perceptual “vowel spaces” of cochlear implant users: implications for the study of auditory...

Perceptual ‘‘vowel spaces’’ of cochlear implant users:Implications for the study of auditory adaptationto spectral shift

James D. Harnsberger, Mario A. Svirsky,a),b) Adam R. Kaiser,a) David B. Pisoni,a)

Richard Wright,c) and Ted A. Meyera)

Speech Research Laboratory, Department of Psychology, Indiana University, Bloomington, Indiana 47405

~Received 2 May 2000; accepted for publication 12 December 2000!

Cochlear implant~CI! users differ in their ability to perceive and recognize speech sounds. Twopossible reasons for such individual differences may lie in their ability to discriminate formantfrequencies or to adapt to the spectrally shifted information presented by cochlear implants, abasalward shift related to the implant’s depth of insertion in the cochlea. In the present study, weexamined these two alternatives using a method-of-adjustment~MOA! procedure with 330 syntheticvowel stimuli varying inF1 andF2 that were arranged in a two-dimensional grid. Subjects wereasked to label the synthetic stimuli that matched ten monophthongal vowels in visually presentedwords. Subjects then provided goodness ratings for the stimuli they had chosen. The subjects’responses to all ten vowels were used to construct individual perceptual ‘‘vowel spaces.’’ If CI usersfail to adapt completely to the basalward spectral shift, then the formant frequencies of their vowelcategories should be shifted lower in bothF1 andF2. However, with one exception, no systematicshifts were observed in the vowel spaces of CI users. Instead, the vowel spaces differed from oneanother in the relative size of their vowel categories. The results suggest that differences in formantfrequency discrimination may account for the individual differences in vowel perception observedin cochlear implant users. ©2001 Acoustical Society of America.@DOI: 10.1121/1.1350403#

PACS numbers: 43.66.Ts, 43.71.Es, 43.71.Ky@RVS#

afidcrse,a

fuinrm

as

thio

ticepalenies.-

lyis-

to-ces-non

ncyhan

ted

are

ec-dobutrodeere-arenor-ch

larmu-fre-therTos isre-

oo

lec

I. INTRODUCTION

Although cochlear implants allow profoundly depeople to hear, cochlear implant users show a very wrange of speech perception skills. The most successfulchlear implant users can easily hold a face-to-face convetion, and they can even communicate on the telephondifficult task because there are no visual cues availablebecause the acoustic signal itself is highly degraded~Gsto-ettneret al., 1997!. On the other hand, the least successcochlear implant users have a difficult time communicateven in a face-to-face situation, and can barely perfoabove chance on auditory-alone speech perception t~Dorman, 1993!.

One long term goal of our research is to understandmechanisms that underlie speech perception by cochlearplant ~CI! users and, in so doing, gain an understandingthe individual differences in psychophysical characteriswhich may explain individual differences in speech perction with a CI. It is important to remember that electrichearing as provided by a cochlear implant is quite differfrom normal acoustic hearing. One important difference lin listeners’ ability to discriminate formant frequencieKewley-Port and Watson~1994! report difference limens between 12 and 17 Hz in theF1 frequency region for highly

a!Also at Devault Otologic Research Laboratory, DepartmentOtolaryngology-Head and Neck Surgery, Indiana University SchoolMedicine, Indianapolis, IN 46202.

b!Also at Department of Biomedical Engineering, and Department of Etrical Engineering, Purdue University, West Lafayette, IN 47907.

c!Now at the University of Washington, Seattle, WA 98195.

2135 J. Acoust. Soc. Am. 109 (5), Pt. 1, May 2001 0001-4966/2001/1

eo-a-a

nd

lg

ks

em-fs-

ts

practiced normal hearing listeners. In theF2 frequency re-gion, they found a frequency resolution of approximate1.5%. For cochlear implant users, formant frequency dcrimination depends on two factors: the frequency-electrode map that is programmed into their speech prosor, and the individual’s ability to discriminate stimulatiopulses delivered to different electrodes. It is not uncommfor some cochlear implant users to have formant frequedifference limens that are one order of magnitude larger tthose of listeners with normal hearing, or even more~Nelsonet al., 1995; Kewley-Port and Zheng, 1998!. It is reasonableto hypothesize that cochlear implant users with such limifrequency discrimination skills will find it quite difficult toidentify vowels accurately because formant frequenciesimportant cues for vowel recognition.

Another important difference between acoustic and eltric hearing is related to the finding that cochlear implantsnot stimulate the entire neural population of the cochleaonly the most basal 25 mm at best, because the electarray cannot be inserted completely into the cochlea. Thfore, cochlear implants stimulate cochlear locations thatmore basal and thus elicit higher pitched percepts thanmal acoustic stimuli. For example, when the input speesignal has a low frequency peak~e.g., 300 Hz!, the mostapical electrode is stimulated, regardless of the particufrequency-to-electrode table employed. The neurons stilated in response to this signal may have characteristicquencies of 1000 Hz or even higher. This represents a raextreme modification of the peripheral frequency map.the extent that the auditory nervous system of CI useradaptable enough to successfully ‘‘re-map’’ the place f

ff

-

213509(5)/2135/11/$18.00 © 2001 Acoustical Society of America

banod

be

eoso

hisctthht

ra

tdthth

-eitgs

gela

fen

eldderre

mit

eahtr

octr

ctdr-ewm

tlethIt

toerar-

rage,tionto-

tionm-ers

of, a

ys ofilarll as

e isslyabil-to

fre-heun-by

onscies

byes.la-ea-

ad-lger-ivenhy-uldcesdy,

erstion

ts

tu-ere

is-by

noight

quency code in the cochlea, the basalward shift provideda CI should not hinder speech perception. On the other han inability to adapt and re-map the place frequency cmay severely limit speech perception in CI users and mayan important source of individual differences in speech pception~Fu and Shannon, 1999a, b!. Although there is con-sensus in the literature that cochlear implants stimulate nrons with higher characteristic frequencies than thstimulated by the same sound in normal ears, there is ctroversy about the amount of this possible basalward sFor example, Blameyet al. ~1996!, in a study where CI userwith some residual hearing were asked to match the elecal percepts in one ear to the acoustic percepts in the oear, concluded that the electrode positions that matcacoustic pure tones were more basal than predicted fromcharacteristic frequency coordinates of the basilar membin a normal human cochlea. However, Blameyet al. ~1996!also acknowledge that ‘‘the listeners may have adapted tosounds that they hear through the implant and hearing aieveryday life so that simultaneously occurring sounds intwo ears are perceived as having the same pitch.’’ Onother hand, Eddingtonet al. ~1978! conducted the only experiment we are aware of, where a unilaterally deaf voluntreceived a cochlear implant and was asked to match the pof acoustic and electric stimuli while still in the operatinroom, before he had much of a chance to adapt to the baward shift in the way described by Blameyet al. ~1996!.Eddingtonet al. ~1978! concluded that their pitch-matchindata were ‘‘consistent with frequency versus distance rtionships derived from motion of the basilar membrane.’’

Several previous studies have addressed the issue oaptation to changes in frequency-to-electrode assignmfor cochlear implant users. Skinneret al. ~1995! showed thatusers of the SPEAK stimulation strategy identified vowbetter with a frequency-to-electrode table that mappemore restricted acoustic range into the subject’s electrothan the default frequency-to-electrode table. The expmental table that resulted in better vowel perception repsented a more extreme basalward shift in spectral infortion than the default table, suggesting that listeners wcochlear implants can indeed adapt to such shifts, at lwithin certain limits. Another study that demonstrates tadaptation ability of human listeners in response to specshifts was conducted by Rosenet al. ~1999!, who usedacoustic simulations of the information received by a hypthetical cochlear implant user who had a basalward speshift of 6.5 mm on the basilar membrane~equivalent to 1.3–2.9 octaves, depending on frequency!. Initially, the spectralshift reduced word identification in normal-hearing subje~1% correct, as compared to 64% for the unshifted contion!, but after only three hours of training, subjects’ perfomance improved to 30% correct. Whether or not such pformance represents the maximum possible by CI usersnot addressed in this study, given the relatively short tispent in training.

Recently, Fuet al. ~submitted! performed an experimenin which the frequency-to-electrode tables of three cochimplant users were shifted one octave with respect totable they had been using daily for at least three years.

2136 J. Acoust. Soc. Am., Vol. 109, No. 5, Pt. 1, May 2001

yd,ee

r-

u-en-ft.

ri-er

edhene

heinee

erch

al-

-

ad-ts

sa

esi--

a-hst

eal

-al

si-

r-ase

areis

important to note that this one-octave shift was in additionthe original shift imposed by the cochlear implant. Aftthree months of experience with the new table, it was appent that adaptation was not complete because, on avesubjects did not reach the same levels of speech percepthat they had achieved before the table change. Takengether, these previous studies show that auditory adaptato a modified frequency map is possible but it may be liited, depending on the size of the spectral shift that listenare asked to adapt to.

In the present study, we investigated the adaptationhuman listeners to a basalward shift using a new paradigmmethod-of-adjustment~MOA! procedure. This methodologwas used to obtain maps of the perceptual vowel spaceadult, postlingually deafened cochlear implant users. Simtasks have been used with normal hearing listeners as weCI users~Johnsonet al., 1993; Hawks and Fourakis, 1998!.In this task, subjects select the region of theF1 –F2 planethat sounds to them like a given vowel, and the procedurrepeated for ten English vowels. This task simultaneouassesses a cochlear implant user’s auditory adaptationity, by comparing the locations of his/her selected regionsthose selected by normal hearing listeners, and his/herquency discrimination skills, by examining the spread of tselected regions. More specifically, a listener who wasable to adapt to the basalward spectral shift introducedtheir cochlear implant would be expected to select regiwhose centers are systematically shifted to lower frequenwith respect to the regions of the vowel space selectednormal hearing listeners in mapping their vowel categoriThe extent to which cochlear implant listeners show retively normal vowel category centers could be used as a msure of their adaptation to basalward spectral shift.

An alternative to this hypothesis predicts that the spreof selected regions~category sizes!, as well as category centers, would increase~or ‘‘smear’’! as a result of perceptuaadaptation. The resulting vowel categories may be lar~show greater spread! reflecting the cochlear implant listener’s need to map a greater range of frequencies to a gvowel. To differentiate between the spectral smearingpothesis and the frequency discrimination explanation worequire a longitudinal study of the changes in vowel spaand frequency discrimination. For the purposes of this stua simple shift hypothesis limited to vowel category centwas tested and compared with the frequency discriminahypothesis.

In addition to the MOA task, two other perceptual teswere administered to CI users, anF1 jnd test with syntheticvowel stimuli and a closed set identification test with narally produced vowels. Taken together, these measures wintended to investigate the role of formant frequency dcrimination and auditory adaptation in vowel perceptionCI users.

II. EXPERIMENT

A. Methods1. Participants

Forty-three Indiana University undergraduates withreported history of speech or hearing problems and e

2136Harnsberger et al.: Perceptual ‘‘vowel spaces’’

rd

2137 J. Acoust. S

TABLE I. Demographic and other information for patients with cochlear implants. CNC5Percentage of cor-rectly identified words on CNC word lists~male talker!. Each patient was administered three 50-wolists.*CI6, CI7 Progressive; CI8, Progressive childhood.

Patient Age Gender

Age at onsetof profound

deafness Age at implantationImplant use

~years! CNC

CI1 67 F 43 61 6 52%CI2 35 M 29 31 3 68%CI3 37 M 34 36 1 46%CI4 74 F 27 71 2 25%CI5 63 M 56 57 5 11%CI6 70 M * 66 3 32%CI7 68 F * 65 2 1%CI8 58 M * 52 6 0%

rsalalpacar

ngxil

nd

ctathnd

phe

ueadiaareo

ndn

ls

able

tsCIco-hehereeIS

cyearallyperaretput

toelsareisCISif-itsb-

adult cochlear implant~CI! users, all monolingual speakeof English, participated in this experiment. The normhearing participants consisted of 20 males and 23 femranging in age between 18 and 28. The normal-hearingticipants were recruited to represent the dialect of AmeriEnglish spoken in central Indiana with a common inventoof vowels. Only normal-hearing listeners who reported livitheir entire lives in central Indiana were included in this eperiment. Central Indiana was defined in terms of a 60-mradius around Indianapolis, roughly covering the Midladialect as described by Wolfram and Schilling-Estes~1998!.This criterion was used to exclude two other regional dialefound at the northern and southern extremes of the sThese other regional dialects are reported to differ fromMidland dialect in terms of vowel quality and degree atype of dipthongization~Labov, 1991!. For participating intwo 1-h sessions, the participants received either $7.50hour or two credits towards their research requirement if twere enrolled in an undergraduate psychology class.

The CI users were recruited from the population of adpatients served by the Department of Otolaryngology-Hand Neck Surgery at the Indiana University School of Mecine in Indianapolis. The demographics of the CI usersgiven in Table I, while information concerning their cochleimplants is provided in Table II. All of the CI users wernative speakers of American English, with the exceptionCI1, who was a native speaker of British English. British aMidland American English are not reported to differ substa

oc. Am., Vol. 109, No. 5, Pt. 1, May 2001

-esr-ny

-e

tste.e

ery

ltd-re

f

-

tively from one another in vowel quality for the ten voweused in this study~Gimson, 1962; Pilch, 1994!. Thus theAmerican English vowel spaces were deemed an acceptbenchmark for CI1 as well as the other CI users.

All of the CI users had received their cochlear implanat least one year prior to participating in this study. Theusers differed from one another in terms of the type ofchlear implant they received: Five were users of tNucleus-22 or Nucleus-24 device, programmed with eitthe SPEAK strategy or the MPEAK strategy, while thrwere users of the Clarion device, programmed with the Cstrategy. The SPEAK strategy~Skinner et al., 1994! filtersthe incoming speech signal into a maximum of 20 frequenbands, which are associated with different intracochlstimulation channels. Typically, six channels are sequentistimulated in a cycle, and this cycle is repeated 250 timessecond. The channels to be stimulated during each cyclechosen based on the frequency bands with the highest ouamplitude. In contrast, the CIS strategy~Wilson et al., 1991!as implemented in the Clarion device filters the signal ineight bands, one for each stimulation channel. All channare sequentially stimulated with pulses whose amplitudesdetermined by the filters’ outputs. The stimulation cyclerepeated at a rate of at least 833 times per second. Thestrategy differs from the SPEAK strategy in its use of a dferent stimulation rate, fewer stimulation channels, and instimulation of all channels in a cycle rather than only a suset of the available channels.

frommappedlationI5 who

TABLE II. Information concerning the patients’ cochlear implants. Insertion depth indicates the distancethe most apical electrode to the round window. Frequency range indicates the acoustic range that wasinto the available stimulation channels. Note: All patients with Clarion devices used the CIS stimustrategy, while those patients with Nucleus devices used the SPEAK strategy, with the exception of Cused the MPEAK strategy.

Patient Implant typeInsertion

depth~mm!Frequencyrange~Hz!

Number ofactive

channelsStimulation

mode

CI1 Clarion 1.0 25 350–5500 8 BipolarCI2 Clarion 1.2 25 350–6800 8 BipolarCI3 Nucleus 24 21.25 120–7800 21 MonopolarCI4 Nucleus 22 19.75 170–4800 14 BP11CI5 Nucleus 22 22 300–4000 20 BP11CI6 Nucleus 22 19.75 150–9200 19 BP11CI7 Nucleus 22 19 120–6300 18 BP13CI8 Clarion 1.0 25 350–5500 8 Bipolar


oleagn

ara-ods

rosoca

trath

ae

ci-pthla

leol

thrtasthsaa

orth

dy

atl-e

lycus

t they an

the

ofco-esti-im-thens

ciseain-ter-ted

var-inthe

just

en

entanngof-

lis--

stIV.

330ng are-

ostee

d in

e

The CI users also differed from one another in termsthe depth of insertion of the electrode array in the cochThe array’s depth of insertion, in turn, determines the mnitude of the basalward spectral shift induced by the implaIt is possible to roughly estimate the size of this basalwshift for an individual CI user with three pieces of informtion: the location of the electrodes, the frequency to electrmapping used by the cochlear implant’s speech procesand estimates of the characteristic frequency of the neustimulated by a given electrode pair. The speech procesof the participants in this study divide the acoustic frequenspectrum into channels. Each channel is specified byacoustic frequency range that is assigned to a pair of incochlear electrodes. Low frequencies are mapped to thecal electrodes, while high frequencies are mapped tobasal electrodes.

In the present study, the basalward spectral shift wcalculated for two channels for every subject, the channcorresponding to 475 Hz and 1500 Hz. These frequencorrespond to theF1 andF2 of a neutral vowel for an average male speaker. First, the place of stimulation for a scific channel was defined as occurring half way betweenelectrodes for that channel. When stimulation was bipo~i.e., both electrodes were intracochlear!, the place of stimu-lation was defined as occurring half way between the etrodes for that channel. For patients who received monopstimulation~i.e., the return electrode was extracochlear! thelocation of electrical stimulation was considered to be atintracochlear electrode. Second, the intraoperative repoinsertion depth was used to adjust this place estimate bon the depth of electrode insertion. Cochlear lengths ofsubjects were not measured individually. Instead, the baward shift was calculated assuming that all subjects haderage sized cochleas with a length of 35 mm~Hinojosa andMarion, 1983!. Third, the electrically assigned frequency fthis location was calculated as the geometric mean offrequency boundaries defining the channel being studiethe speech processor. Finally, the characteristic frequencthe neurons stimulated by a given electrode was calculfrom Greenwood~1961! and compared to the frequency caculated in step three. This discrepancy for each CI user, msured in Hz and octave, is shown in Table III. Alternativethe shift is also reported in terms of the location differenbetween electrical and acoustic stimulation, in mm. The

TABLE III. The estimated basalward shift for the eight CI users for twcochlear locations corresponding to 475 Hz and 1500 Hz. Shifts are lieither as a frequency shift~Hz!, an octave shift, or as the location differencin electrical and acoustic stimulation~mm!.

Patient

475 Hz 1500 Hz

Hz Octave mm Hz Octave mm

CI1 235 0.7 2.7 409 0.3 1.4CI2 235 0.7 2.7 409 0.3 1.4CI3 1301 1.9 8.1 2320 1.3 6.1CI4 1719 2.2 9.2 2068 1.3 6.0CI5 645 1.3 5.2 1669 1.1 5.1CI6 1128 1.8 7.6 1542 1.2 5.3CI7 1707 1.6 9.1 2334 2.1 7.4CI8 235 0.7 2.7 409 0.3 1.4


fa.-t.d

eor,nsrs

yn

a-pi-e

slses

e-er

c-ar

eofedel-

v-

einof

ed

a-,ee

of Greenwood’s equation is based on the assumption thaaverage characteristic frequency of neurons stimulated belectrode pair placedx mm from the round window is thefrequency that would cause maximum displacement ofbasilar membrane at the samex mm from the round window.Clearly, these are only rough estimates of the amountbasalward shift. In particular, the estimate of a 35-mmchlea may lead to substantial overestimates or undermates of the actual basalward shift. Future studies mayprove the precision of these estimates by measuringlength of each individual cochlea using 3D reconstructioof CAT scans~Skinneret al., 1994; Kettenet al., 1998!; byusing the same 3D reconstructions to obtain more preestimates of electrode location in the cochlea; and by obting physiological and behavioral data that may help demine the characteristic frequency of the neurons stimulaby different electrode pairs.

2. Stimulus materials

a. Method-of-adjustment task.The stimulus set con-sisted of 330 synthetic, steady-state isolated vowels thatied from one another in their first and second formants0.377 Bark increments. The vowels were generated usingKlatt 88 synthesizer~Klatt and Klatt, 1990!. The Bark incre-ment size was chosen as a close approximation of thenoticeable difference for vowel formants of Flanagan~1957!.The F1 andF2 values for this stimulus set ranged betwe2.63 Z ~250 Hz!–7.91 Z ~900 Hz! and 7.25Z ~800 Hz!–15.17 Z ~2800 Hz!. These ranges were chosen to represthe full range of possible values for speakers of Americand British English, and were successfully used in pilotithe present experiment and in an earlier method-adjustment study of vowel perception in normal-hearingteners~Johnsonet al., 1993!. All of the other synthesis parameters for this stimulus set also followed Johnsonet al.~1993!. The formulas for calculating the values of the morelevant synthetic parameters are summarized in TableTheF0 parameter was varied to generate two sets of thestimuli, one representing a male voice and one representifemale voice. All of the synthetic speech sounds were p

dTABLE IV. The formulas for an important subset of the parameters usegenerating the synthetic stimulus sets.Bn5the bandwidth of formantsF1 –F4.

Parameter Formulas

Duration 250 msF0 Male: 120 Hz over the first half, dropping to 105 Hz at the

endFemale: 186 Hz over first half, dropping to 163 Hz at thend

F3 $(0.522* F1)1(1.197* F2)157% or$(0.7866* F1)2(0.365* F2)12341% Hza

F4 3500 Hz or (F31300 Hz), whichever higherB1 $29.271(0.061* F1)2(0.027* F2)1(0.02* F3)% HzB2 $2120.222(0.116* F1)1(0.107* F3)% HzB3 $2432.11(0.053* F1)1(0.142* F2)1(0.151* F3)% HzB4 200 Hz

aThe first formula was used for the half of the grid with higherF2 values,while the second was used for the half of the grid with lowerF2 values.


hadti

w

iothticdBern

eurik

li-vethrea

l-1-

oaea

aninrainte

meic

nar

etreeo

/dey

weltons.

ehowOne

toried

gs

orytheheedthens,rs’

rlier

al-d tothat

in

the

-seteache atuli

ereasnta-

ent-

dg-h

be--ere

onecor-

fore70

muliliby

sented at a 70 dB C-weighted SPL listening level. Tstimuli were presented over Beyer Dynamic DT-100 hephones for normal-hearing listeners, and over an AcousResearch loudspeaker for CI users.

b. Vowel identification task.The vowel identificationtask employed a closed-set procedure that used nine voin an ‘‘h-vowel-d’’ format. The stimuli were digitized fromthe female vowel tokens of the Iowa laserdisc~Tyler et al.,1987!. Only steady-state vowels~no diphthongs! were usedfrom this stimulus set. There were three separate productof each vowel. Listeners were administered three listsconsisted of five repetitions of each vowel and three practokens. The stimuli were presented at a level of 70C-weighted SPL over an Acoustics Research loudspeak

c. F1 jnd task. The stimuli for this task were sevesynthetic three-formant vowels, with anF2 value of 1500 Hzand anF3 of 2500 Hz.F1 varied linearly, from 250 Hz forstimulus 1 to 850 Hz for stimulus 7. Three-formant raththan two-formant vowels were used in order to measplace-pitch discrimination with more realistic and speechlstimuli. The stimuli were created using the Klatt 88~Klattand Klatt, 1990! speech synthesizer software. Voicing amptude increased linearly in dB from zero to steady state othe first 10 ms and back to zero over the last 10 ms ofstimulus. Total stimulus duration was 1 s. Stimuli were psented at a level of 70–72 dB C-weighted SPL overAcoustic Research loudspeaker.1

3. Procedures

a. Method-of-adjustment task.The procedures variedslightly for each participant group in the study. Normahearing participants were tested in a quiet room in twosessions. The second session took place approximatelyweek after the first session. In the first session, normhearing participants completed the method-of-adjustmtask using one of the synthetic stimulus sets, either the mor the female voice set. In the second session, participcompleted the method-of-adjustment task with the remainstimulus set. The experiment was balanced for the ordewhich the two stimulus sets were presented to the normhearing participants. In contrast, the CI users were testedquiet room or a sound-attenuated chamber, in a singlesession, varying in length by individual CI user betweenand 3 h. Given the length of time CI users required to coplete the MOA task and other demands on their time, threceived only one of the two stimulus sets, the male voset.

Each participant was presented with a two-dimensio~15 rows and 22 columns! visual grid centered in a computedisplay screen. The grid consisted of the 330 synthstimuli described above. A single English word appeaabove this grid, constituting the target stimulus for a givtrial. The visual target stimulus for a given trial was oneten words, ‘‘heed,’’ ‘‘hid,’’ ‘‘aid,’’ ‘‘head,’’ ‘‘had,’’‘‘who’d,’’ ‘‘hood,’’ ‘‘owed,’’ ‘‘odd,’’ and ‘‘hut,’’ each ofwhich contained one of the ten vowels under study, /i/,(/,/e/, /}/, /,/, /#/, /Ä/, /u/, /)/, and /o/. Subjects were instructeto search the grid, playing out individual sounds until th


e-

cs

els

nsate

.

ree

re-n

hnel-ntletsginl-ast

1-ye

l

icdnf

located one or more synthetic sounds that matched the voin the visual target stimulus. Subjects were not requiredsearch the grid exhaustively before making any selectioAfter making their choice~s!, subjects were asked to giveach synthetic sound a rating on a 1–7 scale, gradingclose a match the synthetic sound was to the target.repetition of the visual target stimulus set was presentedlisteners. The order of presentation of the stimulus set varandomly from participant to participant.

The particular stimuli chosen and their respective ratinwere used to calculatecategory centersand category sizesfor each vowel type for each listener group. The categcenter for a given vowel was determined by averagingF1 andF2 frequencies of the stimuli selected to match tgiven vowel, with their contribution to the average weightby their rating. The category sizes were computed fromstandard deviation of the selected stimuli in both dimensiotaking the weights into account. Normal-hearing listenecategory centers were expected to appear in theF1 by F2space in a similar arrangement to that observed in eavowel production studies with American English~i.e., Peter-son and Barney, 1952; Hillenbrandet al., 1995!. If the cat-egory centers of CI patients deviated from those of normhearing listeners, the extent of that deviation was expectebe dependent on the magnitude of the basalward shiftindividual listeners had to adapt to~see Table III!. Failure toadapt completely to this kind of stimulation would resultan overall space that is shifted lower inF1 andF2 ~i.e., thesubject would select synthetic stimuli that were lower inF1andF2 than those chosen by normal-hearing subjects forsame target vowel!.

b. Vowel identification task.The vowel identificationtask, administered only to the CI users, was a closedspeech perception task in which three separate tokens ofof nine /hVd/ tokens were presented in random order, ona time. The CI users had to say which one of the nine stimthey thought they heard by responding verbally. They winstructed to guess if they did not know which vowel wpresented. All subjects heard a total of at least 15 presetions of each vowel~except CI5 who heard 10!. The sub-jects’ responses were tabulated and scored for total percage of correct responses.

c. F1 jnd task. The F1 jnd task, administered only tothe CI users, required listeners to make an absolute jument. The stimuli for this task were labeled ‘‘1’’ throug‘‘7’’ in order of increasingF1. In this task, all stimuli wereplayed in sequence several times, so the subjects couldcome familiar with the stimuli. The stimuli were then presented ten times each in random order and subjects wasked to identify the stimulus that was presented usingof the seven responses. The subject’s response and therect response were displayed on the computer monitor bethe presentation of the next stimulus. After each block oftrials ~10 presentations of each of 7 stimuli!, the mean andstandard deviations of the responses to each of the 7 stiwere calculated. Thed8 for each pair of successive stimuwas calculated as the difference of the two means dividedthe average of the two standard deviations. Thesed8 mea-surements were then cumulated to calculate a cumulatived8


FIG. 1. The mean vowel space of normal-hearing listeners, calculated~a! using all of the ratings provided and~b! using only ratings of four or above.

ctul

on-

i

a

ald

mdin

leuein

go-fornry,anttionIn

eweleingenat-hecat-

ires.al-ent

ess-ical

curve, which provided an overall measure of the subjeability to discriminate and pitch rank the seven stim~Durlach and Braida, 1969; Levitt, 1972!. To calculate thecumulatived8 curves, we followed the common assumptithat the maximum possible value ofd8 was three. The average jnd~defined as the mean stimulus difference resultingd851) was calculated based on the cumulatived8 curve.Given that theF1 range spanned by the seven stimuli w600 Hz, the jnd was defined as~600/cumulatived8). At leasteight blocks of 70 trials were administered in order forsubjects to reach a plateau in performance as measurethe cumulatived8. The cumulatived8 reported here is theaverage of the best two blocks for each subject. The nortive mean cumulatived8 from this procedure was calculateto be 53 Hz, based on a pilot study with six normal-hearlisteners.

B. Results

1. Normal-hearing participants

The normal-hearing listeners were expected to sevowel category centers with first and second formant valthat corresponded to those typical in vowel production


’si

n

s

lby

a-

g

cts

American English. Figure 1 shows the mean vowel cateries obtained from the group of normal-hearing subjects,the male-voice stimulus set.2 Each category center is showalong with error bars denoting the ‘‘size’’ of each categothat is, the relative spread of the category in both formdimensions. The error bars represent the standard deviafrom the category centers in both formant dimensions.panel ~A! of this figure, on the left, all of the ratings havbeen used to calculate the center and size of all ten vocategories. Panel~B! of this figure, on the right, shows thvowel space of normal-hearing subjects calculated usonly ratings of four and above. The rating of four was chosbecause it was the highest rating that still allowed for cegory sizes to be calculated for all ten vowels of all of tnormal-hearing and CI participants. The center of eachegory was determined by weighting the Bark values~in eachdimension! of all synthetic stimuli that were chosen by thegoodness ratings, and then averaging the weighted valu

The perceptual spaces shown in Fig. 1 for the normhearing listeners demonstrate that the method-of-adjustmtechnique for measuring vowel categories can be succfully used to generate vowel spaces that display the typ

FIG. 2. A comparison of the vowel category centers from the MOA task~in bold, connected by straight lines! with the vowel production centers~in plain text,connected by dotted lines! from ~A! Hillenbrandet al. ~1995! and ~B! Peterson and Barney~1952!.


onnt-

hpo

ob

wobud

n-

thth

rs

e

ersth

y

n inalIndi-ivenel

in-reTherror

theachst

tion

intervowel relationships that have been observed inF1 byF2 spaces generated from vowel production data~Petersonand Barney, 1952; Hillenbrandet al., 1995!. For instance,front vowel category centers have a higherF2 than backvowel centers; high vowel category centers have a lowerF1than low vowel category centers. While no vowel productidata have been reported for English speakers from ceIndiana,3 Hillenbrandet al. ~1995! examined the vowel production spaces of 45 men, 48 women, and 46 children wwere native speakers of American English of the variety sken in southern Michigan. The study by Hillenbrandet al.~1995! was a replication and extension of the classic workthe acoustics of American English vowels carried outPeterson and Barney~1952!. The two panels in Fig. 2 showthe plots of the vowel production centers from these tstudies compared with the MOA vowel category centerstained from the normal-hearing subjects in the present st~calculated using ratings of 4 and above!. Panel~A! showsthe vowel category centers from the male talkers in Hillebrandet al. ~1995!. Panel~B! shows the vowel category centers from the male talkers in Peterson and Barney~1952!.While there are differences in the exact locations ofvowel perception and vowel production centers betweentwo studies, all three vowel spaces@the MOA perceptualvowel space, and the vowel production spaces of Peteand Barney~1952! and Hillenbrandet al. ~1995!# show acommon set ofF1 andF2 relations between vowels. Th

FIG. 3. The vowel space of patient CI1.



ral

o-

ny

o-y

-

ee

on

first and second formants from the normal-hearing listenwere significantly correlated with their counterparts in bothe Hillenbrandet al. ~1995! set (r 510.98,p,0.01 forF1;r 510.89, p,0.01 for F2) and the Peterson and Barne~1995! set (r 511.0, p,0.01 for F1; r 510.83, p,0.05for F2).

2. Cochlear implant patients

The results of the MOA,F1 jnd, and vowel perceptiontasks for the vowel spaces of the eight CI users are showFigs. 3–10. Unlike Figs. 1 and 2, Figs. 3–10 show individuCI user spaces, rather than an averaged group space.vidual spaces for CI users were presented separately gthe large potential individual differences in the CI vowspaces and the stated goal of this research in examiningdividual differences in the CI user population. Each figushows the labeled centers of the ten vowel categories.size of each category in each dimension is indicated by ebars. The centers of the categories were computed usingmean in each dimension weighted by the rating given to esynthetic stimulus, using ratings of only four or above, juas in the normal-hearing listeners’ vowel space in panel~B!of Fig. 1. The category sizes represent the standard deviafrom the category center in theF1 andF2 dimensions. To




h

tletionthep

bhte

aspanthC, dnhehthrly

ht, thels,min

byde

e pa-erfted

CIheyerers.beeirral

re-wefer-andse

the right of each CI user’s vowel space, the results of his/F1 jnd (jndF1), vowel perception~VOWEL!, and CNCword list ~CNC! tests are listed.

An examination of all eight vowel spaces reveals litevidence of any systematic shift due to a lack of adaptato a basalward spectral shift. The one possible exceptioCI1’s space. Instead, we find individual CI user spacesdiffer from the normal space in terms of the sizes of perctual categories, their degree of overlap, and the regionperceptual space that particular categories occupy. Theseservations are supported by measures of the differencestween CI and normal vowel categories. Table V lists tabsolute differences, in Bark, between the category cenof both normal and individual CI users, in bothF1 andF2.Positive differences indicate that the formant frequency (F1or F2) of the normal-hearing listeners for that vowel wgreater than the equivalent formant frequency of the CItient. If a shift were observable with a particular CI user, owould expect to see positive differences in one or both offormants of most of the vowels of that user. Of the eightusers, seven showed positive and negative differencespending on the vowel and formant in question. Only osubject, CI1, showed systematically lowered formants forcategory centers, as would be expected for a listener whonot adapted completely to the spectral shift introduced bycochlear implant. In Fig. 3, CI1’s categories are clea




er

nisat-

ofob-e-

ers

-eeIe-

erase

shifted toward the lowest formant values in the upper rigcorner, resulting in a more compressed space. Howevermagnitude of this shift was not the same for all ten voweor for both formants. The magnitude of the shift varied fro0.28Z to 3.53Z, with on average a greater shift observedF1 thanF2. Surprisingly, this shift was not accompanieda high jndF1 or low scores on the vowel identification anCNC word recognition tests. This subject did differ from thother seven in terms of her native dialect~British English!and in her professional background as a speech-languagthologist. It is possible, but difficult to determine, that hbackground biased her in some manner to provide a shivowel space in the MOA task.

The results shown in Figs. 3–10 indicate that mostusers adapted to their frequency-shifted input, although tvaried widely in the degree to which their vowel spaces wsimilar to the vowel spaces of normal-hearing listeneMoreover, this adaptation by CI users did not seem torelated to the depth of insertion of the electrode array of thimplant. Table III listed the predicted basalward spectshifts of the eight CI users inF1 andF2. If the individualdifferences in the magnitude of the predicted shifts weresponsible for individual differences in vowel spaces, thenwould expect to see a correlation between the actual difences between the normal and CI user category centersthe predicted shifts in each formant. In fact, neither of the




ce

95

TABLE V. The differences in Bark between normal and CI vowel categories, for each formant~F!, for individual vowel categories, and the mean differenacross all categories.

Patient F

Vowel category

Meani ( e } , # Ä u ) o

CI1 F1 0.43 1.06 1.79 2.37 3.53 0.62 2.82 0.45 1.32 1.52 1.5F2 0.28 0.8 1.09 1.46 1.46 0.39 0.57 1.41 2.35 0.66 1.0

CI2 F1 0.05 21.7 0.02 0.14 20.86 21.16 20.93 20.52 0.07 20.92 20.58F2 20.13 0.31 20.77 21.13 0.52 20.19 0.57 0.24 1.97 24.1 20.27

CI3 F1 20.62 20.57 21.47 21.65 20.75 0.62 20.15 0.2 0.09 1.15 20.32F2 0.27 0.2 0.06 0.29 1.16 0.64 0.43 21.29 1.54 0.21 0.35

CI4 F1 20.47 0.39 20.35 0.08 0.09 20.44 1.04 20.11 0.89 0.63 0.18F2 0.24 21.3 20.59 20.81 20.53 20.57 21.83 0.46 20.6 0 20.55

CI5 F1 22.01 22.91 20.01 20.89 20.31 0.29 20.1 20.55 20.69 0.12 20.71F2 0.05 2.08 20.1 0.13 4.31 20.38 20.26 21.07 20.14 22.07 0.26

CI6 F1 22.03 21.23 21.16 21.2 0.44 0.51 0.6 20.14 20.65 0.86 20.4F2 0.3 20.45 0.38 21.04 0.9 0.75 21.08 0.1 0.6 20.38 0.01

CI7 F1 22.09 21.74 20.94 20.27 1.97 0.67 1.49 21.25 20.62 20.49 20.33F2 2.57 0.15 1.33 1.35 2.73 20.27 22.09 22.49 21.01 22.15 0.01

CI8 F1 22.18 20.6 20.06 20.05 0.65 20.08 0.93 22.22 21.28 21 20.59F2 3.42 1.23 2.72 1.1 1.85 21.89 22.47 22.71 22.34 22.93 20.2

dofllyte

oea

ehe

’slacata

tirld

erizenav

ttlceeI2heasef

ales

ggesttraln-eltwotion

ea-s ofes

byr baret aual

ane

two correlations was significant using Spearman rank orcorrelations. Thus despite the different ‘‘starting points’’individual CI users in learning to normalize for spectrashifted input, most patients seem to manage to compleadapt after, at the most, one year of experience.

While most CI users showed little evidence of a lackadaptation to the basalward spectral shift, many of thvowel category centers differed from those of the normhearing listeners in nonsystematic ways. For example, thand /}/ categories of CI3 are shifted lower with respect to tequivalent normal categories; /i/ and /(/ of CI4 overlap al-most completely with one another; and all of CI7 and CI8categories are centralized and involve massive overThere are no obvious explanations for such discrepantegory centers. For some categories at the periphery ofspace, the centers may have differed from those of normhearing listeners if the grid did not encompass the enrelevantF1* F2 space for that listener. However, that wounot account for the ‘‘interior’’ categories~mapped entirelyout in the grid provided! that still differ from those ofnormal-hearing listeners~e.g., CI5’s /)/!.

The CI users’ vowel spaces also differed from the avage normal vowel space in terms of vowel category sThis pattern indicates differences in frequency discrimition ability. Among the CI users’ vowel spaces, we find seeral that are composed of categories that overlap very liand that appear to be in roughly the same regions in pertual space as those obtained from normal hearing listenExamples of such ‘‘normal’’ CI spaces are those from Cand CI3. In these spaces, front and back vowels occupy t‘‘normal’’ regions of perceptual space. The sizes of the cegories in these spaces are not much greater than thonormal hearing listeners. In contrast, the vowel spacesCI7 and CI8 show much larger categories and a great deoverlap between front and back vowels. The size of th


er

ly

firl-/e/

p.t-

hel-e

-.--e,p-rs.

irt-

oforofe

categories and their arrangement in perceptual space suthat these users might have great difficulty in using specinformation to discriminate among vowel sounds, or in idetifying particular vowels in running speech. The vowspaces of the four other CI users fall between theseextremes on a continuum characterized by category locaand size.

These observations are supported by the difference msures displayed in Table V, and also by the mean sizecategories~in Bark! for each CI space. These category sizare shown in Table VI, for both theF1 andF2 dimension,along with the mean sizes averaged over bothF1 andF2.The sizes for each perceptual dimension were calculatedaveraging the Bark distances encompassed by one erroin each formant dimension of each category center to gmean category size for a given subject. Below the individCI user measures are the averages across all CI users~‘‘CI’’ !

TABLE VI. The mean size of individual CI user’s categories in theF1dimension, theF2 dimension, and both dimensions, along with the mecategory sizes across all normal-hearing~NH! and CI users’ categories. Sizrefers to one standard deviation~one error bar! from the vowel categorycenter.

Patient F1 (Z) F2 (Z) F1 andF2 (Z)

CI1 0.23 0.17 0.2CI2 0.22 0.36 0.29CI3 0.37 0.37 0.37CI4 0.75 0.65 0.7CI5 0.7 1.05 0.88CI6 1.01 0.98 0.99CI7 1.6 2.1 1.85CI8 1.33 2.48 1.9

Mean CI 0.78 1.02 0.9Mean NH 0.25 0.40 0.33


ethr

r-

eng

raf tra

nb

y

ll

aa

o

rres

bNornttioono

t tiz

n, tceeetirearobine

weio

lt

telyow-m-im-

edelfre-

er,s,llI8

es.ostith

tle

tent

er-nedofarelant

tro-asap-bere

iftub-CIto

esis

thattedinec-forco-

onlyant

esal-s-tionersforthis

hisncynotin

illser-dult

and the equivalent averages for all normal-hearing listen~‘‘NH’’ !. The differences in all three measures betweenCI users and the normal-hearing listeners was on the ordea 3:1 ratio, and all were significant at thep,0.01 level (F1:t59.1, d f581.1; F2: t56.9, d f581; F1 and F2 com-bined: t510.6,d f5163.6).

The vowel spaces of CI2 and CI3, the relatively ‘‘nomal’’ spaces, have the lowest mean category sizes~alongwith CI1!, while CI7 and CI8 have the highest. Large vowcategories typically overlap with one or more neighboricategories in perceptual space, which may be predictivepotential problems in the perception of vowels in natuspeech. These findings are consistent with the results oF1 jnd and vowel identification tests: the worst performeon these tests were also the subjects whose vowel spincluded the largest, overlapping categories. The patterresults is reflected in Spearman rank order correlationstween the category size measures and theF1 jnd and vowelidentification tests. TheF1 jnd score correlated significantlwith category size inF1 (r 510.76, p,0.05), while thevowel identification score correlated significantly with athree measures (F1: r 520.91,p,0.05; F2: r 520.98,p,0.01; F1 andF2: r 520.95, p,0.05). All of the othercorrelations were only marginally significant. Thus there wa significant positive correlation between category size,measured in this study, and performance onF1 discrimina-tion, and a significant negative correlation between categsize and vowel identification.

In addition, the three category size measures also colated very strongly with the CNC word recognition scorobtained from the CI users (F1: r 520.88, p,0.05; F2:r 520.95, p,0.05; F1 and F2: r 520.91, p,0.05). Inother words, there was a significant negative correlationtween category size and a word identification test. The Cword lists measure listeners’ ability to perform open set widentification. Thus it requires listeners to access their melexicons, making the task more similar to speech percepin natural settings than the MOA or phonetic identificatitests reported here. The correlations between the CNC wrecognition scores and category size results indicate thaMOA procedure provides useful information for charactering the speech perception abilities of CI users.

III. DISCUSSION AND CONCLUSIONS

Despite the large individual differences observed amocochlear implant users in all three tests used in this studyresults revealed an orderly relationship between the pertual vowel spaces of these listeners and their performancthe F1 jnd and the vowel identification tests. In Sec. I, wdiscussed two potential factors that may limit vowel idenfication by cochlear implant users: reduced formant fquency discrimination and lack of adaptation to basalwspectral shift. Overall, the results show that the major prlem in vowel identification in these listeners appears tovolve formant frequency discrimination. Only one of theight CI users, CI1, showed a systematic shift of her vospace that was consistent with limited auditory adaptatFor each vowel, she selected regions with lowerF1 andF2than those selected by normal hearing listeners. This resu


rseof

l

oflhescesofe-

ss

ry

e-

e-Cdaln

rdhe-

ghep-on

--d--

ln.

is

consistent with the hypothesis that she has not compleadapted to the basalward spectral shift and thus selects vels with very low formants as the best exemplars, to copensate for the frequency shift imposed by the cochlearplant.

Other than CI1, none of the other CI subjects showevidence of any systematic shifts in their perceptual vowspaces, suggesting that they were able to adapt to thequency shift introduced by the cochlear implant. HowevCI users clearly differ in the size of their vowel categoriereflecting differences in their ability to discriminate smadifferences in formant frequency. For example, CI7 and Cwere the two CI users with the poorest jndF1 values, vowelidentification scores, and CNC word recognition scorTheir vowel spaces showed substantial overlap among mvowel regions. In contrast, CI users such as CI1 and CI2 wthe best~smallest! jndF1’s, high vowel identification, andhigh CNC word recognition scores, tended to have litoverlap among vowel regions.

Our results on perceptual vowel spaces are consiswith those reported recently by Hawks and Fourakis~1998!.Their cochlear implant subjects also showed widely divgent amounts of overlap among vowel categories obtaiusing an identification task with synthetic stimuli. In termsadaptation to basalward spectral shift, the present resultsinteresting because they suggest that most cochlear impusers are able to adapt to the frequency shifts typically induced by their devices. However, although CI1’s space wabnormal and thus not the best example of incomplete adtation, her results do remind us that not all listeners mayable to adapt in a similar way. In addition, perhaps molisteners would find it difficult to adapt if the spectral shwas greater than that estimated for the cochlear implant sjects in this study. Moreover, the larger categories of theusers could themselves reflect incomplete adaptationbasalward spectral shift, as the spectral smear hypothsuggests.

In summary, the data reported in this paper suggestvowel perception by cochlear implant users may be limiby the listener’s formant frequency discrimination skills,combination with his/her ability to adapt to basalward sptral shift. The present findings also have implicationsunderstanding the perception of consonant sounds bychlear implant users, because formant frequencies notprovide cues for vowel perception but also provide importinformation for consonant recognition. In future studies wintend to explore longitudinal changes in adaptation to baward shift and in frequency discrimination, in order to invetigate the nature of the improvement in speech percepobserved in most postlingually deaf cochlear implant usafter they receive their devices. An additional avenueresearch may involve applying the same methods used instudy to prelingually deafened pediatric CI users. In tcase, however, we would not expect to see major frequeshifts in the vowel spaces of children because they wouldhave prior expectations concerning the location of vowelsthe vowel space. Instead, the frequency discrimination skof prelingually deafened pediatric CI users may be, on avage, even worse than those of postlingually deafened a


hebeichthotioalini

g1,eee

eia

ce

rinith

udli

ns

.,eol.

n

nt,

o

ss.

-

A.ofol.

foc.

o-

S.in

lear

AK

.,

.dt,C.

lant

-

.

CI users. Finally, it may be interesting to determine whetthe listeners who show incomplete adaptation have lesshavioral plasticity than the other listeners, or whether thlack of adaptation is due to a shallower insertion, whwould result in a greater spectral shift to be overcome bylistener. Taken together, these studies should help us accfor the enormous individual differences in speech percepby CI users, which should provide the theoretical rationfor improved devices and intervention strategies in this clcal population.

ACKNOWLEDGMENTS

This work was supported by NIH-NIDCD TraininGrant No. DC00012, NIH-NIDCD Grant No. R01-DC0011and NIH-NIDCD Grant No. R01-DC03937. We would likto thank Dr. Richard Miyamoto for providing the electrodinsertion information for the cochlear implant users. Wwould also like to thank two anonymous reviewers for thhelpful comments, as well as Chris Quillet for his technicassistance.

1Some stimuli were slightly louder than others, but these small differenin loudness should be very hard to perceive for most CI users.

2The female stimulus set results were not included in the normal-healisteners’ averaged space because the CI users were only presented wmale stimulus set.

3However, work is in progress in our laboratory on a vowel production stof Central Indiana English, using as talkers the same normal-hearingteners that participated in this study.

Blamey, P. J., Dooely, G. J., Parisi, E. S., and Clark, G. M.~1996!. ‘‘Pitchcomparisons of acoustically and electrically evoked auditory sensatioHear. Res.99, 139–150.

Dorman, M. F.~1993!. ‘‘Speech perception by adults,’’ inCochlear Im-plants, edited by R. S. Tyler~Singular Publishing, San Diego!, pp. 145–190.

Durlach, N. I., and Braida, L. D.~1969!. ‘‘Intensity perception. I. Prelimi-nary theory of intensity resolution,’’ J. Acoust. Soc. Am.46, 372–383.

Eddington, D. K., Dobelle, W. H., Brackmann, D. E., Mladejovsky, M. Gand Parking, J. L.~1978!. ‘‘Auditory prostheses research with multiplchannel intracochlear stimulation in man,’’ Ann. Otol. Rhinol. Laryng53, 5–39.

Flanagan, J.~1957!. ‘‘Estimates of the maximal precision necessary in quatizing certain ‘dimensions’ of vowel sounds,’’ J. Acoust. Soc. Am.29,533–534.

Fu, Q.-J., and Shannon, R. V.~1999a!. ‘‘Effects of electrode location andspacing on phoneme recognition with the Nucleus-22 Cochlear ImplaEar Hear.20, 321–331.

Fu, Q.-J., and Shannon, R. V.~1999b!. ‘‘Effects of electrode configurationand frequency allocation on vowel recognition with the Nucleus-22 Cchlear Implant,’’ Ear Hear.20, 332–344.

Fu, Q.-J., Shannon, R. V., and Galvin, J.~submitted!. ‘‘Performance overtime of Nucleus-22 cochlear implant listeners wearing a speech procewith a shifted frequency-to-electrode assignment,’’ J. Acoust. Soc. Am

Gimson, A. C. ~1962!. An Introduction to the Pronunciation of English~Edward Arnold, London!.


re-ir

euntne-

rl

s

gthe

ys-

,’’

-

’’

-

or

Greenwood, D. D.~1961!. ‘‘Critical bandwidth and the frequency coordinates of the basilar membrane,’’ J. Acoust. Soc. Am.33, 1344–1356.

Gstoettner, W., Hamzavi, J., and Czerny, C.~1997!. ‘‘Rehabilitation of pa-tients with hearing loss by cochlear implants,’’ Radiologe37, 991–994.

Hawks, J. W., and Fourakis, M. S.~1998!. ‘‘Perception of synthetic vowelsby cochlear implant recipients,’’ J. Acoust. Soc. Am.104, 1854–1855.

Hillenbrand, J., Getty, L. A., Clark, M. J., and Wheeler, K.~1995!. ‘‘Acous-tic characteristics of American English vowels,’’ J. Acoust. Soc. Am.97,3099–3111.

Hinjosa, R., and Marion, M.~1983!. ‘‘Histopathology of profound senso-rineural deafness,’’ Ann. N.Y. Acad. Sci.405, 459–484.

Johnson, K., Flemming, E., and Wright, R.~1993!. ‘‘The hyperspace effect:Phonetic targets are hyperarticulated,’’ Language69, 505–528.

Ketten, D. R., Skinner, M. W., Wang, G., Vannier, M. W., and Gates, G.~1998!. ‘‘ In vivo measures of cochlear length and insertion depthnucleus cochlear implant electrode arrays,’’ Ann. Otol. Rhinol. Laryng175, 1–16.

Kewley-Port, D., and Watson, C. S.~1994!. ‘‘Formant-frequency discrimi-nation for isolated English vowels,’’ J. Acoust. Soc. Am.95, 485–496.

Kewley-Port, D., and Zheng, Y.~1998!. ‘‘Auditory models of formant fre-quency discrimination for isolated vowels,’’ J. Acoust. Soc. Am.103,1654–1666.

Klatt, D. H., and Klatt, L. C.~1990!. ‘‘Analysis, synthesis, and perception ovoice quality variations among female and male talkers,’’ J. Acoust. SAm. 87, 820–857.

Labov, W. ~1991!. ‘‘The three dialects of English,’’ inNew Ways of Ana-lyzing Sound Change, edited by P. Eckert~Academic, New York!, pp.1–44.

Levitt, H. ~1972!. ‘‘Decision theory, signal-detection theory, and psychphysics,’’ in Human Communication: A Unified View, edited by E. E.David and P. B. Denes~McGraw-Hill, New York!, pp. 114–174.

Nelson, D. A., Van Tasell, D. J., Schroder, A. C., Soli, S., and Levine,~1995!. ‘‘Electrode ranking of ‘place pitch’ and speech recognitionelectrical hearing,’’ J. Acoust. Soc. Am.98, 1987–1999.

Peterson, G. E., and Barney, H. L.~1952!. ‘‘Control methods used in thestudy of vowels,’’ J. Acoust. Soc. Am.24, 175–184.

Pilch, H. ~1994!. Manual of English Phonetics~W. Fink, Munchen!.Rosen, S., Faulkner, A., and Wilkinson, L.~1999!. ‘‘Adaptation by normal

listeners to upward spectral shifts of speech: Implications for cochimplants,’’ J. Acoust. Soc. Am.106, 3629–3636.

Skinner, M. W., Holden, L. K., and Holden, T. A.~1995!. ‘‘Effect of fre-quency boundary assignment on speech recognition with the SPEspeech-coding strategy,’’ Ann. Otol. Rhinol. Laryngol.166, 307–311.

Skinner, M. W., Ketten, D. R., Vannier, M. W., Gates, G. A., Yoffie, R. Land Kalender, W. A.~1994a!. ‘‘Determination of the position of nucleuscochlear implant electrodes in the inner ear,’’ Am. J. Otolaryngol.15,644–651.

Skinner, M. W., Clark, G. M., Whitford, L. A., Seligman, P. M., Staller, SJ., Shipp, D. B., Shallop, J. K., Everingham, C., Menapace, C. M., ArnP. L., Antogenelli, T., Brimacombe, J. A., Pijl, S., Daniels, P., George,R., McDermott, H., and Beiter, A. L.~1994b!. ‘‘Evaluation of a newspectral peak coding strategy for the Nucleus 22 channel cochlear impsystem,’’ Am. J. Otolaryngol.15 ~Suppl. 2!, 15–27.

Tyler, R. S., Preece, J. P., and Lowder, M. W.~1987!. ‘‘The Iowa audiovi-sual speech perception laser video disk,’’Laser Videodisk and LaboratoryReport, University of Iowa at Iowa City, Department of OtolaryngologyHead and Neck Surgery.

Wilson, B. S., Finley, C. C., Lawson, D. T., Wolford, R. D., Eddington, DK., and Rabinowitz, W. M.~1991!. ‘‘Better speech recognition with co-chlear implants,’’ Nature~London! 352, 236–238.

Wolfram, W., and Schilling-Estes, N.~1998!. American English~Blackwell,Malden, MA!.


perceptual “vowel spaces” of cochlear implant users: implications for the study of auditory...

Documents