Effect of masker level on overshoot in running- and frozen-noise maskers Regine von Klitzing DrittesPhysikalisches Institut, Universitdt G6ttingen,Biirgerstrasse 42-44,D-37073 GSttingen, Germany Armin Kohlrausch a) Institutefor Perception Research (IPO), P.O. Box 513,NL-5600MB Eindhoven, TheNetherlands (Received23 October1992;revised18 November1993;accepted13 December1993) Maskedthresholdswere measuredwith running-and frozen-noisemaskers.The 5-kHz signal was2 ms in duration.The maskerwaslow-passnoise(20 Hz-10 kHz); its total durationwas 300 ms. The overall level of the masker was 30, 50, or 70 dB SPL. The onsetof the signal was delayedby 0, 3, 8, 18, 198, or 278 ms relativeto the onsetof the masker.In all frozen-noise measurements, thesignalwasaddedto the samefinestructureof the noise.Overshoot in frozen noisewasmeasured for two startingphases of the signalthat led to a 10-dBdifference for large signal-onset delays.In all threeconfigurations (runningnoiseandfrozennoisewithtwodifferent signalphases)maskerlevelhad a similarinfluence on overshoot. At the intermediate masker level (50 dB SPL), a significant amountof overshoot(up to 15 dB) wasobserved in all three conditions. At the low and the highmaskerlevels,overshoot wasverymuchreduced,and even becamenegativein mostconditionsfor the 30-dB-SPLmasker.For the 50-dB frozen-noise masker,the total variationof thresholdswith signalphasewas 8 to 11 dB for long signal-onset delays,but only 3 to 6 dB for shortdelays.For the low- andhigh-levelmaskers,whereonly a small overshootwas observed,the thresholdvariationwith phasefor a signalat maskeronset wasthe sameas that for the long-delaycondition.An explanationfor the variationof signal detectability with maskerlevelis proposed that refersexplicitlyto the compressive input-output characteristic of the basilar membrane at intermediate levels. PACS numbers: 43.66.Dc,43.66.Mk, 43.66.Ba[HSC] -- 15and45dB.Hefoundanincrease ofovershoot from0 INTRODUCTION The term overshoot refersto the fact that a brief signal is sometimes harderto detectif it is presentedat the onset of a maskerthan if it is presentedwith a long onsetdelay. About 30 yearsagothe first experiments to investigatethis phenomenon were conducted(Samoilova,1959; Scholl, 1962;Zwicker, 1965a,b;Elliott, 1965). In the last decade overshoothasreceivedrenewedattentionwhich can partly be attributedto the fact that no satisfactoryexplanation has beenofferedso far (e.g., Baconand Viemeister,1985; Baconand Moore, 1986;Carlyon, 1987, 1989;Carlyon and Sloan, 1987; Champlin and McFadden, 1989; McFadden, 1989;Bacon, 1990;McFaddenand Champlin, 1990;Bacon and Smith, 1991;Schmidtand Zwicker, 1991;Carlyon and White, 1992;Baconand Takahashi, 1992). It is known that overshootis influencedby many parameterslike frequencyanddurationof the signal,spectral composition and level of the masker,prior stimulation, hearing impairment, and even consumptionof aspirin by subjects.Since theseeffectsand the possiblerelation to neuralshort-termadaptationhavebeenreviewedin several recentpublications(e.g.,Baconand Smith, 1991;Carlyon and White, 1992; Bacon and Takahashi, 1992), we will to about10 dB for an increasein the maskerspectrumlevel up to q-5 dB. For all highermaskerlevels,up to a spectrum level of 45 dB, overshoot remained constant at this value. Fastl (1976) useda uniform maskingnoiseof 16kHz bandwidthand an 8-kHz signalof 2-msduration.The spectrumlevel in the 8-kHz regionrangedfrom -15 to q-25 dB. Similar to Zwicker's results, overshoot increased for spectrumlevelsup to 5 dB and remainedconstantfor all higherlevels.The amountof overshoot,however,was much larger than reportedby Zwicker, beingas much as 30 dB. The results of more recent measurements revealed a differentdependence of overshooton maskerlevel.Bacon (1990) measuredovershootfor a 10-ms4-kHz signal.The noise had a bandwidth of 8 kHz and was presentedat spectrumlevelsbetween-- 10 and 50 dB. Overshootgrew nonmonotonically with level,reachingan asymptoteof 10 to 15 dB at spectrumlevels of 20 to 30 riB, before decreas- ing at higher levels.For a majority of the subjects,overshootat a 50-dB spectrumlevel was as small as it was at --10 dB. Carlyon and White (1992) measuredovershootwith concentrate on the influence of masker level on overshoot. broadbandmaskersfor signal frequenciesof 2.5 and 6.5 Zwieker (1965a) measuredovershootof a 5-kHz 2-ms kHz. At 2.5 kHz, overshootvalueswereindependentof the signalin a broadbandnoiseat spectrumlevelsbetween masker level. At 6.5 kHz, overshootfor most subjectswas largerat a maskerspectrumlevelof 25 dB than at either5 a)Allcorrespondence should besenttoArminKohlrausch at thisaddress. or 45 dB. However,two of the subjectsshowedeither no 2192 J. Acoust.Soc.Am.95 (4),April1994 0001-4966/94/95(4)/2192/10/$6.00 @ 1994Acoustical Societyof America 2192 leveldependence or an inversebehavior,with the smallest overshootoccurringat the intermediatemaskerlevel. Bacon and Takahashi (1992) collected data for maskerspectrumlevelsbetween20 and 40 riB. Their signal was a 10-ms 4-kHz sinusold.For the majority of their subjects,overshootdecreasedmonotonicallywith increasing maskerlevel.The rangeof individualovershootvalues at the lowest masker level was 7 to 26 dB, while at the highestlevel it was 2 to 14 dB. The presentstudy is concernedwith level effectsin overshootconditionsfor running-and frozen-noisemaskers.There are two reasonsto investigateovershooteffects for frozen-noisemaskers.We know from recent experimentsthat thresholdsof short signalsin frozennoisecan be much lower than thosein runningnoiseand that they vary stronglywith the temporalpositionof the signalin the noiseand with the signalstartingphase (Langhans, 1991; Langhansand Kohlrausch, 1992). These resultsfor frozen-noise maskers indicate that with the lack of statis- tical variability in the maskerwaveform,additionalcues like slighttimbredifferences causedby the additionof the signal can be evaluatedby the subjects.It is unclear whetherthe useof suchsubtlecuescouldhelp to detecta signalplacedcloseto the onsetof a frozennoiseand thus lead to smallerovershootvaluesthan for a running-noise masker. A second motivation for the use of frozen noise we startedthe experiments described in thispaper,we were interestedin checkingwhether this discrepancybetween the model predictionand the majority of data is a consequenceof usingfrozennoisein the simulationand running noise in the measurements,or whether the model is incor- rectin thisrespect.• I. METHOD A. Stimuli and apparatus All stimuli were generateddigitally,convertedto analog voltageswith a two-channel,16-bit D/A converterat a samplingrate of 30 kHz, and low-passfilteredat 10 kHz. To obtain the noisemasker,a complexspectrumof frequencycomponentsbetween20 Hz and 10 kHz was constructed.All frequencycomponentshad a constantamplitude, the phase values were randomly distributed between0 and 2•r. The inverseFFT of this spectrumresults in a cyclicnoisesignalconsisting of 2•5 samples witha periodof 1.1 s. In experiments with frozennoise,a fixed 300-mspart of this time waveformwas used.This fixed noisesamplewaspresentedin all intervalsof all runsfor a given signaldelay (for more details,seeSec.I C below). In contrast, in experiments with running noise, different300-mssectionsof this 1.1-snoisewerepresented to the listenerin every interval of a run. Thesenoisebursts were copied from the long noisewaveform by randomly stemmed from theoretical predictionsderived from a modelfor temporalprocessing. This modelhas beenappliedto nonsimultaneous-masking dataandothertemporal effectsand is describedin detail elsewhere(Pfischel,1988; Kohlrauschet al., 1992; Dau etal., 1994). The relevant choosing oneof the2• values asstarting point.Forevery elements of the model are a (linear) ba.,filar-membrane levels of - 10, 10, and 30 dB. All noise bursts had a rect- stagefollowedby a seriesof adaptive-gain-control stages angulartemporalenvelope,i.e., the durationof the onset ramp was only influencedby the analoglow-passfilter at with time constants between 5 and 500 ms that result in a loglikecompression for signalswith a stationaryenvelope. Fast fluctuationsin the envelopeof an input signallike in amplitudemodulationor the fluctuationsof a noiseenvelope are, however,transformedmore linearly and lead to largervariationsin the internalrepresentatiion of the stimulus.The detectioncriterionusedto predictthresholdsis derivedfrom the stationarybehaviorin sucha way that, for a stationarysignal,thejust-noticeable changein levelis 1 dB at all levels. For a signal placed at the onset of a frozen-noise masker, the model predictsa somewhatlower threshold than for a signalwith a long onsetdelay. The reasonfor this lowerthresholdis that both the signaland the masker are lesscompressedat the onset of the masker than at a later point. The addition of the signal at t'.heonset of the maskerthus leadsto a greaterchangein the internalrepresentationthan for onepositionedlater in the masker,and thus the signalis detectedmore easily. A negativeovershoot(or undershoot):fornoisemaskers has beenobservedonly in somelistenersunder a lim- run, a new 1.1-snoisewaveformwas calculatedby usinga new set of phasevalues. The noise maskerswere presentedat three different levels:30, 50, and 70 dB SPL, corresponding to spectrum 10 kHz. The signalwas a 5-kHz sinusoldwith a total duration of 2 ms. It was gated with l-ms raised-cosine on and off ramps.The signalwasaddedto the maskerwith oneof the following onset delays:0, 3, 8, 18, 198, or 278 ms. The onsetdelay of the signalwas definedas the time between maskeronsetand the onsetof the signal'sramp. In the running-noise measurements, the startingphaseof the signal, definedwith respectto its envelope,was alwayszero. In the frozen-noisemeasurements, the startingphasewas an additionalparameter. At each masker level, the following three measurementswere performed: ( 1) Thresholdsof a signalplacedwith a 278-msdelay in the noise bursts.This measurementwas performedfor nine valuesof the startingphasebetween0 and 2•r in frozen noise(signalphaseincreasedin stepsof •r/4), and in additionfor the signalin runningnoise.The valuesfor the starting phases0 and 2•r, althoughphysicallyidentical, were measuredseparatelyto testthe reliabilityof the data. ited set of conditions (e.g., Osman and Raab, 1963; Wilson (2) Measurement of the thresholds for six temporal and Carhart, 1971;Bacon, 1990, Figs. 2 and 3; Baconand Smith, 1991, Fig. 1; Baconand Takahashi, 1992, Fig. 2), and little emphasishasbeengivento thesefindings.When positionsof the signal,with onsetdelaysbetween0 and 278 ms. This measurementwas performedfor frozen noisefor thosetwo signal-phase valueswhich gavethe highestand 2193 J. Acoust.Soc. Am., Vol. 95, No. 4, April1994 R. von Kiltzingand A. Kohlrausch: Effectof maskerlevel 2193 the lowest thresholds in the first measurement. (3) Thresholdsof a signalplacedcloseto the onsetof the maskerburst at that temporalpositionfor which the highestthresholdshad beenfoundin the secondmeasurement. The onsetdelay was chosenfor each subjectseparately and was either 0 or 3 ms. This measurement again wasperformedfor nine valuesof the startingphasein frozen noiseand in additionfor the signalin runningnoise. All measurements reported in this paper were performed diotically, i.e., masker and signal were presented identically to both ears of the subjects. 0.05 0.00 -0.05 0.0 0.2 0.4 0.6 0.8 s 1.0 0.0 0.2 0.4 0.6 0.8 s 1.0 o.o 0.2 0.4 0.6 0.8 s 1.0 0.05 0.00 ko.05 0.05 0.00 B. Subjects and procedure -0.05 The experimentswere performedby three subjects, who had experiencein psychoacoustic experimentsand had severalhoursof trainingin overshootconditions.The subjects rangedin agefrom 24-28 years.The subjects were Time seated in a sound-insulated booth and listened to the stimFIG. 1. Illustrationof the generationof the frozen-noisemaskerbursts. uli via headphones(Beyer DT 880M with diffuse-field The top row showsthe 1-sfrozennoise.The signalonset(markedby the equalizer).The computercontrolledthe experimentsand arrow at the bottom) occurred at 0.6 s. The middle trace shows the 300-ms masker burst which was used in the measurement with a 278-ms read the answersof the subjects. delay between masker and signal onset. The bottom row shows the The maskedthresholdswere measuredusing a stanmaskerfor the delayof 0 ms betweenmaskerand signal. dard three-intervalforced-choice(3IFC) procedurewith adaptivelevel adjustment.The three noiseintervalshad a durationof 300 ms eachand were separatedby interstim- the desiredmasker-signalonset delay, a 300-ms section surroundingthis time point was cut out of the long noise ulus intervalsof 500 ms. In one randomlychoseninterval, signalwith a rectangularwindow. This procedureis illusthe signalwas addedto the noisemasker.The task of the subjectwas to specifythe interval containingthe signal. tratedin Fig. 1. The top row showsthe 1-snoisewherethe The signal level was varied accordingto a two-down arrow (bottom) marks the positionof the signal.The situation for a 278-ms delay betweenmasker and signal is one-upprocedure(Levitt, 1971). A run beganwith the shown in the middle trace and the masker used for the signallevelwell abovethe expectedthreshold.The initial step size for the level changewas 10 dB and was halved O-msdelay is shownin the bottom trace. (dB measure)after each secondturnaroundpoint of the II. RESULTS signallevel.When the stepsizebecamesmallerthan 1 dB, it was set to the constant value of 1 dB and the run was extended over 20 more trials. The threshold levels were calculatedas mediansignallevelwithin theselast 20 trials of eachrun. Typically, 4 to 6 reversalsof the signallevel occurredduring the 20 trials. For eachconditionthis procedurewas repeatedfour times in differentsessions. Data in the figuresare calculatedas the median valuesfrom these four runs and are shown together with the correspondinginterquartileranges. A. Signal detectability as a function of masker level In Fig. 2, the influenceof signal phaseon masked thresholds in a frozen-noise masker of 50 dB $PL is shown For differenttemporalpositionsof a short signal in frozen noise, thresholdscan vary by more than 10 dB (Langhans,1991). This valuecorresponds to the order of for a short-delay(a) and a long-delaycondition(b) by the opensymbols.The closedsymbolsto the left in both panels indicatethe thresholdsin runningnoisefor the samesignal delay. The differentsymbolsindicateindividualdata from the three subjects. The resultsfor frozen noisealways show a maximum at phase•r/2 and a minimum at phase3•r/2. In the longdelaycondition,the thresholddifferencebetweenthesetwo phasesis 8 to 11 dB. The signalthresholdsfor running noiseat thisdelayare higherthan the arithmeticaverageof thosefor frozennoise,but they do not form an upperlimit overshoot in running noise. If the signal was added to a for the frozen-noise fixed frozen-noisesample with a variable onset delay, a possibleovershooteffectcouldbecomeinvisibledueto this fine-structureeffect.Hence, it is very important that the perimentsat 1 kHz (Langhansand Kohlrausch, 1992). The measurementsin the short-delaycondition were performedfor that signalpositioncloseto the maskeronset that led to the highestthreshold(cf. Fig. 3). This time point was 0 ms for subjectsAS and SM and 3 ms for subjectRK. The phase dependenceof the thresholdsin frozen noiseis much smallerthan in the long-delaycondition and amountsfor two of the subjectsto no more than 3 dB. The variation of thresholdwith phaseis largestfor C. Construction of the frozen-noise measurements signalis alwaysadded to the samefine structurein all frozen-noise measurements. To meet this requirementa noise sampleof 1 s was generated.Within this noisea fixedtemporalpositionfor the signalwaschosenwherethe phasedependence of the threshold wasmaximal(i.e.,about10dB).2According to 2194 J. Acoust. Soc. Am., Vol. 95, No. 4, April 1994 thresholds as we found in recent ex- R. von Klitzingand A. Kohlrausch:Effect of masker level 2194 7O 70 dBSPI dB SPL 65 o QJ • 6O 60 55 5O 50 I ]•/2 (a) I I I • I 3]z/2 I i I 2]z 10 20 200 (a) ms 280 Signal 13clay SignalPhase 70 d@SPL 70 6• dB SPL 65 .• •0 • o... ......._ 60 oJ • 55 50 5O (b) 0 •[/2 • 3•[/2 2• Signal Phase FIG. 2. Thresholds of a 2-ms5-kHz signalin 300-msburstsof running and frozennoise.The flat-spectrum noisehad a bandwidthof 10 kHz and a levelof 50 dB SPL. Opensymbols indicatethe resultsin frozennoise, filledsymbols indicaterunning-noise data.The abseissa denotes thestarting phaseof the signalfor the frozen-noise results.Circlesare usedfor subject RK, diamonds forsubject AS andsquares forsubject SM. Median andinterquartiles of four runsfor eachparametervalueandsubjectare shown.(a) Data for a short-delay condition(onsetdelayof 0 msfor AS and SM and of 3 ms for RK) and the (b} the data for a Iong-dday dB SPL 6O The amount of overshoot for the frozen noise was 55 terminedby placingthe signalat variousdelaysrelativeto themaskeronset.Theresultsareshownfor eachsubject in a separatepanelof Fig. 3. The measurements were performedfor thephasevaluesrr/2 (squares)and3rr/2 (tri- 5O angles)whichcorresponded to the maximumand the min- The samemeasurements were repeatedat two other 2195 J. Acoust.Soc.Am.,Vol.95, No.4, April1994 200 ms 280 70 that subject(SM) whois mostsensitive in the short-delay condition.In this condition,the running-noisethresholds form an upperlimit for the frozen-noisevalues. masker levels,namely 30 dB SPL and 70 dB SPL. The 20 Signal Detay 65 subjectsmeasureda smallerovershootof 3 to I0 riB. The maximum in thresholddid not alwaysoccur for the O-ms delay [seeFig. 3(a)]. I 10 (b) conditions(onsetdelayof 278 ms}. imumsignalthresholds, respectively. Experiments at the phase3rr/2 showan overshoot of up to 15 riB. At rr/2 all I 0 I I o lO (c) 20 200 ms 280 Signal Detay FIG. 3. Maskedthresholds of the2-mssignalin frozennoiseasa function of the signaldelay.The ,-nasker levelwas50 dB SPL. Squaresindicatethe resultsfor thesignalphaserr/2 andtriangles thedatafor thephase3rr/2. Median and interquartilesof four runs for each parametervalueand subjectare shown.(a) Resultsfor subjectRK. (b) Resultsfor subject AS. (c} Resultsfor subjectSM. R. yonKlitzing andA. Kohlrausch: Effectof rnaskerlevel 2195 80 30 dB ][ 2O ]•..70dB dB o 25 z 60 • 50 •/SOdB 20 15 ;0 30 -- _•_e..• .....• • •/30dB 10 20 /.0 60 dBSPL NaskeF 0 10 20 200 Level ms 280 Signal Delay FIG. 4. Maskedthresholds of the 2-mssignalin frozennoiseasa function of the masker-signal delayfor the maskerlevels30, 50, and 70 dB SPL. Squaresindicatethe resultsfor the signalphasest/2 and trianglesthe data for the phase3sr/2. The data showthe medianand the corresponding interquartilerangesof the 12 individualthresholdvalues. FIG. 5. Masked thresholdsof the 2-ms signal in running noiseas a functionof the maskerlevel.Signalthresholds are expressed as 10 log(E/ No). The parameteris the signaldelay. Squares:short delay condition (medianof valuesfor 0, 3, and 8 ms delay); triangles:18-msdelay; circles:48-msdelay;invertedtriangles:long-delayposition(medianof valuesfor 198-and278-msdelay).Medianandinterquartiles of fourruns for eachparametervalueare shownfor one subject(RK). resembleone periodof a sinusold.(2) The minimumof thispatternoccursfor the70-dBmaskerat a phasevalueof (RK) and includedmore maskerlevelsand signaldelays than the main experiments. The parameterin the figureis the onsetdelay of the signal.At all signaldelays,thresholds vary nonmonotonicallywith maskerlevel. The variation is largestfor the short-delaycondition(squares),but evenat delaysof more than 200 ms (invertedtriangles), the relativedetectabilityvariesby 6 dB with maskerlevel. All three curvesrepresenting a signalpositionwithin the samethree subjectsas beforeparticipatedin the measurementsdeterminingovershoot,while the influenceof signal phaseon the thresholdwas studiedwith two of thesesubjects (AS and RK). The effect of signal phaseon the thresholds at these two levels can be summarized as fol- lows: (1) As at 50 dB SPL, the frozen-noise thresholds 3•r/2 (as for the 50-dB masker), but for the 30-dB masker first 50 ms after masker onset reach their maximum at a at a phasevalueof 5•r/4. (3) The total variationof thresholdswith phaseis about 8 dB at 30 dB and about 5 dB at 70 dB and is the samein the short- and the long-delay positions.(4) Running-noisethresholdsare generallyin the upper-half range of the correspondingfrozen-noise maskerlevelof 50 dB. In the long-delaycondition,thresholdsincreaseup to a maskerlevel of 60 dB. thresholds. The experimentsin frozennoisehave shownthat both the maskerleveland the startingphaseof the signalaffect overshoot.In Fig. 6, the amount of overshootis representedasa functionof themaskerlevelfor bothphases•r/2 (squares) and 3•r/2 (triangles) in frozen noise and for running noise (circles). Individual overshootvalueswere calculatedin the followingway: The referencevalue was the medianthresholdvalue of the eight singlemeasurements for the two longestdelays 198 and 278 ms. The Overshootin frozen noise was again determinedfor two differentphasevaluesof the signal.The resultsfor all threemaskerlevelsare summarizedin Fig. 4. Data in this figureare calculatedas medianvaluesfrom the 12 individual runsof the threesubjectsand are showntogetherwith the corresponding interquartileranges.Squaresindicateresultsfor a signalphaseof •r/2 and trianglesindicateresults for a signalphaseof 3•r/2. Compared to the resultsat 50-dB masker level, the B. Overshoot as a function of masker level value of overshoot resulted from the difference between curves at 30- and 70-dB masker level are flatter. At 30 dB, this referenceand the highestthresholdwithin the first 10 the signal is even easierto detect for a positionat the ms after masker onset. If thresholds were lower in the masker onsetthan at the end. For the signal phaseof 3•r/2 (triangles), such a negativeovershootoccursnot only in the median values, but also in the individual data of each onset region than the referencevalue, the lowest value was subject. The figurealsoindicatesa stronglevel nonlinearityin signal detectability,in particular at short signal delays. This nonlinear aspectof overshootmeasurementsis emphasizedin Fig. 5, wheresignalthresholds[10 log(E/No)] in runningnoisehavebeenplottedas a functionof masker level. These measurementswere performedby one subject overshoot 2196 J. Acoust. $oc. Am., Vol. 95, No. 4, April 1994 chosen to indicate the maximal undershoot. The data pointsin Fig. 6 representthe averagesof theseindividual values. In all three conditions, overshoot is maximal at the intermediate masker level. In frozen noise, the maximum overshootwasobtainedwith a signalphaseof 3•r/2 andthe minimum overshootwas obtained with a signal phase of •r/2. The overshootfor running noise fell between these two extremes. At the lower masker level of 30 dB $PL, the R. von Klitzingand A. Kohlrausch:Effect of masker level 2196 15 dB 10 oi -5 30 50 Masker dB 70 Level FIG. 6. Overshoot as a function of the masker level. Circles indicate the magnitudeof overshootin runningnoise.Squaresindicatethe frozennoisedata for the signalphaserr/2 and trianglesindicatethe resultsin frozennoisefor the phase3zr/2. Averagevaluesof •:hethreelisteners. averageovershootvalues are always negative.From the nineindividualvaluesat this level,only thosein running noiseand in frozennoisefor phase•r/2 of subjectRK are FIG. 7. Vectordiagramsillustratingthe additionofa maskerand a signal of the samefrequency.The solidvectorin both diagramsrepresents IIhe masker,andthethin vectorsrepresent thesignalsfor relativephasevalues of 0 (continuous)andof rr (dashed).The circlesaroundtheoriginof the maskervectorindicatethejust noticeable changein vectorlength.In the left panelthiscriterionis a changeby a factortwo (the diameterhastwice the lengthof the maskervector). The length of the two signalvectors differsby a factorof 3 (9.5 dB). In the right panelthe thresholdcriterion is a changeby a factor of 4. The length of the two signalvectorsnow differs bya factor • (4.4dB). An important additionalparameterin the vectordiagram is the relative length between the masker and tlhe signalvectorfor a signalat threshold.If the criterioncor- dB SPL, the amount of overshoot for the three conditions responds a doublingof the lengthof the maskervector(left panel in Fig. 7), the amplituderelation betweenthe inphaseandthe out-of-phase signalat thresholdis a factorof threeor 9.5 dB. The righthalf of Fig. 7 showsthesituation for a thresholdcriterionthat is equivalentto quadrupling the lengthof the maskervector.In sucha situation,tlhe amplituderelationbetweenthe in-phaseand the out-of- is systematically orderedfor all subjects, thisis not thecase phase signal isonly•, whichcorresponds to a leveldiffer- at the two extreme masker levels. enceof 4.4 dB. Becausehigher thresholdsoccur, for instance,for a signalplacedat the onsetof the masker,tlhe right half of Fig. 7 might representa short-delaycondition, whereasthe left half might representa long-delaycondi- positivewith a maximum of 3 dB. Small overshootvalues are alsofoundat 70 dB SPL, althoughtlaevaluesare on averagea few dB higherthan at 30 dB SPL and they are negativein onlyoneindividualcase(subjectAS for signal phase•r/2). While at the intermediatemaskerlevel of 50 III. DISCUSSION tion. A. Effects of signal phase The resultspresentedin this paperdemonstrate that overshoot values of more than 10 dB can be obtained for a frozen-noise masker at an intermediate masker level. In addition,the nonmonotonicdependenceof overshooton maskerlevelis similarfor frozen-andrunning-noise maskers. A third result is that the amount of overshoot in frozen noisedependsuponthe startingphaseof the signal,althoughthe overallleveldependence is similar. Thresholdsof a shortsignalin frozen,noiseare influencedby thedeterministic interaction of maskerandsignal and thusdependon the signal'sstartingphase.The phase Thuswith increasing signal-to-noise ratioat threshold, the variation of thresholdwith the starting phasedecreases.Another consequence of this reasoningis a differencein overshootfor the in-phaseand the out-of-phase signal.The threshold amplitudefor thein-phase signaldiffers by a factor of 3 betweenthe two vector diagrams which corresponds to a thresholdchangeof 9.5 dB. The threshold of theantiphasic signalchanges onlyby• which correspondsto 4.4 dB. Suchan idealizedpictureis in line with our resultsilar the 50-dB-SPLfrozen-noise masker(Figs. 2 and 3). For the large onsetdelay, the thresholdsof the individualsub- dependence can be illustratedby usinga vector diagram (Fig. 7). The maskerand the signalare bothrepresented by a vectorand the anglebetweenthe two corresponds to the (relative) phaseof the two finestructures.The masked jectsvary with the signalphaseby 8 to 11 dB [Fig. 2(b)]. For signals placedat theonset,wherethresholds in general are muchhigher,the variationwith phaseis muchsmaller [2 to 6 dB, cf. Fig. 2(a)]. The variationat the onsetposithreshold is reached if the sum of the two vectors matches tion is largestfor subjectSM whosethresholdsare lowest. a certain thresholdcriterion (circle in Fig. 7). If the Finally, the amount of overshootis on average6 dB maskerandthe signalare in phase,thiscriterionis reached smallerfor a startingphaseof •r/2 (highersignal-to-noise witha minimal length ofthesignal vector. Since a decreaseratioat [hreshold)thanfor a startingphaseof 3½r/2(lower in the vector length cannot be detectedfor frozen-noise signal-to-noiseratio), a value in agreementwith the above maskers andshortsignals, thresholds arehighestfor a sig- example. nal addedin antiphaseto the masker(ct: Langhansand At the masker levels of 30 and 70 dB SPL, the influKohlrausch, 1992). enceof the signalphaseis similar for the onsetand the 2197 J. Acoust.Soc.Am.,Vol.95, No.4, April1994 R. yonKlitzing andA. Kohlrausch: Effectof maskerlevel 2197 long-delayposition.Sucha resultis expectedfrom the reasoningwith the vectordiagram, if the thresholdcriterion is natedby the high-threshold fibers.Sincethesefibersexhibit only a smallonsetresponse,overshootdecreases for higher not stronglyelevatedfor the signalat the maskeronset. levels. And indeed, overshoot in all conditions, for frozen noise as well as for runningnoise,is smallerthan at 50 dB (with one exceptionin the resultsof subjectSM for frozen noise and phase•r/2). This resultagreeswith other recentstudies showinga relativemaximumof the overshootin runningnoisefor spectrumlevelsbetween20 to 30 dB (Bacon, 1990; Carlyon and White, 1992; Bacon and Takahashi, 1992), althoughthe maximum for our subjectsseemsto occurcloserto a spectrumlevelof 10 dB (of. Fig. 6). The different level dependencein the results reported by Zwicker (1965a) and Fastl (1976) might be a consequence of applyinga differentexperimentalprocedure(method of adjustment),as was pointedout by Bacon (1990). For the 30-dB-SPL masker, we observea negative overshootfor the signalwith startingphase3•r/2 in frozen noise.The effectis not very large, but it is the kind of behaviorthat we expectedfrom our model predictions mentioned in the Introduction. In contrast to our recent simulations mentioned in footnote 1, however, the results with runningnoisealso indicatea negativeovershootfor two of the subjects.Thus it appearsthat a negativeovershootat low levelsis not a particularconsequence of using a deterministic masker, as our model predictions suggested. B. Adaptation in auditory-nerve fibers and overshoot Overshootand especiallythe nonmonotoniclevel effects have been related by severalauthorsto peripheral adaptationin the two populationsof auditory nerve fibers (e.g., Champlin and McFadden, 1989; Bacon, 1990; McFaddenand Champlin, 1990). In simplifiedterms,this argument is as follows. Fibers in the eighth nerve can be subdividedin two groupsaccordingto their physiological properties(e.g., Libermanand Simons,1985). About 75% of the fibershavea low thresholdand a high spontaneous activity, while the remainderhave high thresholdsand a low spontaneous activity. Thesetwo groupsof fibersalso differ in their onsetresponse. Low-thresholdfibersexhibit a large onsetresponsethat decreases over time due to adaptation. High-thresholdfibers,on the other hand, show only a small onsetresponseand thus much lessadaptation (Rhode and Smith, 1985, but see below). For low to mod- eratemaskerlevels,the responseof the low-thresholdfibers is dominant.These fibers respondto a short increment (corresponding to the "signal" in an overshoot measure- ments) in the level of a pulsedstimulus (the "masker") with a constantincrementin firing rate, independentof the temporal position of the increment relative to the "masker" onset (e.g., Smith and Zwislocki, 1975). Since the responseto the maskeradaptsover time, the neural signal-to-maskerratio (Bacon, 1990) increaseswith increasing onset delay, corresponding to a decrease in masked threshold. At high masker levels, the lowthresholdfibers are saturatedand the responseis domi2198 J. Acoust.Soc. Am., Vol. 95, No. 4, April 1994 Althoughit is generallybelievedthat peripheraladaptation is involved in overshoot, the above scheme is cer- tainly not sufficientto explainall or evena major part of the observedeffects(for a thorough discussion,see McFaddenand Champlin,1990). Probablymostpuzzlingare the changes in overshoot dueto aspirinconsumption, prior noiseexposureor with sensory-hearing impairedsubjects (Bacon and Takahashi, 1992; Champlin and McFadden, 1989;McFaddenand Champlin,1990). In all theseconditions, overshootis reducedor disappears,becausesubjects becomemore sensitivein the short-delaycondition. McFadden and Champlin (1990) discussedthe idea that in aspirinas well as in TTS overshootexperiments,an unusual rebalancingeffect betweenlow- and high-threshold auditory nervefiberscould take place.If, at intermediate maskerlevels,subjectscoulduseinformationsuppliedby high-threshold fibers(insteadof informationsuppliedby low-thresholdfibers as in "normal" overshootexperiments), overshootwould disappear. However,this idea as well as the aboveargumentfor the decreaseof overshootat high levelshaveto be questionedin the light of recentphysiologicalresultsby Relkin and Doucet (1991). They showed that with a sufficient interstimulustime of at least500 ms (a typicalintervalin psychophysicalovershootmeasurements),high-threshold fibersshowa similaronsetresponseaslow-thresholdfibers. In the next sectionwe presentan alternative,probably related,idea, which links signaldetectabilitydirectlyto the nonlinearinput-output characteristicof the basilar membrane.Although we haveto admit that our consideration is not an explanationof overshootper se, it seemsto offer a differentview on someof the above-mentioned puzzling effectsin the short-delayconditions. C. Basilar-membrane nonlinearities and signal detectability In the input-outputcharacteristic of the basilarmembrane (BM), we can roughlydiscriminatethree level regions (e.g., Johnstoneet al., 1986; Yates, 1990; Ruggero and Rich, 1991). At low levels up to about 40 dB above absolutethreshold,the responseis rather linear. At intermediate levels between about 40 and 80 dB, the function becomesstronglycompressive.This compressivebehavior is assumedto be a consequence of the saturationof the activeprocesses and to work instantaneously. At the highest levelsabove about 80 riB, the function again tends to be more linear. This compressivecharacteristicis only ob- served for signal frequenciesclose to the resonantfrequencyof a particularplaceon the BM. For frequenciesof about half an octave (or more) below the resonant fre- quency, the input-output characteristicbecomeslinear (e.g., Sellick et al., 1982; Johnstoneet al., 1986). This frequency-dependent function of the mechanicsis also reflected in the rate-intensityfunctionsof auditory nerve fibers (Yates et al., 1990). R. von Klitzingand A. Kohlrausch:Effect of maskerlevel 2198 Such a compressivecharacteristicshouldbe reflected in a reduceddifferentialsensitivityfor levelsin the midlevel range (midlevel"hump"). This changein sensitivitywill, however,only be substantial(in terms of changesin the measuredthresholds),if the detectiontaskrequiresa large changein amplitude,as for shortstimulusdurations.Only the case of sensorineuralhearing loss, the input-output characteristic of the BM will be more linear than in a healthyear (Yates, 1990) and (ideally) no midleveldeterioration should occur. Overshoot measurements for sub- jects with permanentsensorineural hearinglossby Bacon and Takahashi (1992) revealed maximum overshoot val- been observedin intensity discrimination.experimentsfor clicks (Raab and Taub, 1969). Here, a maximal value for uesof only 5 dB, while a controlgroupof normal-hearing subjectsreachedovershootvaluesbetween7 and 26 dB. The differencewasprimarily due to lowerthresholdsin tke short-delayconditionfor the hearingimpairedsubjects.In addition, the short-delaythresholdsvaried much more linearly with maskerlevel for thesesubjects(of. Fig. 1 in AI/I Bacon and Takahashi, 1992). then the differentialeffectof compression will be large comparedto the limitedresolutionof psychoacoustic measurements. Such an increased threshold at intermediate levels has of 1.2 was observed for a click level of 40 dB SL. This valuewaslargerby a factor3 than the resultsat clicklevels of either 20 or 80 dB SL. Carlyon and Moore (1984) reported a much larger midlevel deteriorationin intensity discrimination for 30-mssignalsthanfor 225-mssignals.A similar resultis mentionedby Plack and Viemeister(1992) who observedin preliminaryexperiments that the "midlevelelevationin the Weberfractionis alsolargerfor brief signals"(p. 1909). This reasoningis alsoof relevancefor maskingexperimentsusingshortsignals.The peakamplitudeof a 5-kHz signalof 2-msduration,placedwith a largeonsetdelayin runningnoisewith a spectrumlevelof - 10 dB, is about38 dB comparedto the typical peak amplitudeof the noise within the 5-kHz auditory filter of 30 dB (of. the very instructivewaveformpicturein Zwicker, 1965a,Fig. 12). At a 20-dB highernoiselevel the BM becomescompressive. Assuming,for simplicity,a linear characteristicfor input amplitudesup to 50 dB and a compression ratio of 1:3 for peak amplitudesabove 50 dB, the differencebetweennoisepeaksand signalpeakat the BM outputis no longer 8 dB (as at the lower level), but only 2.7 dB. This differenceoccursbecausethe noiseamplitudesare still in the rangeof linear transformation,while tlhehighersignal amplitude undergoescompression.In order to reach an 8-dB differenceafter compression, the signallevel would haveto be increasedby another 16 dB. As we have shownin Fig. 6, the thresholdsignal-tonoiseratio of a 2-ms signalvariesnonlinearlywith masker levelfor all onsetdelaysandis highestin thelevelregionof strongBM compression.However, the an•tountof nonlinearity decreases with increasingonsetdelay. If one wants to explainthis resultpurelywith the input-outputcharacteristicof the BM onehasto assumethat tlheamplification causedby the activeprocesses is reducedin the courseof a longer noisemaskerand that the input-output characteristic becomes more linear. Such an idea has been discussed by Schmidt and Zwicker (1991) who mention the time constants of the outer hair cells of several milliseconds and the delayedactivationof the efferentsystemas possible sourcefor such a change.In our opinion, however,the current experimental data seem insuffici[entto state a changeof BM mechanicswithin a few ten milliseconds. Nevertheless,the influenceof the various manipulations on signal detectability in the short-delay condition showsmany parallelswith what would be predictedfrom BM properties.If the cochlearamplifieris disruptedas in 2199 J. Acoust.Soc. Am., Vol. 95, No. 4, April 1994 McFaddenand Champlin (1990) observedthat after aspirinadministration, the absolutethresholdfor tonalsignals was increasedand the sensitivity in the short-delay overshootconditionwas improved,i.e., thresholdswere lowered. As the authors discussthoroughly,aspirin is known to affect the active micromechanicalprocesses on the BM in a similar way as doesTTS and permanenthearing loss,i.e., it makesthe input-outputcharacteristic of the BM more linear. BM mechanicsalso offers an explanationfor the smallerlevelnonlinearitiesobservedwith longersignals.If the durationof a signalis increased,its thresholdlewfl decreases. As soonas the peakamplitudeof the signaland the typical noisepeaksare of similar magnitude,the compressionwill affectboth the noiseand the signalsimilarly and the nonlineargrowth of the maskedthreshold--and thus overshoot--at intermediatelevelswill disappear. The last applicationof this argumentis the specific influenceof maskerbandwidth on short-delaythresholds in overshootconditions.Usually, no overshootis observed for maskers with a bandwidth of two critical bands or less (e.g., Zwicker, 1965b;Fastl, 1976/77; Baconand Smith, 1991). From experimentsthat combinedon-frequency narrow-band maskerswith flanking bands of higher and lower frequencyit was concludedthat the presenceof maskercomponents abovethe on-frequency bandwascrucial for a largeovershoot(McFadden, 1989;Schmidtand Zwicker, 1989). On the other hand, Carlyon and White (1992) pointedout that at a frequencyof 6.5 kHz, which is higher than thoseusedin the other cited studies,a substantial overshoot was obtained for a relative masker band- width of only 0.3 times the centerfrequency. A narrow-bandsignalproducesmain excitationon the BM at the place tuned to its center frequencyas well -'rs excitation at more basal placeson the membranethat are tuned to higher frequencies(sometimesreferred to as upward spreadof excitationor as accessory excitation).Since the characteristicof the BM is linear for the accessory excitation, a short increment in a narrow-band noise of intermediatelevel can be detectedmore easilyat a place tuned to a frequencyabovethe noise,sincethere, no compressionoccurs.As long as the subjectscan rely on information from off-frequencychannelstuned abovethe signal frequency, signal thresholdsshould grow linearly with masker level. In overshoot measurements,two conditions have been R. von Klitzingand A. Kohlrausch:Effect of maskerlevel 21õ0 usedthat reducethe availabilityof accessory excitationand that lead to largeovershootvalues:(a) Increasingthe relative masker bandwidth beyond about 20% to 30% (Zwicker, 1965b; McFadden, 1989; Bacon and Smith, 1991;Carlyonand White, 1992); (b) choosinga veryhigh frequencyfor the on-frequency band (Carlyon and White, 1992). A third condition has been applied in intensitydiscriminationexperiments,where similar nonlinearlevel lyon, M. v. d. Heijden, D. Hermes, N. Versfeld and, in particular,the two reviewersSid Baconand Craig Champlin provided helpful commentson earlier drafts of this paper. Technical assistanceby M. v. d. Heijden, L. Martensand G. Kirschmann-Schr6der in preparingthe figures is gratefullyacknowledged. This studywassupportedby a grant from the Deutsche Forschungsgemeinschaft (Ko 1033/3-1). extension ofthismodelto running-noise maskers revealed that effectsas for overshootare observedfor high-frequency IA recent for sucha maskerno undershoot is predicted(Dau, 1992;Dau et al., signals.Surroundingthe signalwith a notchednoiseand 1994). Thresholdsremainconstantfor all (simultaneous)positionsof thus masking spread of excitation, reducesthe perforthe signal.We explain this behavioras follows. For a running-noise mance at intermediatelevelsmarkedly comparedto the masker,the detectionis limitedby the statisticalpropertiesof the noise. condition without a notched noise (Carlyon and Moore, The statisticalvariation in the internal representationis larger for the 1984). onsetpart and smallerfor the later part of the noise.In a parallelmanner, the increasein internalexcitationcausedby the additionof the signalvarieswith its temporalposition.In summary,the relativedetectability of the signalis not increasedcloseto the maskeronset. We would like to finishwith the qualificationthat, in order to understandthe overshootphenomenon,it is certainly not sufficientto consideronly BM properties.For example,overshootwith broadbandon-frequencymaskers as well as the probablyrelatedmidlevelhump in intensity discriminationare largestat frequenciesabove 1 to 2 kHz. 2Thechoice ofsucha temporal position forthesignal wasinspired byour previousexperiencewith frozen-noisemaskerswhere the most interesting effects in frozen noise--like the monaural-diotic threshold difference-•occurredat such positions (Langhans and Kohlrausch, 1992). No BM measurements are known to us that show a differ- ent behavior of active processesbelow and above 1 kHz. Neural adaptation processesor even more central partsof the hearingpathwaymustbe takeninto accountin understanding overshoot.However, we think that on the basisof our aboveargumentsthe contributionof BM mechanicsis of greaterrelevancefor this phenomenonthan arguedin other recentarticles(e.g., McFadden, 1989;Bacon and Smith, 1991;Carlyon and White, 1992). Bacon, S. P. (1990). "Effect of masker level on overshoot," J. Acoust. Soc. Am. 88, 698-702. Bacon,$. P., and Moore, B.C. J. (1986). "Temporaleffectsin simultaneouspure-tonemasking:effectsof signalfrequency,masker/signalfrequencyratio and maskerlevel," Hear. Res. 23, 257-266. Bacon,S. P., and Smith, M. A. (1991). "Spectral,intensiveand temporal factorsinfluencingovershoot,"Q. J. Exp. Psychol.43, 373-399. Bacon, S. P., and Takahashi, G. A. (1992). "Overshoot in normal- hearingand hearing-impaired subjects,"J. Acoust.Soc.Am. 91, 28652871. Noteaddedin proof.It wasbroughtto our attentionby Brian C. J. Moore that somerecentpublicationsconclude that active processesmight be less involved in cochlear tuningat low frequencies(Rosenand Stock,1992;Wilson, 1992). Rosen and Stock measured auditory filter bandwidths between 125 and 1000 Hz at various levels. Gener- ally, filter bandwidthsincreasedwith level, with the strongesteffectat 1000Hz and little or no effectat 125 Hz. In the conclusion,they refer to a remark in a reviewarticleby Wilson that at low frequencies"there appearsto be a much smallerinvolvementof the active process"(Wilson, 1992, p. 81). Unfortunately,thisstatementby Wilsonis not sup- ported by a reference.Nevertheless,such a difference would be consistent with our discussion of the role of basilar-membrane nonlinearities in overshoot situations, sinceovershootseemsto be absentat low frequencies. ACKNOWLEDGMENTS Bacon, S. P., and Viemeister, N. F. (1985). "The temporal courseof simultaneoustone-on-tonemasking," J. Acoust. Soc. Am. 78, 12311235. Carlyon,R. P. (1987). "A releasefrom maskingby continuous,random, notched noise," J. Acoust. Soc. Am. 81, 418-426. Carlyon,R. P. (1989). "Changesin the maskedthresholds of brieftones producedby prior burstsof noise,"Hear. Res. 41, 223-236. Carlyon, R. P., and Moore, B.C. J. (1984). "Intensity discrimination:A severedeparturefrom Weber'slaw," J. Acoust.Soc. Am. 76, 13691376. Carlyon,R. P., and Sloan,E. P. (1987). "The overshooteffectand sensory hearingimpairment,"J. Acoust. Soc. Am. 82, 1078-1081. Carlyon, R. P., and White, L. J. (1992). "Effect of signalfrequencyand masker level on the frequencyregionsresponsiblefor the overshoot effect," J. Acoust. Soc. Am. 91, 1034-1041. Champlin, C. A., and Mc Fadden, D. (1989). "Reductionsin overshoot followingintensesoundexposures," J. Acoust.Soc.Am. 85, 2005-2011. Dau, T. (1992). "Der optimaleDetektor in einemComputermodellzur Simulationyonpsychoakustischen Experimenten," unpublished master thesis,University of G/Sttingen. Dau, T., Piischel,D., and Kohlrausch,A. (1994). "A quantitativemodel of the "effective"signalprocessing in the auditory system:I. Model structure"(in preparation). Elliott, L. L. (1965). "Changesin the simultaneousmaskedthresholdof brief tones," J. Acoust. Soc. Am. 38, 738-746. We would like to thank our colleagues from the psychoacousticresearchgroup at the Universityof G6ttingen for the continuoussupportin the courseof the experiments and for many fruitful discussions about the content of this paper. Special thanks to Andres Sander and Stefan Miinkner for spendingmany hoursin the listeningbooth. Fang-GangZeng and Bob Carlyon both directedour attention to the close relation between basilar-membrane nonlinearity and the two-populationhypothesis.R. Car2200 J. Acoust.Soc. Am., Vol. 95, No. 4, April1994 Fastl, H. (1976). "Temporal masking effects: I. Broad band noise masker," Acustica 35, 287-302. Fastl, H. (1976/77). "Temporal maskingeffects:II. Critical band noise masker," Acustica 36, 317-331. Johnstone, B. M., Patuzzi, R., and Yates, G. K. (1986). "Basilar mem- branemeasurements and the travellingwave,"Hear. Res.22, 147-153. Kohlrausch,A., Piischel,D., and Alphei, H. (1992). "Temporal resolution and modulationanalysisin modelsof the auditorysystem,"in The .4uditoryProcessing of Speech,editedby M. E. H. Schouten(Mouton De Grnyter, Berlin), pp. 85-98. Langhans,A. (1991). "Psychoakustische Messungen und Modellvorstel- R. von Klitzingand A. Kohlrausch: Effectof maskerlevel 2200 lungenzur Wahrnehmung reproduzierbarer undstatistisch fiuktuierender Signale,"Ph.D. thesis,Universityof 06ttingen. Langhans, A., and Kohlrausch,A. (199:2)."Differences in auditoryper- Ruggero,M. A., and Rich, N. (1991). "Furosemidealtersorganof Co•rti mechanics:Evidencefor feedbackof outer hair cells upon the basilar formance between monaural and diotic conditions. I. Masked thresh- Samoilova,I. K. (1959). "Maskingof shorttonesignalsas a functionof the time intervalbetweenmaskedand maskingsounds,"Biophysics 4, olds in frozen noise," J. Acoust. Soc. Am. 91, 3456-3470. Levitt,H. (1971). "Transformed up-downmethodsin psychoacoustics," J. Acoust. Soc. Am. 49, 467-477. Liberman,M. C., and Simmons,D. (198S). "Applicationsof neuronal labelingtechniques to the studyof the peripheralauditorysystem,"J. Acoust. Soc. Am. 78, 312-319. McFadden, D. (1989). "Spectraldifferencesin the ability of temporal gapsto resetthe mechanisms underlyingovershoot,"J. Acoust.Soc. Am. 85, 254-261. McFadden,D., and Champlin,C. A. (1990). "Reductionsin overshoot during aspirin use,"J. Acoust.Soc. Am. 87, 2634--2642. Osman,E., and Raab, D. H. (1963). "Temporalmaskingof clicksby noisebursts," J. Acoust. Soc. Am. 38, 1939-1941. Plack, C. J., and Viemeister,N. F. (1992). "The effectsof notchednoise on intensitydiscrimination underforward masking,"J. Acoust.Soc. Am. 92, 1902-1910. Piischel, D. (1988). "Prinzipien der zeitlichen An:•lysebeim H6ren," Ph.D. thesis,Universityof Gfttingen. Raab, D. H., and Taub, H. B. (1969). "Click-intensitydiscrimination withandwithouta background maskingnoise,"J. Acoust.Soc.Am. 46, 965-968. Relkin,E. M., and Doucet,J. R. (1991). "Recoveryfrom prior stimulation. I: Relationshipto spontaneous firing ratesof primary auditory neurons," Hear. Res. 88, 215-222. Rhode,W. S., and Smith,P. H. (1988). "Characteristics of tone-pip response patternsin relationshipto spontaneous rate in cat auditory nerve fibers," Hear. Res. 18, 159-168. Rosen,S., and Stock, D. (1992). "Auditory filter bandwidthsas a function of levelat low frequencies(125 Hz-I kHz)," J. Acoust.Soc.Am. 92, 773-781. 2201 J. Acoust.Soc.Am.,Vol.95, No.4, April1994 membrane," J. Neurosc. 11, 1057-1067. 44-52. Schmidt,S., and Zwicker, E. (1991). "The effectof maskerspectral asymmetryon overshootin simultaneousmasking,"J. Acoust. Soc. Am. 8% 1324-1330. Scholl, H. (1962). "Das dynamischeVerhalten des Gehfrs bei der UnterteilungdesSchallspektrums in Frequenzgruppen," Acustica12, 101107. Sellick, P.M., Patuzzi, R., and Johnstone,B. M. (1982). "Measurement of basilar membranemotion in the guineapig usingthe M6ssbauer technique,"J. Acoust.Soc.Am. 72, 131-141. Smith, R. L., and Zwislocki,J. J. (1975). "Short-termadaptationand incrementalresponsesof singleauditory-nervefibers," Biol. Cybern. 17, 169-182. Wilson, J.P. (1992). "Cochlearmechanics,"in AuditoryPhysiology and Perception,editedby Y. Cazals,L. Demany,and K. Horner (Pergamon, Oxford), pp. 71-84. Wilson, R. H., and Carhart, R. (1971). "Forward and backward masking: Interactionsand additivity," J. Acoust. Soc.Am. 49, 1254-1263. Yates,G. K. (1990). "Basilarmembranenonlinearityand its influenceon auditory nerve rate-intensityfunctions," Hear. Res. 50, 145-162. Yates, G. K., Winter, I. A., and Robertson, D. (1990). "Basilar membranenonlinearitydeterminesauditoryrate-intensityfunctionsand cochleardynamicrange,"Hear. Res.45, 203-220. Zwicker, E. (1965a). "Temporaleffectsin simultaneous maskinghy white-noisebursts," J. Acoust. Soc. Am. 37, 653-663. Zwicker, E. (1965b). "Temporal effectsin simultaneousmaskingand loudness,"J. Acoust. Soc. Am. 38, 132-141. R. vonKlitzing andA. Kohlrausch: Effectof maskerlevel 2201