Recognition of nonsensesyllables by hearing-impaired listeners and by noise-maskednormal hearers LarryE. Hurries a) Division ofHearingandSpeech Sciences, Vanderbilt University School ofMedicine, Nashville, Tennessee 3 7232 Donald D. Dirks, Theodore S. Bell, and Gail E. Kincaid Division ofHeadandNeckSurgery, U.C.L.,4.School ofMedicine, Los.4ngeles, California90024 (Received 4 February 1986;accepted forpublication 15October 1986) In thepresent study,speech-recognition performance wasmeasured in fourhearing-impaired subjects andtwelvenormalhearers. Thenormalhearers weredividedintofourgroups of three subjects each.Speech-recognition testing forthenormalhearers wasaccomplished in a background ofspectrally shaped noisein whichthenoise wasshaped to produce masked thresholds identicalto thequietthresholds of oneof thehearing-impaired subjects. The question addressed in thisstudyiswhether normalhearers witha hearing losssimulated througha shaped masking noisedemonstrate speech-recognition difficulties similarto thoseof listeners withactualhearingimpairment. Regarding overallpercent-correct scores, theresults indicated' thattwoof thefourhearing-impaired subjects performed betterthantheir corresponding subgroup of noise-masked normalhearers, whereas theothertwoimpaired listeners performed likethenoise-masked normallisteners. A grossanalysis of thetypesof errorsmadesuggested thatsubjects withactualandsimulated losses frequentlymadedifferent typesof errors. PACS numbers:43.71.Ky, 43.71.Es,43.66.Dc conductive pathologythanwith sensorineural impairment. Both filteringand conductivepathologyappearto simply The psychoacoustic and speech-recognition perforattenuate incoming signalenergypriorto stimulation of the manceof listenerswith sensorineural hearinglosshasbeen cochlea.It hasbeensuggested previously (Milner, 1982;Destudiedby severalinvestigators in recentyears(for recent Gennaroetal., 1981;Humes,1982)thatthemostapproprireviews,seeIesteadt,1978;Scharfand Florentine,1982; atecontrolconditionfor comparison to sensorineural hearHumes,1982;andMoore,1983).It hasbeenfrequentlysugingimpairment is the introduction of broadband masking gested thatthespeech•recognition difficulties experienced noiseshapedto producemaskedthresholdsin the normal by listeners havingsensorineural hearinglossaretheresult hearersthat are identicalto thoseof the impairedears.The ofpsychoacoustic abnormalities accompanying thecochlear maskingprocessresponsible for thresholdelevationis bepathology andnotsimplythethreshold elevation (Bonding, lieved to be predominantlyof cochlearorigin (Fletcher, 1979;Tyleretal., 1980;Patterson etal., 1982).Appearing to 1953;Zwicker and Feldtkeller, 1967;Zwicker, 1975, 1982). INTRODUCTION support thisconclusion isthefrequent observation thatnormalhearerslisteningto filteredspeech performbetterthan hearing-impaired subjects havingcomparable hearingloss. 'Such an outcome has been observed in several studies (Wanget al., 1978;Waldenet al., 1981;Kiukaanniemi, In addition, noise-maskednormal listeners demonstrate loudnessrecruitmentcomparableto that observedin ears with sensorineural pathology(Lochnerand Burger,1961; Stevens,1966). The presentstudysoughtto evaluatehow accurately 1979, 1980). It is reasonedthat if the filteringeffectsof noise-masked normal listeners simulate sensorineural listenthreshold elevation wereresponsible for thespeech-recogniers in their speech-recognition performance. If somephetion performance of hearing-impaired listeners, thenthe nomenonother than thresholdelevationof cochlearorigin normals shouldalsoperformpoorlywhenlistening to approandaccompanying loudness recruitmentisat leastpartially priatelyfilteredspeech. Althoughtheperformance of norresponsible for thespeech-recognition difficulties of listeners mal hearersis reducedby filteringthe signal,the normal with sensorineural impairment,thenonewouldexpectthe hearers listening to filteredspeech typicallyoutperform the sensorineural listeners to performmorepoorlythanthenorimpaired subjects. It isfrequently suggested thatsome addi- mal hearerslisteningin noise.If, however,similarperfortionalabnormality beyondthe thresholdelevationmaybe manceis observed for bothgroupsor theimpairedsubjects operating in theimpairedearsto account for thisdifference performbetterthanthenoise-masked normallisteners, then in performance betweengroups. no additional abnormality, beyond threshold elevation of Filtered signalspresentedto normal hearers,however, cochlearorigin and accompanying loudnessrecruitment, providea bettersimulation of hearinglossassociated with needbe invokedto explainthe speech-recognition perfor- a•Present address: Department' ofSpeech andHearing Sciences, Indiana manceof the impairedlisteners.Superiorperformanceby University,.Bloomington, IN 47405. 765 J.Acoust. Soc.Am.81 (3),March1987 the hearing-impairedlistenersover that of noise-masked 0001-4966/87/030765-09500.80@ 1987Acoustical Society ofAmerica 765 normalscloselymatchedfor signalaudibility, however, while not suggestive of additionalprocessing difficultiesin the impairedlisteners,couldcompromisethe utility of noisemaskednormal hearersas a modelof sensorineural hearing loss. It shouldbe notedthat thisapproachhasbeenpursued by othersin recentyears,althoughonlya limitednumberof subjects or a limitedrangeof speech stimulihavetypically beenemployed(Miller, 1982;Braidaetal.,1985;Zurekand Delhorne, 1986;Fabry and Van Tasell, 1986). All possible outcomes, moreover, have been observed in these earlier studies. Some noise-masked normal hearers performed better thanimpairedlisteners, someperformedthesame,while still othersperformedmorepoorly. !. METHODS A. Subjects A total of 16 subjects participatedin the presentstudy. Four subjectshad bilaterally symmetricalsensorineural hearinglossof presumed cochlearorigin.All fourhadmoderate-to-severe high-frequency hearinglosswith varyingdegreesof impairmentin thelowandmidfrequencies. Threeof the four subjects had their hearinglosssincebirth or early childhoodwhile one (HI-02) had suddenonsetof hearing lossapproximately 4 yearspriorto thepresenttesting.Subjectageswere32, 56, 17,and36yearsforsubjects HI-01, HI02, HI-03, and HI-04, respectively. All impairedsubjects had normaltympanograms. Three normal hearers were selected for each of the hear- ing-impairedsubjects to serveas noise-masked normallisteners.By usingthreenoise-masked normalhearersfor each impaired subject,interindividualvariability in the performanceof normallistenerswith identicalmaskedaudiograms couldbe takeninto considerationwhen comparingthe performanceof the hearing-impaired to the noise-masked nor- nontestearofall subjects toassurethatlisteningwasrestricted to the test ear. For thenoise-masked normalhearersit wasnecessary to mix the testsignalwith a shapedmaskingnoise.The latter wasproducedby a random-noisegenerator(G-edRad1390B ) which,followingattenuation( Hewlett-Packard350D), was shapedby a 1/3 octave-bandmultifilter (GenRad 1925). The outputof the multifilterwasthenroutedto the secondchannelof the GS-162speechaudiometerusedfor the testsignal.The outputof this channelwassetto maximum attenuationwhen testingthe hearing-impaired subjeers. For speech-recognition testing,the C.U.N.Y. Nonsense SyllableTest (NST; Resnicket al., 1975) was employed. This testis comprisedof vowel-consonant and consonantvowel nonsensesyllablesspokenby a male talker and arrangedin 11 subtests. Within eachsubtest,consonantposition (syllableinitial or syllablefinal), consonantvoicing (voiced or voiceless),and vowel environment (/a,u,i/) are constantwhile consonantplaceand/or mannerof articulation vary. Nine of the original subtestswere usedin this investigation(subtests1, 2, 3, 6, 7, 8, 9, 10, and 11). These stimulihad beenpreviouslydigitized(Kamm et al.,. 1985). The NST stimuliwereroutedthroughequipmentidentical to thatof thepure-tonesignalsfollowingoutputthroughthe digital-to-analog converter.A 400-Hz calibrationtonedigitizedfromthemastertapeof theNST materialswasusedfor calibrationandspecification of presentation levels.All stimuluslevelsfor puretones,speechsignals,and noisestimuli are referencedto the soundpressurelevelgeneratedin an NBS-9A6-era3coupler. c. Procedures Pure-tonethresholdsfor all subjectswereobtainedusinga two-alternativeforced-choice paradigmdesigned to estimate the 79.6 percent-correct point on the psychometric real-hearing subjects. Differences between* impairedand functionrelatingpercent-correct signaldetectionto stimumasked-normal listeners couldthenbeevaluatedin light of lus level (Levitt, 1971 ). Three successive correctresponses theobserved between-subject variabilityof thenoise-masked were followed by a decrease in signal level while incorrect normalhearers.All 12 subjectsin this grouphad pure-tone responses were followed by an increase in the level of the thresholdsfrom 125-8000 Hz< 15 dB HL (ANSI, 1969). signal. A total of 15 reversals in signal level were employed Tympanogramswere normal and acousticreflexeswere for a single threshold estimate. Signal increments or decrepresentat 100dB HL at 1000Hz. Normal hearersrangedin ments were 10 dB for the first five reversals and 2 dB for the agefrom 19-32 years. last ten. Of the last ten reversals,the first two were discarded For measurement of pure-tonethreshold,stimuliwere generateddigitallyby a PDP-11/23 laboratoryminicomputersoasto havea 350-msduration(from onsetto offset) with the midpointof the last eight representingthreshold. Pure-tonethresholdsweremeasuredfor frequencies at 1/3octaveintervalsfrom 125-8000Hz. The 19 testfrequencies weredividedintotwosubgroups by selecting everyothertest frequencyfrom 125-8000 Hz. Test orderwithin a subgroup with a 25-ms raised-cosine rise-fall time. The stimuli were was randomized. outputthrougha 12-bitdigital-to-analog converterat a rate of 25 kHz and low-passfilteredat 10kHz. The stimuliwere then routedthrougha programmableattenuatorthat was controlledby the computerand sent to one channelof a speechaudiometer(Grason-Stadlermodel 162, GS-162) priorto transduction by a TDH-49 earphonemountedin an MX-41/AR cushion.A secondGS-162speechaudiometer wasusedto producea speech-shaped noiseat an overall soundpressure levelof 65 dB. This noisewasroutedto the subgroupsthe procedure was repeated a second time. Thresholdsreportedare the meansof thesetwo estimates. Identical psychophysicalprocedureswere employed with the hearing-impairedand normal-hearingsubjects. Oncethe quietthresholdsfor the hearing-impaired subjects had beenestablished, the pure-tonetestingfor the normal B. Apparatus and stimulus generation 766 J. Acoust.Sec. Am., Vol. 81, No. 3, March 1987 After thresholds were obtained for both hearersproceeded.Initial multifilter settingsrequiredto shapethenoisesoasto producemaskedthresholds identical to the quiet thresholdsof the corresponding hearing-imHumeseta/.: Recognitionof nonsensesyllables 766 pairedsubject wereestablished fromcalculations. Thecalcu- patedfor 8-12 h. All subjects werepaidfor theirparticipa- lations made use of the critical ratio and the l/3-octave-band tion. noiselevels measured fora flatmultifilter setting? Withthe exception of onegroupof subjects, thosesimulating HI-03, theseinitial settingsrequiredonly minoradjustments to achieve thedesiredgoalthat maskedthreshold bewithin3 dBoftheimpaired subject's quietthreshold foratleast17of II. RESULTS A. Pure-tone thresholds Figure I displaysthe meanpure-tonethresholdsobtainedfromthefourhearing-impaired subjects of thisstudy. For comparison, a setof reference datafromnormal-heating subjectsis alsoprovided(filled circles).The latter values the end of three 2-h sessions,then the last set of multi filter represent meant_hresholds from12normal-hearing subjects adjustments Wereconsidered final. testedpreviously in thesamelaboratory withidenticalpsySpeech-recognition testingproceeded identicallyfor the procedures. As indicatedin thisfigure,all imhearing4mpaired and noise-masked normal-hearingsub- ehophysical paired subjects in the present studyhadmoderate-to-severe jects.The testitemsfor a givenrandomlyselectedsubtest high-frequency hearing loss with varyingamounts of hearwereadministered six timesin completelyrandomfashion ing loss in the low and mid frequencies. beforeproceeding to anothersubtest.All subtests werepreTable I providesthe root-mean-square (rms) error sentedinitially at 86 dB SPL for practice.Testingwasthen between the quiet thresholds of the impaired listenerand performedat 66, 76, and 86 dB for threesubjectgroups(HI01, HI-02, HI-04, and those noise-maskednormals simulateachof the threecorresponding noise-masked normallisteners. Figure 2 illustratesthe agreementbetweenthe-threshing theseconfigurations)and 56, 66, and 76 dB for the renormals mainingsubjectgroup(HI-03 andassociated noise-masked oldsof the poorestmatchof the threenoise-masked hearing-imnormals). Presentationlevelswere selectedto encompass and the thresholdsfrom the corresponding These the linear portion of the performance-intensity function pairedsubjectfor eachof the four impairedsubjects. data,togetherwith the rms-errorvaluesappearingin Table whileattempingto maintainuniformityacrosssubjects. OrI, indicatethat a dose match was achievedbetweenthe imder of testingfor the threetestlevelswas randomizedfor paired subjects andthenoise-masked normalhearers. The eachsubject. poorest match overall was observed in the subgroup simulatThe speech-recognition testingconcluded with a retest ingimpairedsubjectHI-03 (lowerleftpanelof Fig. 2). This of severalsubtests administeredat oneof the lowertwo preimpaired subjecthad the steepest slopein audiometricconsentationlevels.Generally,thosesubtestsfor which the subfiguration. The difference between maximumand minimum jectsrecognizedbetween20% and 50% of the itemscorrecthearing loss for this subject exceeded the workingrangeof ly were retested.This typically amountedto four or five the multifilter. Unable to match both the maximum and subtests per subgroup.This expandedsetof data wasto be the 19 test frequencies.In most instancesthis goal was achieved.If, however,the desiredgoal wasnot achievedat used forsubsequent analysis ofconfusion patterns, ananaly- minimum thresholds,it wasdecidedto match the maximum sisthat isnotreportedhere.Thesedata,however,areusedin thepresentstudyto evaluatethetest-retestreliabilityof the individual subtest scores. Subjects wereseatedin a sound-treated roomfor testing. A response box, undercomputercontrol,was usedto activate appropriatelightsand buttonsfor both the pure-tone andspeech-recognition testing.The 2AFC pure-tonethreshold paradigmutilizeda 500-mswarninglight followedby a 500-msdelay.Eachintervalwasthenmarkedby a light activatedfor 350 ms with a 500-msdelaybetweenintervals.A response light followedthe secondintervaland indicated that the computerwas readyto receivethe subject'sresponse.The subject'sresponseterminatedthe trial. The same warning light parameterswere used for speech-recognition testing.For thistesting,however,no intervallightwasrequired.In addition,oncetheresponse light wasactivated,the subjectwasrequiredto pressthe button corresponding to the syllableheard.A removablelabel,containingorthographicrepresentations of the nonsense sylla- blesin a particularsubtest,waspositioned abovean arrayof buttons.Eachbuttoncorresponded to one of the possible items on that subtest. Trial-to-trial feedbackregardingcorrectresponses was providedonly for the pure-tonetesting.The hearing-impairedsubjectsrequired6-8 h of testingto completethe study.The noise-masked normal-hearinglistenerspartici767 J. Acoust. Sec. Am., VoL 81, No. 3, March 1987 thresholdsascloselyaspossibleat the expenseof the minimum thresholds.The poormatchis apparentin the region between400 and 1000Hz for thedatafromsubjectMN-03B shownin the lowerleft panelof Fig. 2. The maskedthresholds for the other two noise-masked normal listeners of this subgroupalsovariedfrom thoseof the impairedsubjectin this frequencyregion. I 98 o I I HT-81 I I I Ill I I I I I I I A n H•-82 X HT-04 70 x--X-_ X .j 6• 3• 2o le I 2m• i J i S00 Illll I 100• FREQUENCY 28• i •;000 (Hz) FIG. 1. Mean pure-tonethresholds for the four hearing-impaired subjects in this study. For comparison,the filled circles show mean pure-tone thresholdsfor normalhearers()V= 12) testedwith identicalprocedures. Humos eta/.: Recognitionof nonsensesyllables 767 TABLE 1.The tanserrorbetweenmaskedthresholds of a normal-hearing subject andquietthresholds ofthecorresponding hearing-impaired subject. Subject rms error (dB) MN-01A MN-01B 4.5 4.2 MN-01C 4.0 MN-02A 3.5 MN-02B 3.8 MN-02C 2.2 MN-03A 4.9 MN-O3B 6.1 MN-03C 5.1 MN-04A 3.2 3.1 2.9 MN-O4B MN-04C presentation levels employed appear tohaveadequately defined the linear portion of the performance-intensity func- tion.Floorandceilingeffects arenota,pparentandthelevels havedefineda wide rangeof performancefor eachsubject (at least a 35% differencebetweenthe minimum and maxi- mumperformance). Regarding overall speech-recognitionpeformanee, comparisonbetweenthe data from the hearing-impaired subjectand the corresponding noise-masked normalsfor each of the four subgroupssuggestedthe following outcomes: (1) Hearing-impaired subjectsperformed better than noise-masked normalhearers;or (2) hearing-impaired subjectsperformedthe sameas noise-maskednormals. An exampleof the firstoutcomeis depictedin Fig. 4. This figure providesperformance-intensity functionsfor each subtest and for the overall (NST) B. Speech-recognitionperformance Figure3 illustrates theperformance-intensity functions for eachof the four hearing-impairedsubjects.Each data pointrepresents the meanoverallpercentcorrectfor 438 syllablepresentations (73 syllables presented sixtimes).The I g8 J ß I I i I Ill& i i represented by the filled circlesconnected by solidlines whiletheothersymbols represent datafromeachof thethree corresponding noise-masked normalsubjects. Asthedatain the lowerright panellabeled"ALL" indicate,the overall NST scoreof this impairedsubjectwasbetterthan that of everynoise-masked normalhearerat all presentation levels. For the datain eachpanelof Fig. 4, aswell asFigs.5, 6, and 7, the followingcriterionwasusedto defineequalityof per- I I HT-O I score. The data for HI-01 are I I I I t I I I I I I I I I I I gO ß HI-02 • MN-• IA - eooHN-025 7½} u•5½} - '-o. '" - -•6½} •"ø'"ø•ø' •4• RHS ERROR - 28 I 20½} i i I I lilt I see ! eaa 2aao FREQUENCY (Hz) I 4.$ dB 200 -1 • ' RH$ ERROR = 3.8 I seca i i i i Illl I s½}½} t oea 2gee FREQUENCY (Ha) I I I 11tl s½}oo ß H[-04 IIIIIIIIIIIIIIIJlJ f iII•IIIIIIIIIiill oMN-O3B 7½} --0' 20 J i i Ill ß HI-C3 "O. q e½} oMN-•4A 78 •.1_. 6½} _- ,t / 0 -o. o• 2½}½} See} ! OZ½} 2e½}a FREQUENCY (Hz) i 2½} S½}½}e 20½} RI'• ERROR see iees FREQUENCY :'eoe FIG. 2..Comparison of meanquietthresholds of eachimpairedsubject(filledcircles) to themeanmaskedthresholds of the mostpoorlymatchednoisemaskednormallistenerin eachsubgroup(opencircles).Eachpaneldisplaysthedatafor a differentheating-impaired subject. 768 J. Acoust.Sec. Am.,VoL81, No. 3, March1987 Humeseta/.: Recognition of nonsense} syllables 768 • 7e _ bJ n, n, z 66 bJ o 4e • 3e •o lee 2o • 70 I I I n, LI L2 L• o SPL i- SPEECH LEVEL dB i,i FIG. 3.Performance-intensity functions relatingoverallpercent correcton theNSTtospeech level. L1---56dBforHI-03and66dBfortheother three impaired subjects. L2= LI 4- 10dBandL3= L1+ 20dB.Circles: HI-01; • 3e •8 Squares: HI-02;Triangles: HI-03; Xs: HI-04. • 7E •9 • • 7• •10 • SPEECH • 7B L•¾EL x •11 88 86 7B ALL TM •8 e6 76 I• ½cJB SPL• formance between thehearing-impaired listenerandthecor- FIG. 5.SameasFig.4 butforHI-04 andcorresponding subgroup ofnoiseresponding noise-masked normalhearers. If thescores ofthe masked normal listeners. hearing-impaired subject ona particular subtest werewithin therangeofscores observed forthethreenoise-masked norlistening in noise.On a fewoccasions, the malhearersonthatsamesubtest at twoor morepresentation normalhearers oftheimpaired subject relative tothatof the levels, thenperformance oftheimpaired subject wasconsid- performance group ofnoise-masked normal hearers varied withpresentaered to be the same as that of the noise-maskednormal one hearers. Otherwise, performance of the impairedlistener tionlevelsuchthatonelevelyieldedequalperformance, levelyielded superior performance bythehearing-impaired was consideredto be either better or worse than that of the subject, andtheremaining levelrevealed superior performancebythenoise-masked normalhearers. In theseeases, theperformance ofthenoise-masked normal hearers andthe hearing-impaired subjects wasconsidered tobeequal. Using thisdefinition of equalityof performance, examination of 40 3e• 20• a • tee ?e - • x SPEECH LEVEL (dB SPL• FIG. 4. Percent-correct performance oneachof theninesubtests of the NSTplottedversus speech level.Eachpanelshows theresults fora separate subtest withthenumber ineachpanelcorresponding tothesubtest number. SPEECH LEVEL CdB SPL) "ALL" in thelowerrightpanelreferstotheoverallNSTscore. Theconneeted filledcircles represent thedatafromhearing-impaired subject HI-01 whiletheothersymbols represent theperformance of thecorresponding FIG. 6. SameasFi•. 4 butforHI-03 and•ndin• noise-masked normallistenersin this subgroup. 769 J.ACOuSt. Sec.Am.,Vol.81,No.3,Marciq 1987 m•k• no•=l sub•oupofnoise- listeners. Humes eta/.:Recognition ofnonsense syllables 769 individual NST subtest scores obtained from individual lis- • .o x x x x teners(Dubno and Dirks, 1982). Subtest-score reliability measuredpreviously,moreover,wasassessed usingrepetitionsof a singlesubtestcomprisedof 7-9 testitems.In the presentstudy,a singlesubtestscorewasbasedonbetween42 and 54 testitems.Recall that in the presentstudy,subtest scoresobtainedfrom hearing-impairedsubjectsthat fell eo • 46 2•- between 20% and 50% correct were fetested. These same o subtestswere subsequentlyretestedfor the corresponding noise-masked normalhearers.As mentionedpreviously,this wasdoneto increasethe samplesizefor subsequent analysis of confusionpatterns,an analysisthat will not be reviewed • ee here. In addition, somesubtestscoresoutsidetheselimits for the lowesttwo presentationlevelswere arbitrarily selected for retest.Thesedata, therefore,affordedan opportunityto examinethe reliability of the individual subtestscoresobtainedin this study.Figure 8 showsthe test-retestscoresfor the varioussubtests, eachrepresented by a differentsymbol. s• 3• x Individual SPEECH L•EL (rib SPL) FIG. 7. •mc • Fig.4 butforHI•2 •d •nding ;ubgmupo•nois• data from noise-masked normal hearers and im- pairedlistenersare presentedtogetherin thisfigure.Due to the randomizationof presentation level in this study,the amountof practiceontheNST betweentestandretestvaried between1 and 3 h. The datain thisfiguresuggest that the individual subtest scores are reliable. The Pearson-r correla- individualsubtestscoresin the remainingpanelsof Fig. 4 indicates that it is generallythe casethat the hearing-im- tion coefficient between test and retest score was 0.89 (p < 0.01). Again,eachsubtestscorein thisfigurewasbased on 42 to 54 syllablepresentations(7-9 syllablesper subtest paired subjectoutperforms the noise-masked normal repeated sixtimes). hearers. Theexceptions aresubtests 1and6 (subtest number The results were analyzedfurther to providesomeinisindicated bythenumbers at thebottomofeachpanel).On sight into the types of errorsmadeby hearing-impaired and these two subteststhere appearsto be little difference noise-masked normal-hearing subjects. This was accombetween the performance of the impairedsubjectandthe plishedby examiningindividualsubtestscores.Subtests1-3 masked normals. Figure5 displays a similarsetof performance-intensitycontain syllable-initialconsonantspaired with the vowel /o/. The remainingsixsubtests incorporated in thisstudyall functions foranotherhearing-impaired subject(HI-04) that contain syllable-final consonants, two paired with/o/(subtendedtooutperform thenoise-masked normalhearers. Altests 6 and 7), two paired with/i/(subtests 8 and 9) andtwo thoughtheoverallNST performance of theimpairedlistenpaired with/u/(subtests 10 and 11). Comparison of suber,shownin thelowerrightpanel,isconsistently superiorto test-score profiles between the hearing-impaired subjects thatof thenormalslistening in noise,thisisnotthecasefor normal-hearing counterparts would all individualsubtests.Using the criterion of "sameness" andtheirnoise-masked definedpreviously, performance of the impairedsubject is similar to that of the noise-maskednormals for subtests3, 7, 100 I I I I I I I I I/ 11, and 2. In no case,however,doesthis impairedlistener /- perform worse thantheentiregroupofnoise-masked normal listeners. Figures 6 and7 showcomparable datafortheremaining twohearing-impaired subjects, HI-03 andHI-02, respectively. In bothcases,the datain the lowerrightpanelindicate •" that the hearing-impaired and noise-masked normal-hearing subjects performequivalently on the NST. This is also generally truefor individualsubtest scores for bothsubject subgroups, althoughthereareexceptions (subtests 3, 8, 9, and 10 in Fig. 6 andsubtests 1, 2, and 8 in Fig. 7). These exceptions, moreover, arenotconsistently inthesamedirection.On somesubtests theimpairedsubjectperformedbetter than the noise-masked normal hearers whereas on others just theoppositeoutcomewasobserved. Throughoutthisanalysisit hasbeenassumed that indi- L " *•'" - I I I I I I IPERCENT CORRECT--TEST vidual subtestscoresfrom indiviual subjectsare reliable. FIG. 8. Percent-correct performance on rctestis plottedasa functionof performance on theinitialtestfor individualsubtest scores. Data for impairedandnoise-masked normallisteners havebeencombined. Eachsym- Only limited data are availableregardingthe reliabilityof bol represents a differentsubtest. 770 J. Acoust.Soc. Am., VoL 81, No. 3, March 1987 Humosoral.: Recognitionof nonsensesyllables 770 providesomeinformationregardingdifferences in thetypes of errors made. Toevaluate thepatternoferrorsmadeacross subtests of er. Figure9 displays the meanAP(c) valuesobtained for eachofthefoursubgroups ofsubjects, witheachpanelrepresentingdatafrom a differentsubgroup. The threesymbol typesreflectthe threespeech presentation levelsemployed. the NST, the subtestscoresin Figs. 4-7 were convertedto differencescoresin whichthe scoreof a givennoise-masked The dashedlinein eachpanelat AP(c) = 0 (zero) indicates normalhearerwasexpressed relativeto that of the hearing- no differenceon averagebetweenthe performanceof the hearing-impairedsubjectand noise-maskednormalscomimpaired subject. Specifically, AP(c),•= P(c),•i prising that subgroup.Negative valuesof AP(c) indicate --P(e)em, where P(e), = percent correct on subtesti, MN$= masked normal j, andHI = hearing impaired listen- that the hearing-impairedsubjectoutperformedthe noisemaskednormalhearersin the samesubgroup,andconversely for positivevalues.From thedatain thefourpanelsof Fig. 9, it is apparentthat the majorityof AP(c) valuesfor sub- x ¾ I I I I I ß I ß I t t ! 2 3 8 7 8 0 19 II hIST St,.•TEST x ß x • --t ß HZ-82 o x...... •--•-o I 2 3 8 ? 8 0 tO II NST SI,J•TEST x that the relative error rates across subtests were not the same ß ............. groupsHI-01 and HI-04 are negativewhilethoseof HI-02 and HI-03 vary aroundzero. A separateanalysisof variancewith repeatedmeasures havingwithin-subjectfactorsof presentationlevel (three levels) and subtest(nine subtests)was performedon the datain eachpanelof Fig. 9. For subgroups HI-02 andHI-03, the grandmeanAP(c) wasnot significantly differentfrom zero(p > 0.10). That is,collapsed across levelsandsubtests, therewasno differencein performance betweenthe noisemaskednormalsand the hearing-impaired listenerin these two subgroups. For subgroups HI-01 and HI-04, however, therewasa significant difference betweenthe performance of the impaired listener and the noise-maskednormal hearers,with theimpairedsubjectoutperforming thenormal hearerslisteningin noise(p < 0.01). For all four subgroups, therewasa significanteffectof subtest(p < 0.01) on AP(c), whereas presentation levelwasnotsignificant (p > 0.05). In addition,the subtest-by-level interactionwassignificantin all foursubgroups (p < 0.01). The significant maineffectof subtest andthesignificant subtest x levelinteraction suggest x NX-83 for thenoise-masked normalsandthe hearing-impaired listenersandthat themannerin whichtheirperformance differedacrosssubtestvariedwith presentation level. ......... IlL DISCUSSION I I I I I I i ! I I 2 3 6 7 8 9 10 It NST SUBTEST HY-O4 x I• ß normalhearersfor comparisonto hearing-impairedsubjects havealsoreportedmixedoutcomes.Milner (1982), for example,presentsperformance-intensity functionsfrom two listenerswith sensorineural heatinglosslisteningto uniiltered and filtered nonsensesyllables.Only the unfdtered data are appropriatefor comparisonto the data from the presentstudy. For subjectshaving unilateral sensorineural impairment,comparisonwasmadeto two setsof data from 0 x I I I I I I I I I I 2 3 6 7 8 0 18 II HST SOBTEST FIG. 9. Mean deltaP(c), percentcorrectby noise-masked normallesspercentcorrectby thecorresponding impairedlistener,plottedversussubtest. Eachsymbolrepresents a differentspeechlevel (fOledcircles:Ll; untilled circles:L2; and X s:L3). Eachpanelpresents theresultsfor a separatesub- noise-masked normal ears. In one case, the listener's own normal ear servedas the noise-maskednormal ear, while in the other casea secondsubjectwas the noise-maskednor- group. 771 Regardingoverallnonsense-syllable recognition,none of the hearing-impaired subjectsin this investigation performedworsethanthe corresponding noise-masked normal hearers.When there was a differencein performance,the hearing-impaired subjectsperformedbetterthan their normal-hearingcounterpartslisteningin noise.This outcome occurredfor two of the four hearing-impaired subjects with the othertwo subjectsperformingthe sameastheir respective subgroups of maskednormals. Prior studiesof speechrecognitionusingnoise-masked J. Acoust.Soc. Am., VoL 81, No. 3, March 1987 Humosota/.: Recognitionof nonsensesyllables 771 real.For bothcomparisons, thescores obtainedfor thesimulated loss were 20%-25% better than those obtained for the actuallossat all presentation levels.For the otherimpaired subjectin Milner'sstudy,mixedresultswereobtainedwhen comparedto the datafrom a noise-masked normalhearer. drawingconclusions regardingthe accuracyof the match obtained between3. the speech-recognition performanceof maskednormalsandhearing-impairedlisteners.This would alsoaddgreaterfacevalidityto the measureof"speechrecognition." Theoutcome ofthepresent studyregarding theutilityof At speech levelsbelow80dBSPL,performance of thehearing-impaired subject exceeded thatofthemaskednormalby the noise-masked normal car as a model of sensorineural 15%-20%, whereastheopposite wastruefor levelsexceed- hearinglossis equivocal.For two of the four hearing-impairedsubjectsof thisstudy,overallspeech-recognition pering 80 dB. formanee could be accurately simulated by introducing a More recently,Braidaetal. ( 1985) andZurekandDelspectrally shaped masking noise into a normal ear. For the horne (1986) have reported good agreement between thiswasnotthecase.It isimportant speech-recognition scores of hearing-impaired subjects and remainingtwosubjects, to note, however, that the four impairedsubjectsof this comparable datafromnoise-masked normals.All subjects in study neoer performed worse thanthenoise-masked normal thesetwo studies,however,listenedto speechin a backhearers. This is true for overall speech-recognition perforgroundof broadband noise,whereasMilner (1982) tested mance sampled with a wide variety of speech stimuli and his subjectsin quiet. over a wide range of speech levels, extending from conversaIn anotherrecentstudy,Fabry and Van Tasell(1986) reporteddatafrom six unilaterally impairedlisteners in tional levelsto levelsapproachingthoseexperiencedwith amplification.This is true,moreover,despitecloselymatchwhich the listener's own normal ear served as the noiseing the impairedandnoise-masked listenersfor audibilityof maskednormalear for comparison. All threepossible outthe speech signal. If additional secondary psyehoacoustie comeswere observedin this studywhen recognitionscores processing deficits were operating in the four impairedsubfrom the ears with simulatedand actual hearing losswere jects of this study, one would have expected them to perform compared. Thenoise-masked normalearfora givensubject worse than noise-masked normal hearers listening underthe yieldedscores, obtainedat twopresentation levels,thatwere same test conditions. Again, such an outcome was not obeitherbetterthan,the sameas,or worsethanthe scoresfrom served in the present study, suggesting that either the four theimpairedear.Oneofthelevelsused(90 dBSPL) wasthe impaired subjects of this study did not have such secondary samefor all subjects andwascomparable to thehighestlevel deficitsor, if theydid, the deficitsexertedno inusedwith threeof the four subgroups of this study(86 dB processing performance.Not SPL). For this high level presentation, the differencein flueneeon overall speech-recognition measures from overallpercentcorrectbetweenthe impairedear and the havingobtainedadditionalpsychoacoustie maskedear in the Fabry and Van Tasellstudyhad a mean thesesubjects,it is not possibleto decidebetweenthesetwo valueof -- 7.4% anda rangeof -- 1.3% to -- 23.3% with alternatives. negative valuesindicatingP(c)m < P(c}M•. Themeanperformaneedifference at 86 dB in thepresentstudyfor subtests ACKNOWLEDGMEHTS usingCV syllables pairedwith/n/, asin theFabryandVan This work wassupportedin part by an RCDA awarded Tasell study, is --3.6% with a range of q-5.1% to to the firstauthorby the National Instituteof Neurological, -- 26.3%. Thus, for similarstimuli,the resultsof both stud- Communicative Disorders and Stroke (NINCDS) and an iesaresimilarwith noise-masked normallistenerstendingto NINCDS grant awarded to U.C.L.A. Preparationof the outperformthe corresponding hearing-impaired listener. manuscriptwas also supported,in part, by a grant from Milner (1982) reacheda similarconclusionusingCV items NINCDS to IndianaUniversity.The authorsthank Richard pairedwith/a/,/u/, and/i/. Whenthedifference in perfor- Tyler and Dianne Van Tasell for providinghelpful commancebetweenmaskedandimpairedearsat 86 dB isexam- mentson earlierversionsof thismanuscript. inedacross allsubtests fromthepresent study, inel•ding VC syllables pairedwith/o/,/u/, and/i/, themeandifference becomesq- 6.3% and the rangeis q- 21% to -- 15%. For thisbroaderrangeofstimuli,thetendency isforthehearingimpairedsubject to outperform normalhearerslisteningin 'Withthemultifilter setto0 dBateachl/3-octaveband(attenuation range is from + 25 to -- 25 dB at eachfrequency),the SPL outputwasmea- suredat eachl/3-octavein a 6-era 3 coupler fora broadband noiseinput noise. signal. Thel/3-octave bands areequal toorgreater thantheeritic.•l band, The findingsof thepresentstudyareconsistent with the limited data availablein the literature for comparabletest therebyenablinguseof the critical ratio to producean estimatedmasked thresholdat eachcenterfrequency.This was accomplishedby converting theoverallsoundpressure levelfor a given!/3-octavebandintonoisespectrum level,No [No in dB $PL = overalldB SPL-10 log ( 1/3-octavebandwidth). The critical ratio wasthen addedto N o. Thus, for a flat 0-dB setting of the multifilter,a predictedmaskedthresholdwasestimatedat each l/3-octave centerfrequency.Adjustingthe overalllevelof the masking noiseaffected.themaskedthresholdsuniformly acrossl/3-oetave bands, while eachband could alsobe adjustedseparatelyover a 50-dB range conditions.Becauseof the diversityof outcomesin the literature, however,virtually any findingwouldhavebeenconsistent with at leastsomeprior data. The diversityof previous outcomesappearsto be due,in part, to the natureof the CV or VC nonsensesyllablesselectedfor testing.The reasonsfor thisdependence of thegoodness of thematchin performance on the consonantlocationwithin the syllableare not immediatelyobvious.It is advisable,therefore,to includeasmany vowel contextsand consonantpositionsas possiblebefore 772 J. Acoust.Soc. Am., Vol. 81, No. 3, March 1987 ( + 25 to -- 25 dB re: 0-dB measurements). ANSI (1969). "Specifications for audiometers,"ANSI S3.1-1969(Ameriean National Standards Institute, New York). Humeseta/.: Recognitionof nonsensesyllables 772 Bonding,P. (1979). "Frequency selectivity and speechdiscrimination in sensorineural hearingloss,"Scand.Audiel.g, 205-215. Braids,L. D., Zurek,P.M., Durlach,N. I., andMilner,P. (1985)."Appli- Ph.D. dissertation, City Universityof New York. Moore, B.C. J. (1988). "Reviewpaper:Psycheacoustics of normaland impaired listeners," in Hearing:Physiological Bases andPsychephysics, cationof articulationtheoryto thestudyof speech reception by theheareditedby R. KlinkeandR. Hartmann(Springer,Berlin). ingimpaired,"J. Acoust.Sec.Am. Suppl.177, S68. Patterson, R. D., Nimmo-Smith,I., Weber,D. L., andMilroy, R. (1982). DeGennaro,S., Braids,L. D., andDurlach,N. I. (1981). "Studyof multi"Thedeterioration of hearingwithage:Frequency selectivity, thecritical bandsyllabiccompression withsimulated sensorinenral hearingloss,"J. ratio, the audiogram,and speechthreshold,"J. Acoust.Sec.Am. 72, 1788-1803. Acoust.Sec.Am. Suppl.1 69, S16. Dubno,J. R., and Dirks, D. D. (1982). "Evaluationof hearing-impaired Resnick,S. B., Dubno,J. R., Hefthung,S., and Levitt, H. (1975). "Pholistenersusinga nonsense-syllable test. I. Test reliability,"J. Speech nemeerrorsona nonsense syllabletest,"J. Acoust.Sec.Am. Suppi.1Sg, Hear. Res. 25, 135-141. Fabry, D.A.,andVanTasell, D.$.(1986). "Masked andfiltered simulationof hearingloss:Effectson consonfmt recognition," J. SpeechHear. Res. 29, 170-178. Fletcher,H. (1953). Speechand Hearing in Communication(Van Nostrand, New York). Humes,L. E. (1982). "Spectraland temporalresolutionby the hearing impaired,"in The Vanderbilt HearingAid Report,editedby(3. A. StudebakerandF. H. Bess(Monographs in Contemporary Audiology,Upper Darby, PA). Jestcadt,W. (1978). "Frequencyand temporalanalysisin hearing-impairedlisteners," inddvances in Prosthetic Devices for theDeaf.•1 Technical Workshop, editedby D. L. McPherson(NationalTechnicalInstitute S114. Scharf,B., and Florentine,M. (19li2). "Psycheacoustics of elementary sounds,"in The VanderbiltHearing.4id Report,editedby G. A. StudebakerandF. H. Bess(Monographsin Contemporary Audiology,Upper Darby, PA). Stevens, S.S. (1966). "Power-group transformations underglare,masking and recruitment," $. Acoust. Sec. Am. ;39,725-735. Tyler, R. S., Fernandes, M., andWood,E. J. (19811)."Masking,temporal integrationand speechintelligibilityin individualswith noise-induced hearingloss,"in Disorders ofAuditoryFunction,III, editedby I. (3. Taylor (Academic, London). Walden,B. E., Schwartz,D. M., Montgomery,A. A., and Prosek,R. A. (1981). "A comparison of theeffectsof hearingimpairmentandacoustic for the Deaf, New York). filteringon consonant recognition," J. SpeechHear. Res.24, 32-43. Kiukaanniemi, H. (1979). "Simulation oflow-pass hearing loss in'aFin- Wang,M.D., Reed,C. M., andBilger,R. C. (1978). "A comparisonof the nishspeech discrimination test,"Acta Otolaryngol. 87, 445-450. effectsof filteringandsensorineural hearinglosson patternsof consonant confusion,"$. SpeechHear. Res.21, 5-36. Kiukaanniemi,H. (1980). "Speechdiscrimination of patientswith high frequencyhearingloss,"Acta Otolaryngol.89, 419-423. Zurek, P.M., and Delhorne,L. A. (1986). "Speechreceptionin noiseby Levitt,H. (1971). "Transformed up-downmethodsin psychoacoustics," hearing-impairedlisteners,"J. Acoust.Sec.Am. Suppl. 1 79, S23. J. Acoust. Sec. Am. 49, 467-477. Zwicker,E. (1975). "Scaling,"in Handbookof Sensory Physiology, edited by W. D. Keidel and W. D. Neff (Academic, New York), Vol. 6, Pt. I. Lochner,$. P. A., andBurger,J. F. (1961). "Form of theloudness function in thepresence of maskingnoise,"J. AcoustSec.Am. ;3;3,1705-1707. Zwicker,E. (1982). Psychoakustik (Springer,Berlin}. Zwicker,E., andFeldtkeller,R. ( 1967). DasOhralsArachrichtenempfanger Milner,P. (1982). "Perception of filteredspeech by hearing-impaired lis(Hirzel, Stuttgart). tenersandbynormallisteners withsimulatedhearingloss,"unpublished 773 J. Acoust.Sec. Am., VoL 81, No. 3, March 1987 Humos eta/.: Recognitionof nonsensesyllables 773