aim of experiment 4 wasto examinehow

TAF0s varied withoverall F0,withthepassbands offilters 1

and2heldconstant. Previous research, forexample onthe

;•

identification of melodicintervals(Houtsmaand Smurzynski,1990),suggests thattheencoding of thepitchof a

',

I

25

complex sound ismoreaccurate whenitscomponents are

listener': HC

- listener': AB

resolved thanwhen theyareunresolved. Accordingly, it is possible that athigh F0s,when theupper aswell asthelower

group ofcomponents areresolved, TAF0swould decrease. However,thishasnot beendemonstrated for simultaneous

comparisons asrequired bythepresent task, and itisknown that the discrimination of frequencyrelationships depends onwhetherthecomponents arepresented simultaneously or sequentialy(Demany and Semal, 1990; Demany et al.,

50

1991).

50

100 200

F0 (Hz)

The methodof stimulusgenerationand the procedure werethesameasfor thesine-phase condition of experiment 1,exceptthat sevendifferentF0s wereused,andthenumber of components in eachcomplexwasadjusted for eachF0 so that the highestcarrierfrequencywasalwaysat least3250 286

100 200

J. Acoust.Sec. Am.,Vol. 91, No. 1, January1992

FIG. 7. Threshold asa function of overall F0 for the four listenerswho took

partinexperiment 4. Theunconnected points ontherightofthegraphare fora 250-HzF0 withthepassband offilterI setto250-1250Hz andthatof filter2 setto2500-3500Hz. Errorbarsrepresent plusandminusI standard error.

Carlyoneta/.: DetectingF0 differences

286

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j

HoutsmaandSmurzynski,1990),soa comparison ofF0s in two differentfrequencyregionsshouldbe more accurate when both estimatesare drawn from groupsof resolved components than whenone groupof components is unresolved.However,thedatadescribed sofar do notproveconelusivelythat the thresholdreductionat highF0s is dueto peripheralresolution, ratherthanto someotherattributeof highFOs.To providefurtherevidence on theimportance of peripheralresolution,the 250-Hz F0 conditionwasre-run with thefiltercutofffrequencies doubled,sothat the degree to whichthecomponents wereresolvable wasroughlysimilar to that in theoriginal125-HzF0 condition.The datafor the new conditionare shownin the unconnectedsymbolsto

the right of eachpanelin Fig. 7. They showthat thresholds are at leastashighasin the original125-Hzcondition,supportingthe assertionthat the originalimprovementat high F0s wasdueto increases in theresolutionof thecomponents.

the detection of FM incoherence is better for resolved than

for unresolved harmonics.This predictionof the "FM incoherence"hypothesis wasusedin experiment 5 to distinguish it fromthe assumption underlyingexperiments 1-4, that listenersweredetectinginstantaneous differences in F0. Two conditions were run. In condition 100/225 the low-

er groupof components hada cartierF0 of 100Hz, andthe highergroupa carrierF0 of 225 Hz. In condition100/100, the carrierF0 of both groupswas 100 Hz, and the stimuli weresimilar to thoseof the 100-Hz F0 conditionof experiment4. If listenersweresensitive to across-frequency differencesin F0, thenperformance shouldbeworsein condition 100/225 than in condition100/100, becausethe two groups of componentsare out of tune with eachother on both the standardand signaltrials (cf. Carlyon, 1991). If, however, listenerswere primarily sensitiveto FM incoherence,and coulddetectit independently of anybaseline mistuning, then performanceshouldbe betterin condition100/225 than in condition 100/100, becauseonly in condition 100/225

Vl. EXPERIMENT 5: DETECTION OR FM INCOHERENCE?

OF R

DIFFERENCES

wouldbothgroupsof components beresolved. The assumptionsunderlyingthis predictionare discussed in Sec.VI C.

A. Rationale B. Method

In experiments 1-4, weusedcohercntlyandincoherently modulatedcomplextonesto investigatethe detectionof across-frequency differences in F0. Theassumption underlying our paradigmis that listenersdetectthe F0 differences causedby incoherentFM, ratherthan the FM incoherence perse.This assumption is supported by thedataof Carlyon (1991), whoselistenerscould not detectacross-frequency FM incoherencein inharmonicsounds,and whosepsychemetricfunctionsfor the detectionof a simplemistuningimposedononecomponent era harmonicsoundcouldaccount for the corresponding functionsdescribing the detectionof

The methodof stimulusgenerationwassimilarto that for the sine-phase conditionof experiment1, exceptasfollows:in condition100/225 the outputfrom DAC 1 wasa frequency-modulated 33-component complexwithanF0 of 100Hz, andtheoutputofDAC 2 wasa frequency-modulated 15-componentcomplexwith an F0 of 225 Hz. In the standardinterval,the outputsof the two DACs weremodulatedcoherently; in thesignalintervaltheoutputsweremodulated incoherently (•r modulator delay}. Condition

100/100wasthe sameascondition100/225exceptthat the

carrierF0 of the complexesplayedout throughboth DACs FM incoherence. However, there are differencesbetween the was 100 Hz. Four-pointpsychemetricfunctionswere meastimuli usedhere and thoseusedin the Carlyon ( 1991) experiments,mostnotablythat therearefar morecomponents suredfor eachcondition,with modulationdepthsof 2.5%,

in eachgroupin the presentstudy.Withoutprolongingthe

5.0%, 10%, and 20%.

debate over whether FM incoherence can be detected under

any circumstances, it seemsworth confirmingthat, in the presentstudy,listenerswereindeeddetectinginstantaneous differences in F0 ratherthan in FM coherence per se. There are a numberof findingsfrom experiments1-4 whichsuggest thatourlisteners wereencoding theF0s of the two component groupsand comparing themin someway. Theseincludethedependence of thresholds on thephaseof the uppercomponentsreportedin experiment1, and the reductionin thresholds at high FOSin experiment4. This latterfindingwasattributedto the greaterresolutionof the components at high FOSthan at lowerFOS:as mentioned earlier,comparisons ofFOs aremoreaccuratewith resolved thanwith unresolvedharmonics(HoutsmaandSmurzynski 1990). Unfortunately,the possibilityremainsthat listeners weredetectingdifferences in the way that theFOsof the two groupsof componentschangedover time (FM incoherence), rather than instantaneousdifferencesin F0 (mistuning). In order to maintain an "F1VIincoherence"hypothesis,

onemustassume that differences in performance acrossconditionsarisefrom differencesin sensitivityto FM incoherence;thusonewouldhaveto assumefrom experiment4 that 287

J. Acoust.Sec. Am.,Vol. 91, No. 1, January1992

C. Results

Psychemetricfunctionsfor listenersRC, HC, and JD areshownin thethreepanelsof Fig. 8. The resultsshowthat, whereas in condition 100/100 the functions rise monotoni-

cally with increases in modulationdepthand reacha maximum of at least two for each listener, those in condition

100/225 are much flatter and indicatelower sensitivity.The differencein performancebetweenthe two conditionsis largerfor listenersRC andHC thanfor listenerJD. The resultsof experiment5 confirm that listenersare sensitiveto at leastthe large across-frequency differences in F0 usedin condition100/225;thisF0 difference impairsperformance relative to condition 100/100. The data also show

that FM incoherence doesnot providea sufficientbasisfor detection,at leastin the presenceof a largeF0 difference. However,it couldbe arguedthat listenersweresensitiveto FM incoherence in experiments1-4, but that the largemistuning in condition 100/223 rendered this cue useless.

Whilst agreeingthat experiment5 doesnot provide sufficientevidenceto rulethisoutcompletely,wenotethatlisteners can detectdifferencesin other "grouping"cues,even Carlyoneta/.: DetectingF0 differences

287

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

listener':

RC

listener:

JO

listenec:

HC

condition:

A []

lOO/1OO 100/225

modulationdepth (•)

375 Hz passesonly the third harmonicof the 125-Hz F0. However,similaranalysesof the outputsof filterswith CFs between two adjacentand slightlyhigherharmonics(e.g., harmonics fourandfive) showbeatingbetweentheharmonicsat a rateequalto theseparation betweenadjacentcomponents,which,for thesestimuli,isequalto F0. The beatingis degraded,but not completelyeliminated,by the pink noise. Thuslisteners couldhavecompared therateofbeatingin the filterswith low CFs to that in the filtersresponding to the highergroupof harmonics.Thispossibilityneedsto betaken seriouslybecausethe shapeof the waveformproducedby interactingcomponents isonlymarkedlyinfluencedby their phaserelationship whenthereare morethan two componentsinteracting.Thusphasemighthavebeenrelativelyunimportantat low frequencies not becausethe components wereresolved,but because listenerswereattendingto beating in filtersthat responded to only two components. While stressing thatwecannotruleouta rolefor beatsin

the presentexperiments, we haveonereasonfor thinking that they are not essentialfor the detectionof across-frequencydifferences in F0. Carlyonet al. ( 1991) haveshown

FIG. 8. Psychometric functionsfor conditions100/100 (triangles)and 100/225(squares) ofexperiment 5.Thehorizontal dashed linecorresponds that listeners can detect such differences when the F0 of the tod' = 0.78,whichisthevalueto whichtheadaptive procedures ofexperilowergroupof components isconveyed by onlytheoddharments1, 2, 3, and 4 converged.

when alternativecuesindicatethat the soundsin question shouldbe segregated. For example,listenerscan detectAM incoherencebetweentwo pure tonesthat are well separated in frequency,regardlessof whetherthey are harmonically related (Yost and Shift, 1989; Stricklandet el., 1989), and candetectmistuningof a puretonefrom an otherwiseharmoniccomplex,evenwhenit is presentedto the contralateral ear or with an onsetdisparity(Darwin, 1991;Darwin and Ciocca, 1991). This is not to deny that the effectsof different cues can counteract each other in more central

tasks,suchasthoserequiringlistenersto judgethe number of sourcesheard (Broadbentand Ladefoged,1957), or that thetendencyof soundsto fusemighthavea minoreffecteven on discriminationexperimentslike this one;it is simplyto notethat discriminationof onecueby trainedlistenersis not usuallydrasticallyaffectedby the stateof other, independent,cues.Accordingly,webelievethat thenear-chanceperformancein condition100/225 providesreasonablystrong evidence,in additionto that of Carlyon ( 1991), that listenerswerenot detectingFM incoherencefor the stimuliof the presentstudy. VII. DISCUSSION

A. The roles of amplitude modulationand of pitch in grouping tasks

We interpretedthefindingof experiment1,that TAFOs arenotmarkedlyaffectedby thephaseof thelowergroupof components,as evidencethat thesecomponentswere resolvedby the peripheralauditorysystem.However,thereis an alternativeexplanationfor the relativelysmalleffectof phaseat low frequencies. Figure 3(c) showsthat an auditory filter centeredon 288

J. Acoust.Sec. Am., Vol. 91, No. 1, January1992

monicsof the fundamental.For such stimuli, beatingin filterstunedbetweenadjacentharmonicsoccursat a rate of twice F0, so Carlyon et al.'s resultsindicatethat it is not necessary for adjacentcomponents to beatat a rateequalto F0 in orderforlisteners to detectacross-frequency F0 differences.This conclusionis differentfrom that of Bregmanet al. (1985), whopresented listenerswith complextonesconsistingof two groupsof threecomponents. They reported

that judgmentsof how "decomposed" the two groups soundeddependedon the similarityof their AM rates,and

notof theirF0s or pitches.In theirstudy,theuppergroup alwaysconsisted of components at 1400,1500,and 1600Hz, and thereforehad a pitchand an enveloperepetitionrate (f,•) bothequalto 100Hz. Theyfoundthatwhenthisgroup wascombinedwith oneconsisting of 392, 497, and 602 Hz Oc,•= 105Hz, pitchof about100Hz) theresulting complex soundwas perceivedas decomposed, but that, when it was

combinedwith a groupconsisting of components at 428, 528, and 628 Hz br,• = 100Hz, pitchof about105Hz), a fusedperceptresulted. An importantdifferencebetweenBregmanet al.'sstudy andthat reportedhereis that their stimuliwerepresented at a higherlevel (approx.64- to 74-dB/component)than ours (45 dB/component),and were presentedin quiet whereas ourswerepresentedin pink noise.It is possiblethat performancein their taskwasmediatedby a strongerinteraction (compared to our stimull) betweenindividual components

of eachgroup,or evenby an interactionbetweenthe upper and lower componentsin channelstuned betweenthem. Stricklandet al. (1989) reportedthat detectionof across frequency envelope disparities atf,, = 128Hz wasstrongly affectedby manipulations, suchaspresenting the two carri-. ersto oppositeearsor at very differentlevels,that restricted the use of within-channel

cues. This was true even for fre-

quencyseparationsof the two carrierssimilarto that in the Bregmanet al. study. Carlyoneta/.: DetectingF0 differences

288

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j

B. Effect of phase of lower componentson threshold

posedby Sruloviczand Goldstein (1983). The useof freAlthoughthe phaseof the low-frequency components quencymodulatedstimulimeansthat, if listenerswerecom-

had a muchsmallereffecton performance than did that of thehigh-frequency components, it did havea significant effect for somelisteners(Table I). This difference,between thresholds in thesine-andalternating-phase conditions, was greatestfor thetwo listenerswith thehighestthresholds(JC and AB), and requiresexplanation. It is known that listeners can detect differences in the

phasesof components or of groupsof components that drive different auditory filters, even though their sensitivityto suchacross-channel phasedifferencesis lessthan that to within-channel differences(Patterson, 1987a, 1988). How-

ever, differences in across-channel phaseare, presumably, perceivedasdifferences in timbre,ratherthan in pitch,soit is not obvioushow they wouldaffectthe presenttask.One possibilityis that putting the lower componentsin ALT phaseintroducedan additionaltimbre differencebetween themand the uppercomponents, therebyimpairinglisteners'abilityto fusethe two groups.As the experimentaltask canbe viewedas oneof "groupingversusseparation,"this could have impaired performanceslightly. Although we haveno definiteevidencefor suchan explanation,JC and AB'sperformance did improvewhentheadaptiveprocedure waschangedto a constant-stimulus one:the thresholdsestimatedfrom their psychometricfunctionswere 11.6% (listenerJC) and7.5% (AB), comparedto adaptivethresholds of 19.4% and 24.4%. For the other three listeners, who were

much lessaffectedby the phaseof the lower components, performance waslessaffectedby the changein procedure.It is possiblethat, in the slightlyeasierparadigm,listenersJC andAB learntto attendto theF0s of bothgroupsof components,andto ignoreirrelevantaspects of thestimulussuchas

timbredifferences. Thusit maybethatwhereasthephaseof the upper components affectspitch strength,and severely degradesperformancefor all listenerswhateverthe paradigm,that of the lowercomponents affectstimbreandonly degrades performance moderately,andonlyfor somelisteners in someparadigms. C. Implications for models of pitch perception

The data presentedhereindicatethat listenerscan detect differencesbetweenthe F0s of two groupsof componentsoccupyingdifferentfrequencyregions,underconditions where only one group is resolvedby the peripheral auditorysystem.The effectsof varyingthe phaseof the differentcomponentgroups(experiment1), and of changing the degreeby which the uppercomponentsare resolvedby

the peripheralauditorysystem(experiment4), are consistent with thoseof similarmanipulationsin pitch perception tasks(Plomp, 1964;MooreetaL, 1986;HoutsmaandSmurzynski, 1990). Our data are, to a first approximation,consis-

tent with modelsof pitchperceptionthat derivepitchestimatesfrombothresolvedandunresolved components. Such modelsinclude the "crude sketch" describedby Moore (1989), "autocorrelogram" schemes (Licklider, 1951; Assmann and Summerfield, 1990; Meddis and Hewitt,

1991a), Patterson's( 1987a,b,1991) "pulseribbon,"and,as Houtsmaand Smurzynskihavepointedout, the modelpro289

J. Acoust.Soc. Am., VoL91, No. 1, January1992

paring the outputsof two separatemechanisms--apure "pattern recognizer" (e.g., Goldstein, 1973; Terhardt, 1974) and a purelytemporalmechanism(Schouten,1940, 1970), then the outputsof thesetwo separatemechanisms wouldhaveto be rapidlyconvertedto a form in whichthey couldbecompared.However,althoughit is parsimonious to concludethat listenerswereanalyzingthe outputof a single pitch mechanism,we note that the brain is capableof very fastconversions betweendifferenttypesof sensoryinformation. These include the combination

of interaural

time and

intensitycuesin auditorylateralization( SternandColburn, 1978), and the combinationof auditoryand visualinformation in the perceptionof placeof articulation(McGurk and MacDonald, 1976). Thus thestrongestnewconclusionto be drawnfromour experiments is that,if therearetwo separate pitch mechanisms,their outputsmustbe rapidly converted to a mutuallycomparableform. Sucha conclusioncouldnot havebeendrawnfrom experiments usingsteadysounds,in

whichlistenerscouldanalysethe pitchof eachcomplexin turn, and convert these measures to a common metric

(Schouten,1940;Plomp, 1967;Ritsma,1967;Moore, 1977; HoutsmaandSmurzynski,1990). Note that sucha strategy is possible evenwith simultaneous tonesprovidedthat they havea constantF0, in whichcaselistenerscanprocess each pitchin turn by switchingtheir attentionfromonecomplex to the other [e.g., Broadbentand Ladefoged,1957 (exp. 2); seeDemany and Semal, 1990]. Not all aspectsof the presentfindingsare consistent with the autocorrelogram models.It is knownthat, comparedto high (unresolved)harmonics, low (resolved)harmonicsproduce stronger pitchpercepts, lowerF0 discriminationthresholds,and are dominantfor the perceptionof pitch(Plomp,1967;HoekstraandRitsma,1977;Hoekstra, 1979;Moore eta!., 1985;Houtsmaand Smurzynski,1990). Meddis and Hewitt's (1991a) model accounts for the greatercontributionto pitchperception of thelow harmonicsby the greaterchanneldensityat low frequencies. However, the resultsof experiment4 indicatethat TAF0s drop markedlywhenF0 is increasedsothat bothgroupsof components areresolved, evenwhenthefrequency regionof the component groups(passbands of filters1 and 2) are held constant. Thisindicates animprovement in thecodingofF0 that cannotbeattributedto an increasein channeldensity.If thegreaterdominance, pitchstrength,anddiscriminability of resolvedcomponents arisefrom the sameauditoryprocess,then experiment4 indicatesthat they are inconsistent withtheexplanation offeredby MeddisandHewitt'smodel. D. Implications for perceptual sound segregation

Carlyon and Stubbs(1989) comparedFMTs for harmonic sounds,inharmonicsounds,and "partially harmonic" soundsthat eachcontaineda subsetof harmonicallyrelated components spanning a (different) restricted frequencyregion.They reportedthat, when the stimuli were

presentedin burstsof pink noise,FMTs were lower for the harmonicthanfor eithertheinharmonicor thepartiallyharmonic sounds.They suggested that listenerscan combine Carlyonot at'.:DetectingF0 differences

289

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

information on the F0 of the harmonic stimuli across a wide

de Cheveign•,J. Culling, C. Darwin, A. Q. Summerfield,

rangeof frequencycomponents, andthat their listenersused thisF0 informationto extractthesignalfromthepinknoise. The resultspresented hereindicatethat listenerscanindeed extract such across-frequencyinformation, which may be usedto perceptuallyseparateconcurrentcomplexsounds.

and an anonymousreviewer for useful discussionsand/or

The practicalimplications of theabilityto compareF0s derivedfrom resolvedand unresolved components depend at leastpartiallyon the accuracywith whichlistenerscando this.In orderto ruleout peripheralinteractions,our stimuli werepresented in pink noiseat a fairly low SNR. In reallife the SNR is likely to be higher,at leastoverpart of the frequencyrange(few noisesourcesmaskasuniformlyasdoes pink noise). Our best estimatethereforecomesfrom the

commentsonpreviousversionsof thisarticle,andP. Russell for programming andsignal-processing advice. According to Srulovicz andGoldstein's ( 1983) model,thefrequencies of

translatesinto an rms mistuningof 6.1%. Thesevaluesare clearlylarger than thosefor detectingmistuningof single resolvedcomponents (Moore et al., 1986;Demanyet al.,

individualharmonicsarederivedby passingthe interspikeintervalhistogram (ISIH) of eachauditorynervefiberthrougha filter matchedto the CF of that fiber.In effect,the ISIH is multipliedby a squaredcosinewith peaksat integermultiplesof 1/CF, andintegrated. Theoutputofthefilters matchedto eachCF formtheinputto a "centralspectrum,"fromwhich theœ0isestimatedbya patternrecognition process (harmonictemplates). SruloviczandGoldstein(1983) concludedthat "discriminationof periodicitypitchwith closely-spaced harmonics...involves a completelydifferent and muchlessefficientmechanismthan frequencypatternrecognition" (p. 1274).However,in pointingoutthatthemodelcanin factderiveF0s from unresolvedharmonics,Houtsmaand Smurzynski(1990) notethat, for fibersresponding to unresolved harmonics,therewill be a peakin the ISIH at a periodcorrespondingto F0. This peak will be passedonly by filtersmatchedto harmonicsof thatF0 (whichhaveperiodsequalto submultiplesof 1/FO), and, accordingly,therewill be peaksin the "central spectrum"at frequencies corresponding to unresolved(as well as to re-

1991), which are between 1% and 2% (d' = 1). However,

solved) harmonics.

TAFOs measuredin barrier noise,where the SNR was most favorable, and whose mean acrosslisteners was 4.3%. This

Thespeech synthesizer usedgenerated a formantof frequency F kHz by

this doesnot necessarily meanthat across-frequency commultiplyinga periodicsourceby a 10-kHz sinusold,filteringthe resultant, parisonsarenot usefulfor theprocessing of concurrentcomand thenmultiplyingit by anothersinusoldof frequency( 10 + F) kHz. This resultedin a formantwith centerfrequencyFand slopesequal (in plex sounds:thresholdsfor the detectionof mistuningbedB/Hz) to thehigh-frequency rolloffoftheoriginalperiodicsource.Only tweencomponents whichare resolved from eachotherwill a roughsketchof this sourceis given,resembling a repeatedexponential beelevatedby the presence of competingsound,whoselowthat had decayedto about 11% of its initial (maximum) valueby the end frequencycomponents may have frequencies very closeto of eachperiod.We synthesized the sourceusinga periodof 125Hz, and thoseof the target.Under suchcircumstances, across-fre- calculatedits powerspectrum.This had a low-passcharacteristicthat droppedby 20dBwithin500Hz, andby 26 dBwithin1000Hz, of its0-Hz quencycomparisons basedon groupsof componentsmay maximum. aid segregation. 3Theattenuation of filter 1 re:thelevelin its passbands, withthecorreThe sizeof our TAF0s are roughlyconsistent with the spondingfrequencyin parentheses, is as follows:- 12.9dB (750 Hz), --23.4dB (875 Hz), --32.6dB (1000 Hz), --41.1 dB (1125 Hz), data of Culling (1990). He measuredthe identificationof -- 48.8dB ( 1250Hz). For filter2, corresponding measures were - 42 dB pairsof simultaneous vowels,and foundthat F0 differences (750Hz), - 30.3dB (875Hz), -- 21.2dB (1000Hz), -- 13.3dB (1125 betweendifferentformantsof the samevowelimpairedidenHz), -- 6.7 dB ( 1250 Hz), -- 5.2 dB (2000 Hz), -- 8.6 dB (2125 Hz), tification slightly, but only when the F0 differencewas -- 12.3 dB (2250 Hz), -- 16 dB (2375 Hz), -- 19.5 dB (2500 Hz), greater than one semitone(5.9%). Our data indicate that

- 22.9 dB (2625 Hz).

4Whenmanipulating thephase ofthecomponents ofa complex soundit is

smaller F0 differences would not have been detectable.

necessary to takeinto accountthe phaseresponse of the headphones, and alsothat of other analogequipmentused,suchas filters.Thereforethe waveforms usedin thesimulations of Fig. 3 wereobtainedby playingthe stimulithroughthesameequipmentasusedfor theexperiments, andmeasuringthe outputof the headphones with a B&K condenser microphone mountedin a B&K artificialear (seeSec.I B for details).The microphone outputwascapturedinstantlyusingoneof the ADC channelsof the CED 1401laboratoryinterfacethat alsoincorporatedthe DACs. The waveformswerethenprocessed offlineusingan implementation of the gamma-

E. Conclusions

The resultspresentedpreviouslyshowthat listenerscan detectacross-frequency differencesin F0. Performanceis onlymoderatelyaffectedby increases in the frequencyseparationbetweenthetwogroupsof components, changes in the phaseof the lower (resolved)components, and the presence tone auditory filter modeldescribedby Pattersonet al. (1988). The resultingsimulations werealmostidenticalto similaronesapplieddirectlyto the of noisein the frequencyregionoccupiedby the two groups calculatedwaveforms of thestimuli.The reasonthat theanalogequipment of components.The data are consistentwith, but do not didnotmarkedlyaffectthephasemanipulations usedhereisthatthephase prove,modelsof pitchperception whichproposethat pitch responses of mostpiecesof equipment changesmoothlyacrossfrequency whereasthe alternatingphaseusedhere is extractedfrom bothresolvedandunresolved components (exceptnearsharpresonances), resultedin abruptphasechangesbetweenadjacentcomponents. by a commonmechanism.It is suggested that the combina• Althoughthewidthoftheexcitation patternof thestimuliremained aption ofF0 informationacrossa widefrequencyregionmay be proximately constantthroughoutthe modulation, at the lowestF0s proa usefulstrategyfor the perceptualseparationof concurrent ducedduringthe largestmodulations, therewereinsufficient components harmonic

to fill the upperskirtsof filter 2. The high-frequency portionof the noise maskedcomponentsin the highestfrequencyregionsof theseskirts.

sounds.

ACKNOWLEDGMENTS

ANSI (1969). ANSI S3.6-1969,"Specifications for audiometers"( Ameri-

This researchwassupportedby a Royal SocietyUniversity ResearchFellowship and by a project grant from the Medical ResearchCouncil (U.K.) to the first author. The

Assmann,P. F., and Summerfield,Q. (1989). "Modeling the perceptionof concurrent vowels:Vowels with the same fundamental frequency," J.

can National Standards Institute, New York).

second author is affiliated with the Centre National

de la

RechercheScientifique. We wouldlike to thankE. Burns,A. 290

J. Acoust.Soc. Am., Vol. 91, No. 1, January 1992

Acoust. Soc. Am. 85, 327-338.

Assmann,P. F., andSummerfield, Q. (1990). "Modelingthe perceptionof concurrentvowels:Vowelswith differentfundamentalfrequencies,"J. Acoust. Soc. Am. 88, 680-697.

CarlyonetaL: DetectingF0 differences

290

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j

Bregman,A. S., Abramson, J., Doehring,P., and Darwin,C. I. (1985). "Spectralintegration basedon commonamplitude modulation," Percept.Psychophys. 37, 483-493. Broadbent, D. E., andLadefoged, P. (1957). "On thefusionof sounds

reaching different sense organs," 1. Acoust. Soc.Am.29,708-710. Brokx,I. P. L., andNooteboom, S.G. (1982)."Intonation andtheperceptualseparation ofsimultaneous voices," I. Phonetics 10,23-36. Bums,E. M., andViemeister, N. F. (1981). "PlayedagainSAM: Further observations onthepitchofamplitude-modulated noise," J.Acoust. Soc.

cation,"J. Acoust.Soc.Am. 89, 2866-2882.

Meddis,R., andHewitt, M. (1991b). "Virtual pitchandphasesensitivity

studied using acomputer modeloftheauditory periphery: Phase sensitivity," J. Acoust.Soc.Am. 89, 2883-2894. Meddis,R., and Hewitt, M. (1991e). "Modelingthe identificationof concurrentvowelswith differentfundamental frequencies," J. Aeoust.Soc. Am. {to bepublished). Moore,B.C. J. ( 19771."Effectsof therelativephaseof thecomponents on

thepitchofthree-component complex tones," inPsychophysicsandPhysiology of Hearing,editedby E. F. EvansandJ.P. Wilson(Academic,

Am. 70, 1655-1660.

London,England). Carlyon, R. P. (1991)."Discriminating between coherent andincoherent Moore,B.C. I. (1989).An Introduction tothePsychology ofHearing(Acafrequency modulation ofcomplex tones," 1.Acoust. Soc.Am.89,329demic, New York), 3rd ed.

340.

Carlyon, R. P., Demany, L., andSemal, C. (1991)."Therolesof pitch differencesand modulation incobemneein concurrent sound segrega-

Moore,B.C.J.,andGlasberg, B.R. ( 1983). "Suggested formulae forcalculatingauditory-filter bandwidths andexcitationpatterns," J. Acoust.

tion," 1. Acoust.Soc.Am. 89, 1912{A).

Soc.Am. 74, 750-753.

J. Acoust. Soc. Am. 85, 2563-2574.

Aeoust. Soc. Am. 77, 1853-1860.

B.R.,andPeters, R. W. (1985)."Relative domiCarlyon, R.P.,andStubbs, R. 1.(1989)."Detecting single-cycle frequency Moore,B.C.J.,Glasberg, nance ofindividual partials indetermining thepitchofcomplex tones," J. modulation imposed onsinusoidal, harmonic, andinharmonic carders," B.R.,andPeters, R. W. (1986)."Thresholds for Chalikia,M. H., andBregman, A. S. (1989)."Theperceptual segregation Moore,B.C.J.,Glasberg, hearing mistuned partials asseparate tonesin harmonic complexes," J. of simultaneous auditorysignals: Pulsetrainsegregation andvowelsegAcoust. Soc. Am. 80, 479-483. regation," Percept. Psychophys. 46,487-496. T. W. (1976)."Separation of speech frominterfering speech by Culling, J.(1990)."Perceptual separation ofconcurrent vowels: Effects of Parsons, meansof harmonicselection,"J. Acoust.Soc.Am. 60, 911-918.

across-formant inconsistencies in F0," Brit. 1. Audiol. 24, 194.

R. D. (1987a)."A pulseribbonmodelof mortaural phaseperDarwin, C.J. (1981)."Perceptual grouping ofspeech components differing Patterson, ception," J. Acoust. Soc. Am. 82, 1560-1586. infundamental frequency andonset time,"Q.J.Exp.Psychol. 33A,185-

Patterson, R. D. (1987b)."A pulseribbonmodelof peripheral auditory

287.

processing," in.4uditory Processing ofComplex Sounds, edited byW.A.

Darwin,C. I. (1991). "Pitchof dieboriccomplex toneswitha mistuned

frequency component," inAuditory Physiology andPerception edited by Y. Cazals, L.Demany, andK. Homer(Pergamon, Oxford, England) (in

Yost and C. S. Watson (Erlbaum, Hillsdale, New Jersey).

Patterson, R. D. (1988)."Timbrecuesin mortaural phase perception: dis-

tinguishing within-channel cues andbetween-channel cues," inBasic Issuesin Hearing, edited byH. Duifhuis, J.W. Horst,andH. P.Wit (Aca-

press).

Darwin,C. J.,andCiocca, V. (1991)."Grouping in pitchperception: Effectsofonsetasynchrony andearof presentation ofa mistuned component,"J. Acoust.Soc.Am. (submitted).

demic,London,England).

Patterson, R. D., Nimmo-Smith, I., Holdsworth, J., andRice,P. (1988).

Demany,L., andSemal, C. (1988)."Dichoticfusion of twotones oneoc-

taveapart: Evidence forinternal octave templates," J.Acoust. Soc.Am. 83, 687-695.

Demany, L., andSemal,C. (1990)."Harmonic andmelodic octavetemplates," J. Acoust. Soc.Am.88,2126-2135.

Demany, L., Semal, C., andCarlyon, R. P. (1991)."Ontheperceptual limitsof octaveharmony andtheirorigin,"1. Aconst. Soc.Am. (to be published).

Gardner,R. G., Gaskill,S.,andDarwin,C. J. {1989)."Perceptual group-

ingofformants withstatic anddynamic differences infundamental frequency," J. Acoust. Soc.Am.85, 1329-1237. Goldstein, J.L. {1973)."Anoptimum processor theory forthecentral formationof thepitchof complex tones," 1. Acoust. Soc.Am.54, 14961516.

Hoekstra, A. (1979}."Frequency discrimination andfrequency analysis in hearing," Ph.D.thesis, Institute ofAudiology, University Hospital, Groningen,Netherlands.

Hockstra, A., andRitsma, R. J. (1977)."Perceptive hearing lossandfre-

quency selectivity," inPsychophysics andPhysiology ofHearing, edited by

"Spiral VosFinalReport, PartA: TheAuditory Filterhank," Applied Psychology Unit,Cambridge, England. Contract Rep.APU2341. Patterson, R. D., Robinson, K., Holdsworth, J.,McKeown,D., Zhang,C., Allerrand,M. (1991)."Complex sounds andauditory images," inAudi-

toryPhysiology andPerception, edited byY. Carols, L. Demany, andK. Homer (Pergamon,Oxford,England)(in press).

Piszczalski, M., andGaller,B. A. (1979)."Predicting musical pitchfrom

component frequency ratios," J.Acoust. Soc.Am.66,710-720. Plomp, R. (1964)."Theearasa frequency analyser," J.Acoust. Soc.Am. 36, 1628-1636.

Plomp, R. (1967)."Pitchofcomplex tones," J.Aeonst. Soc.Am.41,15261533.

Ritsma, R.J. (1967)."Frequencies dominant intheperception ofthepitch ofcomplex sounds," I. Acoust. Soc.Am.42, 191-198. Scheffers, M. T. M. { 1983). "Siftingvowels:auditorypitchanalysis and

sound segregation," Doctoral dissertation, Univ.of Groningen, The Netherlands.

Schouten, J.F. (1940)."Theresidue andthemechanism ofhearing," Proc. Kon. Akad. Wetenschap. 43, 991-999.

Schouten, J. F. (1970)."Theresidue revisited," inFrequency Analysis and Periodicity Detection inHearing, edited byR. Plomp andG. F. SmoorenHorst,J.W. (1989)."Detection anddiscrimination offrequency modulaburg(Sijthoff,Lieden). tionofcomplex signals," 1. Acoust. Soc.Am.85,2022-2030. M.,andLyon,R.F. (1990)."A perceptual pitchdetector," Proc. Horst,J.W.,Ritsma, R.J.,andWit,H. P. (1984)."Frequency discrimina- Sinhey, E. F. EvansandJ.P. Wilson(Academic,London,England).

tioninquietandin noise forsignals withtriangular spectral envelopes," J. Acoust. Soc. Am. 76, 1067-1075.

Houtsma, A. •. M, andSmurzynski, J. (1990)."J.F. Schouten revisited: Pitchofcomplex tones having manyhigh-order harmonics," J.Acoust. Soc. Am. 87, 304-310.

Lawrence, W. (1953). "Thesynthesis of speech fromsignals whmhhavea lowinformation rate,"in Communication Theory, editedbyW. Jackson

lnt. Conf.Acoust.Speech, andSig.Process. 357-360. Srulovicz, P., andGoldstein, J. L. (1983)."A centralspectrum model:A

synthesis ofauditory-nerve timing andplace cues inmonaural communication offrequency spectrum," J.Acoust. Soc.Am.73,1266-1276. Stern,R. S.,andColburn, H. S. (1978)."Theoryofbinaural interaction based onauditory-nerve data.IV. A modelfor subjective lateralposition," J. Acoust.Soc.Am. 64, 127-140.

Strickland, E., Viemeister, N. F., Fantini,D. A., andGarrison,M. A. (1989)."Within-versus cross-channel mechanisms indetection ofenveLevitt,H. (1971)."Transformed up-down methods in psychophysics," J. (Butterworths,London,England).

Acoust. Soc. Am. 49, 467447.

Licklider,J. C. R. (1951)-"A duplextheoryof pitchperception," Experientia 7, 128-133.

MeAdams, S.(1984)."Spectral fusion, spectral parsing andtheformation of theauditoryimage,"Ph.D thesis,StanfordUniversity.

McGurk,H., andMacDonald, J. (1976)."Hearing lipsandseeing voices,"

lopephase disparity," 1.Acoust. Soc.Am.86,2160-2166.

Stubbs, R. 1.,andSummerfield, Q. (1988)."Evaluation oftwovoice-sepa-

rationalgorithms using normally-hearing andhearing-impaired listeners,"J. Acoust.Soc.Am. 84, 1236-1249.

Summerfield, Q.. andAssmann. P. F. (1991). "Perception of concurrent vowels: Effects of harmonic misalignment andpitch-pulse asynchrony,"

Nature 264, 746-748.

J. Acoust. Soc. Am. 89, 1364-1377.

studied using acomputer model oftheauditory periphew: Pitchidentifi-

Am. 55, 1061-1069.

andharmony," J. Acoust.Soc. Meddis, R., andHewitt,M. {1991a)."Virtualpitchandphase sensitivity Terhardt,E. (1974)."Pitch,consonance,

291

J.Acoust. Soc.Am., Vol.91,No.1,January 1992

Carlyon etal.:Detecting F0differences 291

ownloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.js

Terhardt, E., Stoll,G., and Seewann,M. (1982). "Algorithm for extraction of pitch salleneefrom complextonal signals,"J. Aeoust.Soe.Am. 71, 679-688.

Warren,R. M., andWrightson,J. M. (1981). "Stimuliproducingconflicting temporaland spectralcuesto frequency,"J. Aeoust.Soc.Am. 70, 1020-1024.

292

J. Acoust. Soc.Am.,Vol.91,No.1,January 1992

Yost, W. A., and Sheft,S.S. (1989). "Across-critical-band processing of amplitude-modulated tones,"J. Aeoust.Soe.Am. 85, 848-857. Zwicker,E. (1952). "Die GrenzenderHorbarkeitderAmplitudenmodulation und der Frequenzmodulation einesTones,"Acustiea2, 125-133. Zwicker,U. T. (1984). "Auditoryrecognitionof dioticanddichotievowel pairs,"SpeechCommun.3, 265-277.

Carlyon eta/.:Detecting F0 differences

292

wnloaded 28 Apr 2011 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j