Perceptual learning in frequency discrimination Laurent Demany LaboratoiredePsychologie Expdrimentale, /lssocid au C. N. R. •, Universitd Ren• Descartes et E. P. H. E. section3, 28 rue Serpente,7õ006Paris,France
(Received12September 1984;accepted for publication 4 June1985) This studywasconcernedwith the effectsof training on the frequencydiscriminationability of humanlisteners.Frequencydiscriminationat 200 Hz wastestedbeforeand aftertraining.Four groupsof listeners receivedtrainingin fourdifferentfrequencyregions,200, 360,2500,and6000 Hz. It wasfoundthattrainingat 200, 360,and2500Hz all providedcomparable improvement in discrimination performance at 200 Hz whereastrainingat 6000Hz providedlessimprovement. This resultis consistent with the ideathat frequencydiscrimination and pitchperceptionare mediatedby differentprocesses at high( > 5000Hz) andlow ( < 5000Hz) frequencies.
PACS numbers:43.66.Fe,43.66.Hg
INTRODUCTION
The 12blocksfor eachsubjectwerepresentedin two 75min sessions, separatedby a meantime of 49.0 h. Each trial consisted of two pairsof tones.Subjects knewthat thesecond
Many investigators of the humanabilityto detectsmall differences in frequencybetweensinetoneshavenotedthat toneof oneof the two pairswouldbe higherin frequency; thisabilitycouldbemarkedlyimprovedwith practice(Wytheyhad to pressa buttonindicatingwhichpair endedwith att, 1945;Harris, 1952;Campbelland Small, 1963;Moore, 1973, 1976;Turner and Nelson, 1982).Weeksof practice thetoneof higherpitch.The highertoneoccurredrandomly andthousands of trialsmaybenecessary to reacha subjeet's in the firstor secondpair and its frequencyvariedwithin a block.In eachblock,individualdiscrimination thresholds asymptotic discrimination threshold. However,verylittleis were determined by means of the classic staircase method knownaboutthe learningprocessinvolved.Oneof the most (Levitt, 1971)withoutfeedback. basicquestionsto ask about this processconcernsits freThe pre-trainingtestandthepost-training testeachconquencyspecificity: Doespracticein a givenfrequencyregion tained 50 trials and were identical in every aspect. The initial improvediscrimination performance onlyin thatfrequency frequencydifference betweenthe highertoneand the three regionor alsoin remoteregions? Data presented in an unidentical standard tones was set at 100 Hz. This difference publishedreportby Heimer and Tatz (1966)suggestthat until the subjectmade frequency discrimination learninghassomefrequency speci- washalvedfor eachcorrectresponse the first error; thereafter,it washalvedafter two consecutive ficity.However,HeimerandTatz did not explorethisspecicorrectresponses and doubledfor eachincorrectresponse. ficity in detailand it seemsthat no oneelsehasexploredit The first two reversals of the staircasewereignoredand for with modernpsychophysieal methods. {an even number of} the remainingreversalpointsthe freThe goalof the presentstudywasto determinewhether quency differences were averaged to find the subject'sdisthelearningof frequency discrimination isfrequencyspecifcrimination threshold. The procedure usedin the training ic by parametricallyvaryingthe differencebetweenthe freblocks was identical to that used in the pre-training andpostquencyrangeof the trainingand the frequencyrangeof the training tests except for the following points: (1) these blocks test. contained70 trials;(2)followingthe firstincorrectresponse, the frequencyincrementwasmultipliedby factorsof 2/3 or 3/2; (3} initial frequencydifferencesfor standardtonesof 360, 2500, and 6000 Hz were,respectively,180, 1250,and I. METHOD
1500 Hz.
The experimentwasconductedin a sound-treated stuSeventysubjects,without known hearingproblems, dio at the Institut de Recherehe et Coordination Acoustibetween theagesof 19and36,weredividedintofourgroups que/Musique.Stimuliweregeneratedby a 16-bitdigitalsynbalancedby age.All subjectsperformeda pre-trainingtest {block1}anda post-training test{block12}in thediscrimina- thesizer,controlledby a computerwhichalsocollectedthe tion of sine tonesnear 200 Hz. The subjectsin different response data.The two tonesof eachpair had a durationof groupsreceived10 trainingblocksin four frequencyranges 400 ms (plus20 ms of riseand fall time)and were separated by 200 ms.The separationof the pairswithin eachtrial was asfollows:16subjects at 200 Hz (groupT200), 16 subjects at intervalwassubjectcontrolled.The 360Hz (groupT360), 16subjects at 2500Hz (groupT2500}, 500msandtheresponse and 22 subjectsat 6000 Hz (groupT6000}. Twenty other threestandardtoneswerepresented at 35 phons;thevariable subjectswere eliminated because their discrimination frequencytonehadthesameSPL asthestandardtones.Subjectslistenedwith the right ear to the tonesvia an earphone thresholdson the pre-trainingtest were lessthan 2 Hz, with flat frequencyresponsein both the test and training whichpromisedlittle chanceof significantimprovement,or becausetheir thresholdswere greaterthan 25 Hz. regions. 1 1118
J. Acoust.Soc. Am. 78(3), Sept. 1985; 0001-4966/85/091118-03500.80;
¸ 1985 Acoust. Soc. Am.; Letters to the Editor
1118
ownloaded 29 Jun 2010 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j
1.4
f.• 12 I
Figure3 showsthem•an slopeof thelinesin Fig. 2 for eachof the four groups.Very similarmeanslopevaluesare obtainedfor trainingat 200, 360,or 2500Hz, but thereis a sharpdrop in meanslopebetween2500 and 6000 Hz. An analysisof varianceconfirmsthat the meanslopesobtained for groupsT200,T360, andT2500donotsignificantly differ
fromeachother[F(2,45)< 1]but,whentakentogether, are reliablydifferentfrom the meanslopefor groupT6000 [F
0,68)= 6.34,P<0.02].2 IlL DISCUSSION
.4
FIG. 1.Merli•nvalueof thestatistic givenby thediscrimination threshold onexperimental blockn relativeto thethreshold on thefirsttrainingblock (block2},foreachgroupandforn = 2,•..... 1I. Solidcurve:groupT200; dashed curve:groupT360;dot-dashed curve:groupT2500;,dottedcurve:
groupT6000.ForthegroupT200,trainedinthefrequency rangeofthetwo t•t blocks,thedataobtainedfor n = I andn = 12arealsoshown.The gap
between blocks6 and7 reflects theseparation between thetwosessions.
II. RESULTS
Let us firstconsiderthe data obtainedduring the train-
ingperiod,startingwithblock2 andendingwith block11. Block-2medianthresholds were:2.8 Hz (groupT200},3.4 Hz {groupT360), 18.6 Hz (groupT2500), and 138.0 Hz (groupT6000).FigureI showsthemedianvalueof thestatistic givenby thethresholdonblockn relativeto thethreshold on block2 for eachgroup.It canbeseenthat the thresholds showa generallydecreasing trend with negativeacceleration, an expectedresultfor perceptuallearning{Gibson,
The data shownin Figs.2 and 3 suggestthat trainingat 6000Hz doesnot facilitatefrequencydiscrimination at 200 Hz as effectivelyas doestrainingat 200, 360, or 2500 Hz. This differenceis unlikelyto be the resultof systematicvariationsamongthe subjects because performance on the pretrainingtestwasessentially identicalfor all fourgroups(see Fig. 2}.
Onepossible explanation for thedifference isthatduringtrainingat 6000Hz, but not theotherfrequencies, the subjects basedtheirjudgments uponloudness changes rather thanpitchchanges (of.Coninx,1978).An alternativeand moreplausible explanation isthattheperceptual learningof frequency {i.e.,pitch)discrimination is to someextentfrequencyspecific. Selective learningeffectsobserved in the visualsystem, for stereopsis (Ramachandran andBraddick, 1973}andfor gratingacuity{FiorentiniandBerardi,1981), havebeenattributedto a separation of the neuralchannels involvedin trainingandtest.In the presentcase,however,
tonesof 200 and2500 Hz are certainlyprocessed in different neuralchannelsand yet trainingat 2500 Hz facilitatedperformante in the test at 200 Hz. Evidently, the perceptual 1969}. learningof frequencydiscrimination canbetransferredfrom Figures2 and 3 are the primary focusof this paper. one neuralchannelto another.To accountfor the markedly Figure2 showsthesubjects' individualthresholds in thetwo smallerfacilitationof the testtaskby trainingat 6000Hz, it testsat 200Hz. In thisfigure,theslopeof thelinesrepresents maybe proposed that frequency discrimination at low frelog {Aft/Aft2);takingthe logarithmof the ratio resultsin quencies, 2500Hz andbelow,andfrequency discrimination groupdistributions whicharemorenearlynormalthanthe at highfrequencies, e.g.6000Hz, are mediatedby qualitadistributions fortheratio,thedifference, or thelogarithmof tivelydifferentprocesses. the difference.For groupsT200, T360, and T2500 altoOn thebasisof physiological dataonneuralsynchrony gether,thereisoneascending lineoutof48;forgroupT6000, at the levelof the auditorynerve(Roseetal., 1967;Johnson, in contrast,thereare fiveascending linesout of 22.
GaOUP
GROUP T3•O
GROUP T2500
GROUP T6000
.5
'3' A
.3
.2
.1
i
i
t
200
360
2500
6000
Training frequency in Hz (log scale)
FIG. 2. Individualdiscrimination thresholds in thepre-trainingtest(block
1}andthepost-training test(block12}.Foreachsubject, a straight lineconneeisthe pre-trainingtestthresholdon the left to the post-training test thre•old on the right. 1119
J. ^coust. Soc. Am., VoL 78, No. 3, September 1985
FIG. 3. Averagevaluesoftheslopeofthelinesin Fig.2, indicatingtheeffect of discrimination trainingin frequency rangesgivenalongthe horizontal axis.The verticalbarsaretwostandarderrorsin overalllength. Letters to the Editor
1119
wnloaded 29 Jun 2010 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.
F.,andOlson,R. K. (1971)."Pitchasamedium: A newapproach 1980),it canbehypothesized thatfrequency discrimination Attneave, to psychophysical scaling,"Am. J. Psychol.84, 147-166. ismediated mainlybya temporalcodingprocess upto 5000 Bachem,A. (1954)."Timefactorsin relativeandabsolutepitchdeterminaHz and exclusively by a tonotopicalcodingprocess beyond tion," J. Acoust. Soc. Am. 26, 751-753. Campbell,R. A., andSmall,A.M. (1963)."Effectof practiceandfeedback 5000 Hz. The resultsreportedherefit this hypothesis very on frequencydiscrimination,"J. Acoust.Soc.Am. 35, 1511-1514. well.In thisrespect,theytallywith psychophysical dataconConinx,F. (1978)."The detection of combined differences in frequency and cerningthe frequencydependence of frequencydiscriminaintensity,"Acustica39, 137-150. tion thresholds(Moore, 1973)and "tonechroma"or "musiFiorentini,A., andBerardi,N. (19•1)."Learningin gratingwaveformdiscrimination:Specificityfor orientationand spatialfrequency,"Vision cal pitch" perception(Bachere,1954;Attneaveand Olson, Res. 21, 1149-1158.
ACKNOWLEDGMENTS
Gibson,E. J. (1969).Principles of PerceptualLearningand Development (Prentice-Hall,EnglewoodCliffs,NJ). Harris,$. D. (1952)."Pitchdiscrimination," J. Acoust.Soc.Am. 24, 750755.
Theauthorisgratefulto Professor W. M. Hartmannfor hismultiplecontributions to thiswork.He alsothanksB. K. Smith,who programmedthe digitalsynthesizer to run the experiment,and Professor E. J. Gibsonfor a copyof the
Heimer,W. L, andTatz, S.J. (1966)."Practiceeffects, knowledge of results andtransferin pitchdiscrimination," ReportIH-52, US NavalTraining DeviceCenter{PortWas•.ngton,NY), unpublished. Jesteadt'W. H., and Sims,S. L (1975)."Decisionprocesses in frequency
reportby Heimer and Tatz. This work wassupportedby an ATP grantfrom the CNRS.
Johnson,D. H. (1980)."The relationshipbetweenspikerateandsynchrony in responses of auditory-nerve fibersto singletones,"J. Acoust.Soc.Am.
•Twodifferent earphones wereused(Sennheiser 430forgroups T200,T360,
Levitt,H. {1971}."Transformed up-downmethods in psychoacoustics," J.
discrimination," J. Acoust. Soc. Am. 5'/, 1161-1168.
68, 1115-1122.
andT2500;Sennheiser 414 for groupT6000}.Accordingto measurements Aconst.Soc.Am. 49, 467-477. carriedoutwitha B & K artificialear,theresponse ofeachearphone in the Moore,'B.C. J. (1973)."Frequency difference timensfort short-duration frequency region(s) whereit wasusedvariedby0.3 dBor lessfora frequentones," J. Aconst. Sec. Am. 54, 610-619. cychangefivetimeslargerthantheblock-2mediandiscrimination threshMoore,B.C. J. {1976)."Comparison of frequencyDL's for pulsedtonesand olds.
modulated tones," Br. I. Audiol. 10, 17-20.
:Themeanvalueoflog[Af•/A f •2)is0.21forgroupT6000and0.42forthe
Ramachandran, V. S.,andBraddick,O. (1973)."Orientation-specific lcam-
otherthreegroupscombined.Thesetwo values,respectively, correspond to 1.62and 2.63 for Af•/Af• 2. Accordingto Jesteadtand Sims{1975)and Turner and Nelson(1982},individualpsychometri½ functionsfor frequency discrimination canbeapproximatedby the equationd' ----k ßd.f,where d' isthediscriminability indexof signaldetectiontheoryanddfis themagnitudeof the frequencydifference.This wouldimply that for a fixedfrequencydilference, d;2/d • (i.e.,thegainin discriminability fromblockI to
ing in stereopsis," Perception2, 371-376. Rose,J. œ.,Brugge,I. F., Anderson,D. I., and Hind, J. I• {1967)."Phaselockedresponse to low-frequency tonesin singleauditorynervefibersof the squirrelmonkey,"J. Neurophysiol.30, 769-793. Turner, C. W., and Nelson,D. A. (1982)."Frequencydiscriminationin regionsof normaland impairedsensitivity,"J. SpeechHear. Res.24, 34-
block12)hasan averagevalueof 1.62for groupT6000and2.63for the otherthreegroupscombined.
Wyatt, R. F. {1945). "lmprovnbilityof pitch discrimination,"Psychol.
41.
Mono. 58, No. 267 entire.
Transformation of sound-pressure level from the free field to the eardrum presented in numerical form E. A. G. Shaw and M. M. Vaillancourt
Divisionbf Physics, NationalResearch Councilof Canada,Ottawa,CanadaK1A OR6
(Received23 November1984;accepted for publication14May 1985} In an earlierwork [E. A. G. Shaw,J. Acoust.SOc.Am. $6, 1848-1861(1974}],self-consistent familiesof curveswhichbestfittedthe experimentaldatafrom 12studieswerepresented to show the averagesound-pressure transformationfrom the freefieldto the humaneardrumin the horizontalplane.Tabulatedvaluesof azimuthaldependence at 15-degintervalsin azimuthand transformation to theeardrumat 0• azimutharegivenat 43 selected frequencies between0.2 and 12 kHz. These values are self-consistent within 0.1 dB.
PACS numbers:43.63.Hx,43.66.Lj,43.66.Pn,43.66.Qp
INTRODUCTION
Sometenyearsago,measurements of pressuretransformation, azimuthal dependence,interaural level difference, and ear canal pressuredistribution from 12 studieswere broughttogetherin a commonframeworkIShaw,1974).The pool of data covered100 subjects,the majoritymale, mea1120
suredin fivecountries, overa 40-yearperiod.Logicalproceduresweredevelopedto identifythe surfaceswhich bestfitted theseessentiallythree-dimensionaldistributionsof data, makingallowancefor the manydisparitiesbetweenstudies. The results were presentedas self-consistentfamilies of curvesshowingthe averagepressuretransformationfrom
J. AcOuSLSOC.Am. 78(3), Sept. 1985; 0001-4966/85/091120-04500.80;
1120
ownloaded 29 Jun 2010 to 193.50.102.40. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.j