Individual Variability in Recognition of Frequency-Lowered ...

Individual Variability in Recognition ofFrequency-Lowered Speech

Joshua M. Alexander, Ph.D.1

ABSTRACT

Frequency lowering in hearing aids is not a new concept, butmodern advances in technology have allowed it to be performed moreefficiently and on select portions of the spectrum. Nonlinear frequencycompression reduces the frequency spacing in a band of high-frequencyenergy so that more information is carried in the audible bandwidth.Frequency transposition and translation techniques lower only the partof the high-frequency spectrum that likely contains important speechinformation. These advances may help overcome the limited bandwidthin conventional hearing aids, which restrict access to high-frequencyinformation even for those with mild to moderate hearing loss. This isespecially important for young children learning speech and language. Aframework is advanced in which factors that influence individualdifferences in speech recognition can be divided into extrinsic factorsthat affect the representation of the frequency-lowered speech at theauditory periphery, including the specific technique and the settingschosen for it, and intrinsic factors that contribute to an individual’sability to learn and benefit from this signal. Finally, the importance ofelectroacoustically verifying the output to avoid too little or too muchlowering and the importance of validating effectiveness of outcomes inindividual users of the technology are emphasized.

KEYWORDS: Hearing aids, frequency lowering, frequency

compression, frequency transposition

Learning Outcomes: As a result of this activity, the participant will describe how different frequency-lowering

technologies alter the speech signal, how characteristics of the hearing aid wearer might influence the ability

to benefit from this signal, and what to watch for when verifying output.

1Department of Speech, Language, and Hearing Sciences,Purdue University, West Lafayette, Indiana.

Address for correspondence: Joshua M. Alexander,Ph.D., Department of Speech, Language, and HearingSciences, Purdue University, West Lafayette, IN 47907(e-mail: [email protected]).

Individual Variability in Aided Outcomes; Guest Editor,Jason A. Galster, Ph.D.

Semin Hear 2013;34:86–109. Copyright # 2013 byThieme Medical Publishers, Inc., 333 Seventh Avenue,New York, NY 10001, USA. Tel: +1(212) 584-4662.DOI: http://dx.doi.org/10.1055/s-0033-1341346.ISSN 0734-0451.

86

Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Individuals with high-frequency sensori-neural hearing loss (SNHL) are denied accessto potentially important speech information. Formilder losses, this can occur if the miniatureelectronics in hearing aids are unable to providesufficient high-frequency amplification or can-not do so without audible whistling and over-tones caused by feedback. For more severe losses,the inner hairs cells that code these frequenciesmay simply be “dead,” possibly rendering ampli-fication in this region less useful.1–5 As theseverity of loss increases to include more low-frequency content, the amount of “lost” speechinformation increases along with the challengesfor the listener. Most of the sounds we will beconcerned about in this discussion involve thefricatives, affricates, and the initial segments ofthe stop consonants, which are primarily char-acterized by aperiodic mid- to high-frequencyspectral information. Fig. 1 shows a spectrogramof the sentence “children like strawberries” withthe previously mentioned sound classes denotedby arrows. As can be seen, the acoustic energy inthese sounds can be quite high in frequency, withpeak energy sometimes around 9,000 Hz, espe-cially for women and children talkers.6–8 Incontrast, the vowels and other consonants areprimarily characterized by bands of energy, for-mants, at relatively lower frequencies. Unlike thediffuse spread of energy that is characteristic ofthe fricatives, affricates, and initial stop conso-nant segments, changing the frequency relation-ship of the formants can have serious

consequences for the identity of the correspond-ing speech sound.

Speech is linguistically and acousticallyredundant and, with varying degrees of success,listeners can identify high-frequency phonemesusing only the transitions from the lower-frequency formants of the coarticulated pho-nemes that precede and follow them.9,10 De-spite this, there is evidence that speechperception improves for both adults and chil-dren when an effort is made to preservethe high-frequency noise energy associatedwith frication, especially when identifying/s/.7,11–21 These findings and the fact thathearing aids have limited usable bandwidthhave been used to explain the continued diffi-culty experienced by young children usinghearing aids when perceiving and producingthese sounds compared with vowels and otherconsonant sound classes.22–25

The gravity of this problem is compoundedby the regularity with which /s/ and its voicedcognate, /z/, occur in the English language(� 8% of all spoken consonants) and by theirlinguistic importance.26 Rudmin identifies over20 linguistic uses for /s/ and /z/, includingplurality, third-person present tense, past versuspresent tense, to show possession, possessivepronouns, contractions, and so on.26 Develop-mentally, inconsistent exposure to these pho-nemes for a child with SNHL may have long-term consequences for morphosyntactic devel-opment.24,27 Findings like these have been the

Figure 1 Spectrogram of the sentence “children like strawberries” before processing as spoken by a femaletalker. Arrows denote the fricative, affricate, and stop sound classes that contain significant high-frequencyenergy and are therefore the subject of frequency-lowering techniques.

RECOGNITION OF FREQUENCY-LOWERED SPEECH/ALEXANDER 87

Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

inspiration for a variety of frequency-lowering(FL) techniques (i.e., methods of moving high-frequency speech information into lower-fre-quency regions) in commercially available hear-ing aids. In contrast to a few years ago when FLseemed like just a signal processing novelty, nowmore than half of the world’s major hearing aidmanufacturers include FL as an optional feature.

Table 1 provides a summary of peer-re-viewed research on modern, commerciallyavailable FL technologies along with a fewstudies under review in which the author haspersonal involvement. A supplementary de-scription of early FL techniques and a summaryof related research findings can be found inSimpson.28 Specific studies in Table 1 arediscussed where appropriate and specific tech-nologies are discussed in the next section.

At the outset, it is important to make notethat the data indicate that we, as a field ofclinicians and researchers, do not yet knowenough to predict who will and who will notbenefit from FL technology. First, becausemodern FL techniques have only been aroundfor a short period compared with other hearingaid processing strategies, there just simply is nota lot of data by independent researchers as canbe inferred from Table 1. Furthermore, we havethe same difficulties as we do when trying tounderstand individual variability with conven-tional hearing aids in addition to nuancesassociated with FL technology. Specifically,we now have the added difficulty of under-standing (1) how the different technologiesalter the speech signal that is transduced bythe impaired auditory periphery and (2) howthis interacts with individual characteristics thatinfluence the ability to learn it.

EXTRINSIC FACTORS RELATED TOTHE SIGNAL PROCESSING ANDIMPAIRMENTFig. 2 provides a framework for understandingthe different factors that likely influence indi-vidual variability in recognition of frequency-lowered speech. The first set of factors to bediscussed are extrinsic to the listener and relatesto the representation of the signal at the audi-tory periphery. The specific speech sound andits environmental context serve as input to the

digital signal processor that alters the acousticrepresentation in a way that is specific to thesignal and to the FL technology and its settings.How this newly coded input is then “seen” bythe central auditory system, and beyond, de-pends on the integrity of peripheral processing(e.g., threshold elevation and broadened audi-tory filters associated with varying degrees ofouter and inner hair cell impairment). It isargued here that these factors contribute themost to differences in individual outcomes.Because there is nothing one can do to controlthe input signal or the peripheral processing, itis incumbent upon the hearing aid dispenser tochoose the appropriate technology and settingsthat match the speech perception deficit withthe impairment. As will be discussed, with a fewexceptions, there are no firm guidelines on howthis should be done during fitting, which putsthe onus on the dispenser to use appropriatemeasures to validate individual outcomes.

Signal Processing Details

To make informed choices about when, what,and how to implement FL in a hearing aidfitting, it is critical that the dispenser under-stands the technology and how the handles inthe fitting software manipulate the signal.Unfortunately, these details are sometimeshard to come by. Presented below is a sum-mary of information gathered from detailedacoustic analyses by the author for all but themost recent FL techniques. Industry interestin FL had a slow start, but has recently surged.Currently, five FL techniques are imple-mented in commercially available hearingaids:

� Linear frequency compression by AVR So-novation of Israel (introduced 1991)

� Linear frequency transposition (LFT) byWidex of Lynge, Denmark (introduced2006)

� Nonlinear frequency compression (NFC) byPhonak of Stafa, Switzerland (introduced2008). Also used by Unitron of Kitchener,Ontario, Canada (starting 2012)

� Spectral envelope warping by Starkey Hear-ing Technologies of Minnesota (EdenPrarie, MN), USA (introduced 2011)

88 SEMINARS IN HEARING/VOLUME 34, NUMBER 2 2013

Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Table

1Summary

ofStudiesUsingModern

Frequency-LoweringTechnology

Reference/T

echnique

ParticipantDetails

Methods/O

utcomeMeasures

Results

Kuketal(2009)29/LFT

8adultswithsevere

toprofound

SNHL>

2kHz

�Daily

trainingforfirstmowithLFT;nocontrolfortrainingwithout

LFT

�Consonantrecog(nonsenseCVCVCs)in

quiet(50-&

68-dBSPL)&

babble

noise,assessedafterinitialfit,1mopostfitþ

training,2mo

postfit

�At50-dBSPL,fricativesim

proved5–10%

initially

&an

additional10%

after2mo

�At68-dBSPL,fricativesim

proved10%

after2mo;an

initialdecreasein

stopswasbackto

baselineafter2mo

�Nodifferenceforspeechin

noise,exceptfricatives(>

15%

improvement)

Auriemmoetal(2009)41/LFT

10childrenwithsevere

toprofound

SNHL>

3kHz

�Phonemerecog(nonsenseCVCVs)in

quiet&

fricativeproduction

�Conditions:participant’sownaid,new

devicewithoutLFT(3

wk),&

withLFT(6

wk)

�Weekly

auditory

trainingwithoutLFT&

withLFT

�After6wk,consonantrecogwithLFTat30-dBHLwas

>20%

betterthanwithoutLFT;nosignificantdifference

at50-dBHL

�Worstperform

ers

withoutLFThadgreatest

improvementwhenactivated

�Novowelrecogdifferences

�Productionaccuracyof/s/&

/z/im

provedby>

10%

whenactivated

Smithetal(2009)96/LFT

6childrenwithsevere

toprofound

SNHL>

1kHz

�Audio

&audiovisualrecogoflivevoicemonosyllables(CVCs)

�Consonantproductionin

words&

sentences

�Parental&

teacherreports

�Norandomizationorcontrolformaturationeffects

�Im

provements

inCVCrecogat3&

6moforaudio

only

(notaudiovisual)&

productionat6mo

Alexanderetal(2008)42/LFT&

NFC

24adultswithmild

tomoderate

SNHL

&24norm

al-hearingcontrols

�Recogoffricatives&

affricates(VCs)in

noiserecordedthroughaids

withLFT&

NFC

�Againstmanufacturers’candidacy

�Conditions:noLFT/noNFC(conventional),LFT,NFC,wideband(by

addinghigh-passfilteredspeechto

therecordings)

�Wideban

dim

provedoverconventional

�NFCwithinputbandwidth

�9kHzim

provedover

conventional

�LFTsignificantlydegradedrecog,especially

/s/&

/z/,for

both

groups

Sim

psonetal(2005)34/NFC

17adults(experiencedHAusers)with

moderate

toprofoundSNHL

�RecogofCVCmonosyllabic

words

�Conditions:conventionalaid,experimentalNFC(4–6wk)

�Noevidencethataudibility

fortheloweredspeechwasmeasured

�8hadim

provementin

high-freq.word

recogwithNFC,8

showednodifference,&

1perform

edworse

Sim

psonetal(2006)35/NFC

7adultswithprecipitousSNHL

�Recogin

quiet(open-setmonosyllabic

words,closed-setVCVs)&

noise(open-setsentences)

�Subjectivemeasure:APHAB

�2conditions:conventionalaid,experimentalNFC(4–6wk)

�Startfreq.rangedfrom

1.0

to1.6

kHz

�Noevidencethataudibility

fortheloweredspeechwasmeasured

�Nosignificantdifferencein

monosyllabic

word

&

consonantrecogin

quiet

�Only

1listenershowedim

provementin

sentencerecog

innoisewithNFC

�APHAB:higherglobalscoresfor4listeners

withconven-

tionalaid

(Con

tinue

d)


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Table

1(Continued)

Reference/T

echnique

ParticipantDetails

Methods/O

utcomeMeasures

Results

Glista

etal(2009)40/NFC

13adults&

11childrenwithsloping

high-freq.SNHLrangingfrom

moder-

ately

severe

toprofound

�Aidedspeechdetection,speechrecogforconsonants,plurals,&

vowels

�3phases:acclim

atizationto

conventionalprocessing,treatm

ent

withprototypeNFC,treatm

entwithdrawal(conventional

processing)

�5adultshadsignificantim

provement,mostlyforplurals

�7childrenhadsignificantim

provement,mostlyfor

plurals

�Benefitforplurals

dependedonagegroup&

high-freq.loss

�Children>

adultspreferredNFC

�1adulthadworsevowelperception

Bohnertetal(2010)47/NFC

11adultswithsevere

toprofound

SNHL

�NFCcomparedwithlisteners’owndeviceafter2&

4mo

�Startfrequencies�

2kHzfor7listeners

�SRT:nonsensesentencesin

noise

�Subjectivequestionnaires

�Comparisonbetw

eendevicesnotvalid;also,noblindingto

the

treatm

ent

�ReportsoffricativesassoundingunnaturalwithNFC,

despitebettersound-qualityratings

Wolfeetal(2010)78/NFC

15childrenwithmoderate

tomoder-

ately

severe

SNHL

�NFCenabledversusNFCdisabled(6-w

kacclim

atizationperiod

each)

�Aidedspeechdetection,pluralrecogin

quiet,sentencerecogin

multitalkerbabble,high-freq.consonantdiscriminationin

quiet

�PluralrecogsignificantlybetterwithNFC(84%

versus

99%)

�Im

provedconsonantdiscrimination(/s/&

/d/)withNFC

�Nosignificantdifferenceforsentencerecogin

noise

Wolfeetal(2011)25/NFC

SeeWolfeetal(2010)78

�6-m

ofollow-uponWolfeetal2010usingthesamemeasures

�Nocontrolorwithdrawalcondition,thus,cannotrule

outmatura-

tion/learningeffects

�Im

provedconsonantdiscriminationwithNFCafter6mo

versus6wk

�Im

provedsentencerecogin

noiseonly

after6mo

McCreery

etal(2012)45/NFC

20adultswithnorm

alhearing

�Processingfor3audiogramswithvaryingdegreesofhigh-freq.

hearingloss

�Recogoffricatives&

affricates(nonsenseCVCs)

�3conditions(usingthePurduehearingaid

sim

ulator):conventional

processing,manufacturer’sdefaultNFCsettings,NFCsettingswith

optimizedbandwidth

�Speechrecogim

provedacrossconditionsasestimated

audibility

&bandwidth

increased

�Optimizedfittingmethodresultedin

significantlyhigher

speechrecogcomparedwithotherconditions

Glista

etal(2012)51/NFC

6childrenwithatleastmoderately

severe

high-freq.SNHL

�Aidedspeechdetection,pluralrecog,/s/-/R/discrimination(CV

pairs),consonantrecog

�3phases:conventionalprocessing(baseline),NFC(treatm

ent�

16

wk),conventionalprocessing(w

ithdrawal)

�Single-subjectdesign;cannotdiscountmaturationeffects

�Significantbenefit(upto

20%)from

NFCfollowingaccli-

matizationfor5listeners

onatleast1test,especially

fricatives;theotherlistenerwasnearceilingwith&

withoutNFCformosttests

�Im

provedpluralrecog(3

listeners)atvaryingtimesdur-

ingtreatm

entphase

�Im

proved/s/-/R/discrimination(3

listeners)overtime

�Acclim

atizationtrendswere

variable


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Table

1(Continued)

Reference/T

echnique

ParticipantDetails

Methods/O

utcomeMeasures

Results

Ellis&

Munro

(2013)95/NFC

15adultswithnorm

alhearing

�Sentencerecogin

noise

�Cognitivemeasures:workingmemory

span(readingspantest),

executivecontrolfunction(trailmarkingtest)

�3conditions:noNFC&

2NFCwitha1.6

kHzstartfreq.(2:1

or3:1

compressionration)

�NFCsignificantlydecreasedspeech

recog

�Significantcorrelationbetw

eenrecogofunprocessed

speech&

3of4cognitivemeasures

�Nosignificantrelationship

betw

eenrecogofNFCspeech

&cognitivemeasures

Kopunetal(2012)97/NFC

12children&

24adultswithmild

to

severe

SNHL

�Recogofmonosyllabic

wordswith&

withoutNFCusingPurdue

hearingaid

sim

ulator,assessedbefore

&aftershort-term

audiovisu-

alexposure

withNFC

�NFCsignificantlyim

provedword

recog

�Nopreferentialeffectforexposure

orforagegroup

�Only

2adultsperform

edworsewithNFC

Brennan

etal(2012)48/NFC

19children&

24adultswithmild

to

severe

SNHL

�Sound-qualityratingsforspeech(sentenceswith3fricatives)&

for

avariety

ofmusic

genres

�3conditionsprocessedwithPurduehearingaid

sim

ulator:restricted

bandwidth

at5kHz,

extendedbandwidth

at10kHz,

&NFC

�Im

provedsoundqualitywithextendedbandwidth/NFC

comparedwithrestrictedbandwidth

forspeech;nodif-

ferencesformusic

�Listeners’preferenceforextendedbandwidth/NFCposi-

tively

correlatedwithamplificationexperience,butno

relationship

withaudiogram

Alexander(2012)46/NFC

28adultswithmild

tomoderately

severe

SNHL

�Recogofconsonants

&vowels

from

nonsensesyllablesin

noise

�Purduehearingaid

sim

ulatorusedto

create

acontrol&

6NFC

conditionswithfixedoutputbandwidth

(2groupsat3.3

&5.0

kHz)

butvaryingstartfreq.&

compressionratio

�Im

provementin

fricativerecogformostNFCsettings

relativeto

control

�1.6

kHzstartfreq.decreasedvowel&

nonfricativecon-

sonantrecog(increasingstartfreq.wasmore

effective

atrestoringrecogthandecreasingcompressionratio)

Abbreviations:APHAB,AbbreviatedProfile

ofHearingAid

Benefit;C,consonant;Freq.,frequency;HA,hearingaid;HL,hearinglevel;LFT,linearfrequencytransposition;NFC,

nonlinearfrequencycompression;recog.,recognition;SNHL,sensorineuralhearingloss;SPL,soundpressure

level;SRT,speechrecognitionthreshold;V,vowel.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

� Frequency compression by Siemens ofErlangen, Germany (introduced 2012)

Linear Frequency Compression

One of the earliest commercially available FLtechniques was linear frequency compressionintroduced by AVR Sonovation in analog de-vices in 1991 and later in digital devices in 2004.A key element of this technique is its switchingbehavior. A spectral balance detector is used atthe front end of the processing to determinewhen FL should occur. Specifically, if theenergy above 2,500 Hz is greater than thatbelow 2,500 Hz, then FL occurs, otherwiseamplification is provided as usual. Hence, FLis dynamic because of its all or none behavior.At least for speech in quiet, this is a goodmethod for identifying the speech sounds char-acterized by high-frequency aperiodic energy.FL is implemented quite simply using a “slowplay” effect; in fact, the early analog version useda magnetic tape that was played at a slow rate ofspeed. Digitally, this is done by using twodifferent analog-to-digital converters (ADCs)at the frontend. The first, ADC1, has a sam-pling rate equal to the sampling rate of thedigital-to-analog converter (DAC). The other,ADC2, has a programmable sampling rate thatis an integer multiple (2, 3, 4, or 5) of the DACsampling rate, called the dynamic frequencycompression coefficient. The former, ADC1,is used when no lowering occurs (i.e., when thesignal is low-frequency dominated, as withvowels) and the other is used for lowering(i.e., when the signal is high-frequency domi-nated, as with fricatives, etc.). Because the

sampling rate of the DAC (output) is slowerthan the ADC2 (input), the entire frequencyrange is shifted proportionally lower by a factorequal to the dynamic frequency compressioncoefficient (Figs. 3 and 4).

Linear Frequency Transposition

In 2006, Widex introduced LFT to the marketas the “Audibility Extender.” Whereas AVRSonovation hearing aids were primarily nicheproducts, this was the first time that the conceptof FL went mainstream. LFT was simply anoptional add-on feature of an already fullydeveloped line of products. When the LFTfeature is activated, the algorithm continuallysearches for the most intense spectral peak in alimited frequency range known as the “source

Figure 2 A framework for understanding factors that likely influence individual differences in speechrecognition. Extrinsic factors affect the representation of the frequency-lowered speech at the auditoryperiphery, including the specific technique and the settings chosen for it, and intrinsic factors contribute to anindividual’s ability to learn and benefit from this signal.

Figure 3 Relationship between input and outputfrequencies for linear frequency compression. Whenthe input is dominated by energy below 2,500 Hz, nolowering occurs. When it is dominated by energyabove 2,500 Hz, the entire frequency range is com-pressed by a factor determined by the dynamicfrequency compression coefficient. Abbreviation:DFC, dynamic frequency compression.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

region.” The source region is determined by aprogrammable “start frequency” that includesthe one-third-octave band frequencies from630 to 6,000 Hz. The source region begins ahalf octave below the start frequency and ex-tends one octave above it or to the limit of theinput bandwidth of the microphone and/orADC. The frequency region to which the inputis transposed is called the “target region” and isone octave below the source region (Table 2).An octave-wide band (relative to the targetdestination) is filtered around the dominantspectral peak in the source region and is then

resynthesized one octave down (a factor of two),thusmixing with any low-frequency energy thatmight be present.29 An optional expandedmode exists for start frequencies � 2,500 Hzwhere the source region begins a half octaveabove the start frequency and extends for anadditional octave (Table 3). As with the basicmode, an octave-wide band (relative to thetarget destination) is filtered around the domi-nant spectral peak in the source region, but isthen resynthesized down by a factor of threeinstead of two. Hence, below the nominal startfrequency there is a potential mixing of energyfrom the original input signal, the transposedsignal from the basic mode, and the transposedsignal from the expanded mode (Fig. 5A).Table 2 The Approximate Source and Target

Regions of the Basic Transposed Signals forDifferent LFT Start Frequencies (in Hz)

Start Freq. Source Region Target

Region

630 445–1260 223–630

800 566–1600 283–800

1,000 707–2000 354–1000

1,250 884–2500 442–1250

1,600 1131–3200 566–1600

2,000 1414–4000 707–2000

2,500 1768–5000 884–2500

3,200 2263–6400 1131–3200

4,000 2828–8000 1414–4000

6,000 4242–(max) 2121–(max/2)

Abbreviations: Freq, frequency; LFT, linear frequency;Note: Actual values will depend on the audibility of thelowered signal. The value for “max” corresponds to themaximum frequency represented by the analog-to-digitalconverter.

Figure 4 A spectrogram of the sentence “children like strawberries” from Fig. 1 after processing with linearfrequency compression and a dynamic frequency compression coefficient of three. Energy from 0 to 8 kHz iscompressed down to 0 to 2.67 kHz when the spectral balance of the input segment is high-frequencydominated. Boxes highlight the visually identifiable energy altered by the processing.

Table 3 The Approximate Source and TargetRegions of the Expanded Transposed Signalsfor Different LFT Start Frequencies (in Hz)

Start Freq. Source Region Target

Region

630 891–1782 297–594

800 1131–2263 377–754

1,000 1414–2828 471–943

1,250 1768–3536 589–1179

1,600 2263–4525 754–1508

2,000 2828–5657 943–1886

2,500 3536–(max) 1179–(max/3)

Abbreviations: Freq, frequency; LFT, linear frequency;Note: Actual values will depend on the audibility of thelowered signal. The value for “max” corresponds to themaximum frequency represented by the analog-to-digitalconverter.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Unlike the previous technique, when activated,FL is continuous, although its behavior isdynamic because what is lowered and wheredepends on the spectral content of the inputsignal (Fig. 6).

Nonlinear Frequency Compression

In 2008, Phonak introduced their first hearingaids with NFC, known as “SoundRecover” tothe market. In some respects, frequency com-pression as implemented by Phonak is oppositeof frequency compression as implemented byAVR Sonovation. Whereas the latter is selec-tive in time and is implemented linearly in thetemporal domain (i.e., by exploiting samplingrate) across the entire frequency range, theformer is frequency selective and is imple-mented nonlinearly in the spectral domainacross time. That is, when activated, NFC isalways operating, but it only does so over alimited analysis band that is determined by aprogrammable start frequency that ranges from1,500 to 6,000 Hz. Frequencies below the startfrequency do not undergo FL (Fig. 7). This is akey difference between how LFT and NFC arecontrolled by the handles within theWidex andPhonak programming software. With Widex,all the FL occurs below the start frequencywhereas with Phonak, all the FL occurs abovethe start frequency.

With the first generation product (NaıdaUP), only frequencies up to 6.3 kHz weresubject to frequency compression. With thesecond generation of products, a band about4.5 to 4.8 kHz wide beginning just below thestart frequency was subject to frequency com-pression, with an upper limit of 10 kHz. Withthe latest generation of products introduced in2011, all frequencies beginning with the startfrequency and continuing through 10 kHz un-dergo frequency compression. The relationshipbetween input and output frequencies is deter-mined by the compression ratio, which can varyfrom 1.5:1 to 4.0:1. As confirmed by calcula-tions performed by the author, because FLoccurs on a log scale, the compression ratiocorresponds to the psychophysical reduction inspectral resolution in terms of auditory filters (e.g., 2.0:1 means that information that wouldnormally span two auditory filters in the

Figure 5 Relationship between input and outputfrequencies for three different start frequencies(different panels) for linear frequency transposition.The dotted line in each panel corresponds to ahypothetical loss in which the maximum audibleoutput frequency with amplification is 1,800 Hz. Theuntransposed signal is represented by circles, thetransposed signal by squares, and the expandedtransposed signal by triangles. (A) The boxes indicatefrequencies of overlap between the untransposed,transposed, and expanded transposed signals whenthe start frequency is 1,600 Hz. (B) The box indicatesan island of intermediate input frequencies that areinaudible when the start frequency is 3,200 Hz. (C)The arrows indicate an ideal scenario when the startfrequency is 2,500 Hz in which the input frequencycorresponding to where audibility for the untrans-posed signal ends is close to the start frequency fortransposition and where audibility for this signal endsis close to the start frequency for expandedtransposition.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

unimpaired ear before processing will only spanone auditory filter after processing). To preventdispensers from being overwhelmed, options inthe programming software are limited to 12 to15 preset combinations of start frequency andcompression ratio. The exact combinationsdepend on the hearing loss entered into thesoftware. Shown in Fig. 8 is a family of fre-quency input-output curves that were measuredempirically using a hearing aid with the secondgeneration algorithm.

Siemens has recently implemented a formof FL simply known as “frequency compres-sion.”30 Details about its signal processing areunknown at this time. Dispenser control overFL is different from SoundRecover in that twohandles, fmin and fmax, determine the start and

end frequencies of the target region, respective-ly. The compression ratio is then determined bythe settings for the two handles, which havelower and upper limits of 1.5 and 8.0 kHz,respectively.

Spectral Envelope Warping

Starkey is also one of the latest companies toimplement FL, known as “Spectral iQ,” in itshearing aids. The algorithm is described as“spectral envelope warping.” The term spectralfeature detection is used in the algorithm de-scription because a classifier looks for spectralfeatures in the high-frequency spectrum thatare characteristic of speech. The term transla-tion has been used to describe the algorithm

Figure 6 A spectrogram of the carrier sentence “children like strawberries” from Fig. 1 and the stimulus“eeSH” after processing with linear frequency transposition in which peak energy from � 4,242 to 7,000 Hzis transposed down to 2,121 to 3,500 Hz. Boxes highlight the visually identifiable energy altered by theprocessing.

Figure 7 A spectrogram of the sentence “children like strawberries” from Fig. 1 after processing withnonlinear frequency compression in which the entire band from 1.5 to 6.0 kHz is compressed to 1.5 to3.5 kHz. One consequence of a low start frequency that can be seen from the figure is that formanttransitions become flattened compared with the input in Fig. 1.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

behavior as these features are added to the low-frequency signal (Fig. 9) in a way that preservestheir natural harmonic structure (Fig. 10). An-other key feature is that, unlike other FLtechniques, this technique does not roll offthe high frequencies beyond the two upperchannels (> 5.7 kHz). This is done to minimizethe risk of the dispenser unintentionally limit-ing audible bandwidth by choosing too aggres-sive a FL setting, the importance of which willbe discussed later. Dispenser control of SpectraliQ takes the form of a seven-point scale thatcorresponds to the bandwidth of the sourceregion, with higher settings reserved for more

severe hearing losses. A gain control is alsoprovided; this control independently adjusts thelevel of the translated spectral feature.

Summary of Techniques

Fig. 11 provides a visual schematic for compar-ing the commercially available FL techniques.Frequency is represented by the color map—thelowest frequencies are red (bottom) and thehighest frequencies are violet (top). The firstbar shows the input band and indicates that thesource region is the frequency range for whichthe hearing aid cannot provide sufficient audi-bility. The following four plots show how eachtechnique approaches the goal of bringing downthis information into a region of aided audibility.

Fig. 12 shows a classification of thesetechniques along two dimensions: algorithmactivation (input dependent versus always ac-tive) and technique that serves as the basis forFL (compression versus transposition). Keycharacteristics of frequency compression are(1) the target region (the frequency range whereinformation is moved to) is contained withinthe source region (the frequency range oranalysis band that is subject to lowering), (2)the bandwidth of the source region is reduced,and (3) the start frequency (which can be 0 Hz)is like an anchor that does not move. Keycharacteristics of frequency transposition/trans-lation are (1) there is less overlap between targetand source regions, (2) the bandwidth of thesource region is not reduced, and (3) the startfrequency is moved to a lower frequency (i.e.,there is a mixing of lowered and unloweredsignals).

Figure 8 A family of frequency input-output curvesfrom the second generation implementation of non-linear frequency compression for one particularhearing loss. For each setting, the first number is thestart frequency, followed by the compression ratio.The dotted line represents the maximum audibleoutput frequency achievable with the hearing aid fora hypothetical hearing loss. Asterisks indicate theonly appropriate settings for this hypothetical lossbased on criteria outlined in the text (i.e., audibilityfor the lowered signal in a way that does not restrictthe audible bandwidth of the processed signal).

Figure 9 A spectrogram of the sentence “children like strawberries” from Fig. 1 after processing withspectral envelope warping. Boxes highlight the visually identifiable energy altered by the processing.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Modeling the Interaction between the

Signal Processing, Hearing Loss, and

Speech

Historically, attempts to implement FL havebeen limited to individuals with severe toprofound SNHL.29,31–35 For these individua-ls, arguments for the use of FL are relativelyeasy to make because deficits in speech recog-nition increase as high-frequency audibilitydecreases.2,36–38 However, as the cutoff fre-quency of audibility decreases, the challengesinvolved with FL increase. Reasons for thisinclude the fact that there is more lost informa-tion that needs to be recovered and a corre-spondingly smaller region for recoding it. In

addition, the recoded information must bemoved to regions where critical low-frequencyspeech information (i.e. formants) might al-ready exist.

On the other hand, arguments justifyingthe use of FL for individuals with mild tomoderate SNHL are more difficult to makebecause the overall deficit attributed to band-width reduction for these losses is less, whichmeans that potential benefit is also less. Inaddition, early FL technology involved drasticalterations of the signal and possible audibleartifacts.33,39 As such, the likelihood of “doingmore harm than good” was relatively high.Modern techniques and advances in signalprocessing have reduced some of these risksassociated with FL and, in general, may bemore amenable to listeners withmilder degreesof SNHL, especially children who depend onthe full bandwidth of speech for normal speechand language development (see followingdiscussion).

As just highlighted, the relationship be-tween SNHL and FL seems to be give and take.As the severity of loss increases, the deficit andcorresponding potential benefit increase alongwith the risks associated with using FL.35,40,41

With less severe loss, there is less deficit andpotential benefit, but also less risk involved withFL.42 This relationship can be better under-stood using the schematic in Fig. 13. On theabscissa is the amount of FL. Because theacoustic fidelity of the signal is radically alteredby FL, the abscissa also is labeled “distortion.”

Figure 10 Author’s rendition of how spectral enve-lope warping adds information from the high-fre-quency spectrum to the low-frequency spectrumwhile maintaining the harmonic structure of thesource signal and the original high-frequency spectralcontent.

Figure 11 Visual schematic for comparing the four frequency-lowering techniques. See text for details.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

FL might be considered “constructive distor-tion” when it aids in speech recognition and“destructive distortion” when it does not,43 sothe term is appropriate either way. On theordinate is information, which can be quanti-fied using information-theoretic terms like bits,or units from the Speech Intelligibility Index,44

and so on. The dashed line represents thepotential information gained by moving other-wise inaudible high-frequency content to low-er-frequency regions of audibility. Up to acertain point, the amount of high-frequencyinformation increases with increases in FL,beyond which, information decreases as cochle-ar limitations take over (e.g., severe amounts offrequency compression). The line does not go

below zero because it represents only thatinformation that is inaccessible with conven-tional amplification, so presumably FL cannotmake this information worse than not having itat all. The dotted line represents the potentialdecrease in information for that part of thespectrum that can be amplified normally. Itnever goes above zero because it is hard toimagine that FL can actually make this infor-mation better. For small degrees of FL, thisinformation may be undisturbed; for example,when NFC has a high start frequency.42,46

However, as FL increases, the informationcontained in the low-frequency spectrum be-comes degraded. The magnitude of the line forthe high-frequency content is intentionally lessthan that for the low-frequency content torespect the differences in the amount eachcontributes to overall speech recognition.44

The absolute magnitude and the underlyingform of each function will then shift dependingon the severity of loss, the specific FL technolo-gy and settings, and the speech sounds involved.

The thick solid line in Fig. 13 representsthe summation of information gained andinformation lost attributable to FL. The goalof the hearing aid dispenser is to choose the FLsetting that maximizes this function for theindividual. Although one cannot know whatthe underlying function is, or where each set-ting is at on the function, probe microphonemeasurements or subjective listening tests canhelp guard against the two extremes.40 The firstextreme occurs when the FL information ismoved to a region that is still inaudible for thelistener. If no additional information is madeaudible, then no benefit should be expected.Although this seems obvious, confusion aboutthe technology or a failure to verify aidedaudibility can result in this scenario. For exam-ple, as already mentioned, with Widex all theFL occurs below the start frequency whereaswith Phonak all the FL occurs above the startfrequency. If the understanding for the formerwere confused for the latter, it is likely that littleto none of the FL information would be audi-ble. Verification of audibility for frequency-lowered speech should also be a considerationwhen evaluating research outcomes asin Table 1. For example, the earliest reportedinvestigations of NFC with adults did not

Figure 12 Classification of the frequency-loweringtechniques. Inserts correspond to the plots shownin Fig. 11.

Figure 13 Schematic of how information in thespeech signal is affected by the distortion introducedby frequency lowering. The dashed line representsthe potential increase of information gained bymoving otherwise inaudible high-frequency contentto lower-frequency regions of audibility. The dottedline represents the potential decrease in informationfor that part of the spectrum that can be amplifiednormally. The solid black line represents the summa-tion of information gained and information lostattributable to frequency lowering. It is hypothesizedthat the underlying forms of the functions willdepend on the factors depicted in Fig. 2.


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

report if or how audibility for the frequency-lowered speech was obtained, thereby limitinginferences or comparisons one might draw.34,35

The other extreme that must be guardedagainst is unintentionally limiting the audiblebandwidth by choosing an overly aggressive FLsetting. After all, the treatment should followthe principle of “do no harm.” Fig. 13 illustratesthe possible negative effects of FL. Shown bythe solid black line, as FL is made moreaggressive and a broader input bandwidth isaffected, a patient’s ability to extract informa-tion from the lowered signal decreases. Not onlydoes the “information value” of the recodedhigh-frequency content decrease, but the infor-mation from low-frequency content is progres-sively degraded as well. The exact details will, ofcourse, depend on the specific technology.Adults being fit with this FLwill verbally objectif they feel that they are losing too muchinformation, but young children cannot. Thisrisk of impeding information extraction haslead manufacturers to intentionally limit theadjustment parameters associated with FL and/or provide the dispenser with recommendedcandidacy guidelines. Regardless of who isbeing fit with FL hearing aids, a guidelinethe author recommends is to use probe micro-phone measurements to obtain the maximumaudible frequency after fine-tuning the hearingaid with FL deactivated, and then do the samewith FL activated using the settings underconsideration to ensure that audible bandwidthis not limited by FL.

To assist the dispenser in choosing FLsettings for NFC and LFT, the author hasdeveloped online tools that visually plot howfrequencies are altered by the different settings(available at www.tinyURL.com). For NFC,the basic principles can be visualized in Fig. 8.For a given audiogram, only a limited numberof combinations of start frequency and com-pression ratio are available to the dispenser. Inthis example, we will assume that the maxi-mum audible frequency in the output that canbe obtained with NFC deactivated is 3.6 kHz,as indicated by the dashed line. From thefigure, it should be clear that only a selectnumber of the available options, as indicatedby the asterisks, avoid the two extremes justdescribed. On the one extreme, none of the FL

signal will be audible for startfrequencies � 3.6 kHz. On the other extreme,settings with start frequencies � 1.7 kHz be-gin to restrict the audible bandwidth of theoutput. Of the remaining settings, the dis-penser might reasonably choose the one thatobjectively maximizes the bandwidth of theinput signal that is made available in theoutput after FL or the one that is subjectivelymost pleasing to the listener.40,45 For LFT, inaddition to these two extremes, one mightwant to consider how the information in theinput is repackaged in the output. Forexample, Fig. 5A shows how a low startfrequency might lead to less than optimaloutcomes because too many frequencies inthe input are over represented in the outputdue to the overlap between the untransposed,transposed, and expanded transposedsignals. Fig. 5B on the other hand, showsthe opposite in which a high start frequencycan lead to an island of intermediate inputfrequencies that are still inaudible afterFL. Fig. 5C shows an ideal scenario in whichthe input frequency corresponding to whereaudibility for an untransposed signal ends isclose to the start frequency for transpositionand where audibility for this signal ends isclose to the start frequency for expandedtransposition.

Some Data

Two studies by the author highlight how thechoice of technology and its settings can influ-ence outcomes for different hearing losses.Alexander et al investigated the efficacy ofLFT and NFC using 24 adults with mild tomoderate SNHL and 24 normal-hearing con-trols.42 Participants listened monaurallythrough headphones to a series of nine fricativesand affricates spoken by three women in avowel-consonant context that had been mixedwith speech-shaped noise at 10-dB signal-to-noise ratio and recorded through hearing aidswith LFT or NFC that were programmed for amild to moderate loss such that FL occurredonly for input frequencies � 4 kHz. It shouldbe noted that the uses of LFT and NFC forthese losses were outside the recommendedcandidacy guidelines established by the


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

manufacturers. Control stimuli included re-cordings made while FL was deactivated ineach hearing aid (restricted bandwidth condi-tion) and these same recordings mixed withhigh-pass filtered versions of the input stimuli,so that average audibility extended out to9 kHz (wide bandwidth condition). Consis-tent with previous findings that demonstratedthe importance of high-frequency informationfor fricative identification,7,19 performance forthe hearing-impaired listeners in the widebandwidth conditions was significantly betterthan the restricted bandwidth conditions. In-dividual performance for conditions whereNFC provided audibility for input frequenciesup to 8 to 9 kHz was similar to the widebandwidth conditions, indicating that the ben-efit observed with increasing bandwidth alsocan be obtained using NFC. In contrast, per-formance with LFT for both hearing-impairedand normal-hearing listeners was significantlyworse compared with the restricted bandwidthcondition by about 10 and 20%, respectively.Significant differences between most condi-tions could be largely attributed to an increaseor decrease in confusions for the phonemes /s/and /z/. Differences in outcomes between thetwo FL techniques might be attributed to thedegree to which LFT altered the low-frequen-cy spectrum compared with NFC in this par-ticular population of listeners. As discussedlater, it is possible that this difference wouldhave become less following extended experi-ence with the technology.

Using simulated hearing aid processing inMATLAB (i.e., “the Purdue hearing aid sim-ulator”) withNFCmodeled after Simpson et aland flexible multichannel wide dynamic rangecompression, 34Alexander investigated the ef-fect of varying the start frequency and com-pression ratio for two fixed outputbandwidths.46 Twenty-eight listeners withmild to moderately severe SNHL identifiedconsonants and vowels from nonsense syllablesin noise. All speech output was low-passfiltered at 3.3 or 5.0 kHz across two groupsof listeners to control for high-frequencythresholds when simulating two clinical sce-narios whereby the dispenser has a variety ofNFC options for repackaging differentamounts of high-frequency information in a

limited band of audibility. For both groupsthere was significant improvement in fricative/affricate identification for most NFC settingsrelative to the low-pass control conditions.However, when start frequency was low (1.6kHz), there was a decrease in vowel and non-fricative consonant identification. Recognitionof these sounds improved when the startfrequency was increased (� 2.2 kHz) eventhough the compression ratio also had to beincreased to provide audibility to the sameband of input frequencies, a process thatreduced spectral resolution within the FLsignal. Alternatively, when less compressionwas used with the low start frequency, voweland nonfricative consonant identification wascloser to that for the low-pass control, al-though this came at the expense of bringingless high-frequency fricative informationdown into the range of audibility(cf. Fig. 13). Overall, the results of this studyindicate that many factors likely determinehow much information individual listenerscan extract from frequency-lowered speech,including the frequency regions altered byFL and the severity of loss in the regions towhere information is moved.

INTRINSIC FACTORS RELATED TOTHE ABILITY TO LEARN ANDBENEFIT FROM FREQUENCY-LOWERED SPEECHThe second set of factors in Fig. 2 to bediscussed are highly interconnected and relateto intrinsic characteristics of the individuallisteners and their abilities to make use of thenew and altered speech cues associated with FL.These factors include age of fitting, age ofhearing loss onset, listening experience, andcognitive factors. The preceding section indi-cates that each FL strategy is a form of speechrecoding that uniquely alters the informationcontained in individual speech sounds at theauditory periphery in a complex manner. Forlisteners to benefit from the recoded speechinformation, they must first learn how to inter-pret the new signal. For some FL settings, littleto no learning may be necessary and immediatebenefit can observed, even in the laborato-ry.29,45,46 For example, if NFC has a high start


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

frequency (e.g., > 3,500Hz), FL will primarilyaffect information contained in the aperiodichigh-frequency spectrum by making it nar-rower in bandwidth. Because the processedsignal can have at least the same sound qualityas conventional amplification,47,48 listenerswith SNHL may not even notice the alterationfor most speech sounds. The potential forbenefit is related to the extent that the newinformation made available with FL reducesuncertainty about the spoken message. Thedegree to which listeners have uncertainty inthe first place will depend on the linguisticcontext of the message and the listeners’ abili-ties to use this context. The role of linguisticcontext was discussed earlier. Listeners’ abili-ties to use context to bootstrap understandingof a message fragmented by “information drop-outs” (inaudibility of critical acoustic cues)depends on their knowledge of grammar, se-mantics, and pragmatics. This is where wemight expect to see differences between chil-dren and adults. It also introduces a furtherdistinction between whether the loss is pre- orpostlinguistic, because listeners who have au-ditory experience with the natural productionsof the recoded sounds might adapt morequickly to them.

For other FL settings, more implicit and/orexplicit learning may be necessary before fullbenefit can be realized. These settings likelyinvolve a manipulation of the primary formantsof speech, which generally reside in the part ofthe spectrum below 3,500 Hz or so.49 For thesesettings, altered sound quality might be ob-served along with new perceptual confu-sions.29,40,48,51 Whether listeners can adapt tothe altered sound quality is an important con-sideration because the technology will likely berejected if they cannot. There is some indicationthat sound quality or preference ratings canimprove for both adults and children over timewith modern FL strategies.50,52 Assuming thatsound quality is not an issue, other factors thatmight determine benefit are listeners’ experi-ence with the technology, which can be gainedimplicitly through acclimatization and/or ex-plicitly through training, and the listeners’underlying cognitive ability to apply the neces-sary effort involved with the perceptual learningprocess.

Children versus Adults

Children might experience greater benefit fromFL compared to adults simply because theyappear to have a greater “deficit” when identify-ing speech under identical conditions. In otherwords, in conditions where adults are perform-ing near the ceiling of their abilities usingconventional amplification, children might stillbe able to benefit from additional informationgained via frequency-lowered speech. As men-tioned, when linguistic knowledge confers anadvantage, adult–child differences in speechrecognition might be expected. For example,when identifying words or sentences in noise,children require more favorable signal-to-noiseratios for similar performance.53–57 However,even when linguistic knowledge is of littleapparent value, as when identifying nonsensesyllables, children still underperform adults andrequire higher levels of audibility.18,58–61 Rea-sons for this are beyond the scope of this article,but likely include differences in phonologicaldevelopment and phonotactic knowledge.62

That is, adults have relatively stable categoriesfor the various acoustic forms sounds in alanguage can take and have an implicit knowl-edge about the allowable sound sequences. Thishelps to restrict the range of possible responseoptions even when nonwords are tested.63Morerelevant to this discussion are findings thatsuggest that children benefit more than adultswhen the availability of speech information isaugmented by increases in signal bandwidth.7,64

Whether this finding holds when the sameinformation is made available via FL has yetto be demonstrated.

Only a small number of studies have testedadults and children using identical methodolo-gies, which is an important evaluation criterionbecause different procedures, especially differentmethods for selecting individual FL settings,can significantly affect outcomes as noted pre-viously. Glista et al reported outcomes for bothchildren and adults with sloping high-frequencySNHL ranging from moderately severe to pro-found.40 Awithdrawal design was used in whichperformance following exposure to NFC for aminimum of 3 to 4 weeks was compared withterminal performance following exposure toconventional processing for another 4 weeksor so. Age group was a significant predictor of


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

performance on a test of plural recognition, with4 of 13 adults and 7 of 11 children showingsignificant improvementwith FL.However, it isunknown whether this result is due to develop-mental differences or to the fact that childrenwere given 5- to 10-dB more gain than adultsper the Desired Sensation Level v5.0 prescrip-tive guidelines.65

Using the Purdue hearing aid simulator,Kopun et al tested adults and children withmild to severe SNHL on monosyllabic wordrecognition before and after 22 minutes of au-dio-visual exposure to two children’s stories thathad been processed with customized amplifica-tion andNFCprocessing.97 They found that bothchildren and adults performed significantly betterbefore and after exposure with NFC compared towithout. Exposure did not provide additionalbenefit for NFC because performance also im-proved for the processing condition withoutNFC. Importantly, they did not find a significantdifference in benefit for children and adults.

In a study that exclusively examined sound-quality judgments for speech and music follow-ing processing using the Purdue hearing aidsimulator with NFC and without NFC at twobandwidths, Brennan et al did not find prefer-ence differences between adults and childrenwith mild to severe SNHL.48 For speech stim-uli, both age groups preferred wide bandwidthprocessing (10,000 Hz) or NFC more oftenthan restricted bandwidth processing (5,000Hz). This preference was positively correlatedwith amplification experience, but there was norelationship with the audiogram. For musicstimuli, no significant differences were foundbetween the three conditions. It should benoted that the start frequency of NFC was> 3,000 Hz for most listeners, which limitedthe processing to frequencies that do not con-tribute much to pitch. These results are some-what in contrast to those of Glista et al,40 whofound that children had a greater preference forNFC than adults, and to those of Auriemmo etal,41 who noted that children more often pre-ferred LFT than adults after initial fitting, butnot after 2 weeks of usage due to increases inpreference by the adults.29

In summary, despite several valid reasons toexpect adult–child differences in benefit and/orpreference for FL, supporting evidence is not

available within studies that used a matchedexperimental design across age groups.

Age of Hearing Loss Onset

Discussion of child–adult differences impliesmore than auditory and linguistic development.Most of the adults in the studies described inthe preceding section had acquired losses(namely presbycusis and noise exposure), andmost of the children had congenital losses (e.g.,genetic causes, meningitis, hyperbilirubinemia,anoxia at birth, etc.). Differences in etiologyhave implications for the degree of involvementof the various cochlear structures (e.g., outerversus inner hair cell loss) and the audiometricconfiguration—factors discussed in the firstpart of this article. For example, Pittman andStelmachowicz analyzed almost 500 audio-grams from a clinical population and foundthat children compared with adults had agreater variety of audiometric configurationsother than the classic sloping loss, had greatervariability in thresholds across frequency, andhad a greater prevalence of asymmetrical hear-ing losses.66 As discussed, audiometric differ-ences influence the amount of unaided deficitfor individual speech sounds and the choice ofFL technology and its settings.

Another factor to consider in this discus-sion is the role of prelinguistic versus postlin-guistic hearing loss. It has been documentedthat young children who are deprived of richauditory input because of untreated SNHL orunaidable high-frequency SNHL often havedelayed phonological development comparedwith their normal-hearing peers.22–25,67,68 Inother words, the acoustic properties that help tocategorize one sound as being different fromanother are less defined (acoustic-phoneticboundaries are more variable), resulting inperceptual confusions and expressive deficits.It is unknown what effect this could have onperception of frequency-lowered speech relativeto conventional hearing aid processing, but theamount of deficit and the ability to learn thenew speech code are likely mediating factors.For example, listeners with better phonologicalrepresentations of the phonemes /s/ and /z/ dueto prior auditory experience might have fewerdeficits with conventional hearing aids, hence


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

less room for improvement as performancenears the ceiling. These listeners might alsobe able to learn to use the new and alteredspeech cues more quickly, whereas those withpoorer phonological representations might re-quire an extended period to acclimatize and/orrequire explicit training.

Acclimatization and Training

Perceptual acclimatization refers to the processby which individuals adapt to altered sensoryinput to maintain optimum performance intheir environment. For speech perception, theeffects of acclimatization are often gauged byimprovement in recognition scores overtime,69–76 but they also could include gradualimprovements in rated sound quality, decreasedlistening effort, and so on.77 The latter are oftenoverlooked, but are important to consider be-cause initial reactions to frequency-loweredspeech, which contribute to the overall experi-ence, might exhibit even greater change overtime compared with conventional amplificationbecause of the increased amount of signalalteration. These considerations are mentionedbriefly elsewhere in this article.

Two studies25,51 that have explicitly exam-ined acclimatization effects for frequency-low-ered speech without training have beenconducted on children. Both studies reportedsignificant improvements in speech recognitionfollowing extended experience listeningthrough hearing aids with NFC. Wolfe et alpresented results from a 6-month follow-up on15 young children (ages 5 to 13 years) withmoderate to moderately severe SNHL whowere initially tested after two counterbalanced6-week intervals in which they listened throughthe same hearing aids with NFC activated ordeactivated during their daily routines.25,78

Discrimination of singular/plural contrasts inquiet, which tests perception of /s/ and /z/, anddiscrimination of high-frequency consonantsembedded in nonsense vowel-consonant-vowelutterances in quiet were significantly improvedby NFC after 6 weeks. Sentence recognition inmultitalker babble did not improve during thistime, however. When tested again at 6 months,plural discrimination maintained at ceiling per-formance levels and there were continued im-

provements in the discrimination of high-frequency consonants relative to testing at6 weeks. Interestingly, there was significantimprovement the signal-to-noise ratio thresh-old for sentence recognition that was not pres-ent after 6 weeks. These results indicate thatsome children may undergo an acclimatizationperiod lasting at least 6 months for NFC.However, maturation and/or learning effectscannot be ruled out as contributing factors inthe Wolfe et al study because there was nocontrol condition or withdrawal conditionwhereby listeners were tested following anacclimatization period withNFC deactivated.25

To minimize the likelihood of maturationeffects, Glista et al tested 6 older children (ages11 to 18 years) with at least moderately severehigh-frequency SNHL in a single-subject de-sign on phoneme detection, plural recognition,discrimination between /s/ and /

R/, and conso-

nant recognition at regular intervals over a 4-month period.51 One listener did not showimprovement over time due to ceiling effects.For the other five listeners, they found signifi-cant improvement (up to 20%) with NFC on atleast one test, especially with fricatives, althoughthe pattern of improvement over time was quitevariable within and between listeners. There-fore, the authors caution about extrapolatingfrom individual results and warn that matura-tion effects are still possible, although unlikely.

Explicit auditory training is another way,and in some cases may be the only way, thatlisteners can get the experience they need tolearn the new and altered speech cues intro-duced by FL. The concept of training onfrequency-lowered speech goes back to theearliest techniques described in the literature,where it was more commonplace than moderntechniques (see Simpson28 for a review). Onereason for this shift in approaches may be thatthe modern techniques described in the recentliterature are usually implemented in wearabledevices, which provides listeners with an op-portunity to acclimatize to the processing. Theonly exceptions that incorporated formal train-ing procedures utilized LFT.29,41

Auriemmo et al tested 10 children (ages 6 to13 years) with severe to profound SNHL onconsonant and vowel recognition in quiet beforeand after a 3-week control period during which


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

they wore a hearing aid with LFT deactivatedduring their daily routines.41 Then they repeat-ed the same testing after two additional 3-weektreatment periods during which LFT in thehearing aid was activated. Throughout thecontrol and treatment phases of the experiment,listeners attended weekly auditory training ses-sions. Themain findings were that during the 6-week period when LFT was activated, conso-nant recognition at a low presentation level(30-dB hearing level) significantly improvedby more than 20% and the production accuracyof /s/ and /z/ improved by more than 10%.There was no significant difference in vowelrecognition when LFT was activated versusdeactivated because performance was near ceil-ing levels during initial testing for both con-ditions. Because listeners received training alongwith their daily exposure to LFT, it cannot bedetermined if the observed effects were due toacclimatization and/or to training.

Kuk et al tested eight adults with severe toprofound SNHL on consonant recognition inquiet and in babble noise, first with LFTdeactivated and activated after the initial fit,then after 1 month of daily exposure and train-ing (� 20 to 30 minutes per a day), and thenafter another month of daily exposure butwithout training.29 A feature analysis of theconfusions revealed that the main source ofimprovement with LFT was fricatives. At apresentation level of 50-dB sound pressure level(SPL), fricative recognition improved 5 to 10%initially and then an additional 10% after2 months. At a presentation level of 68-dBSPL, a significant difference was only seen forfricatives, which improved by 10% after2months. At this level, there was also a decreasein recognition of stop consonants after theinitial fit, which significantly improved after2 months to performance levels slightly betterthan baseline. For speech in noise, the onlysignificant difference was about a 15% improve-ment for fricatives after 2 months of training.Like the Auriemmo et al study, one cannotdetermine if the observed effects were due toacclimatization and/or to training, because lis-teners received training along with their dailyexposure to LFT.41 Furthermore, because therewas no control group that trained on the samehearing aid with LFT deactivated, one cannot

determine if the observed main effects are dueto the additional information introduced byLFT or to simple practice effects. To addressthis concern, a brief follow-up report by Kukand Keenan indicates that there were no train-ing effects per se when using the hearing aidswith LFT deactivated.79

In summary, the few reports that haveexamined the effects of extended listeningexperience with frequency-lowered speech sup-port the notion that at least some listenersrequire a period of acclimatization and/or train-ing with the new speech code before full benefitcan be realized. One can speculate that themoredrastic the alteration of speech cues followingFL, namely involvement of low-frequencyspeech cues, the greater the need for extendedor explicit listening experience for the listener tolearn how to most effectively process the newlyintroduced and altered information.

Cognitive Factors

There has been recent interest in understandinghow cognitive processes influence outcomes forhearing aid algorithms other than FL (seeLunner et al80 for a review). Cognitive process-ing has been implicated in at least two ways.One way is as a moderating variable (e.g., verbalworking memory) that influences who benefitsand under what circumstances, as with widedynamic range compression.81–85 Another wayis as a dependent variable; that is, the constructby which benefit is measured (e.g., in terms oflistening effort, recall, reaction time), as withdigital noise reduction.86–88 As noted in theprevious paragraphs, FL techniques and set-tings that alter more of the natural speech codemay require a learning period consisting ofexposure with or without training to achievemaximum potential. On the one hand, if learn-ing puts a high demand on cognitive resources,then FL might not be beneficial or might evenbe detrimental to those listeners who have fewerresources available, such as the elderly.89–93 Onthe other hand, if FL can provide additionalinformation that helps reduce uncertainty aboutthe spoken message without the need for learn-ing, then it might be able to reduce the cogni-tive load associated with listening in real-worldscenarios, thereby benefiting these same


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

listeners in terms of listening effort, recall,comprehension, multitasking, and so on.92

Two recent studies have explored the rela-tionship between cognitive processing and rec-ognition of speech with NFC.94,95 Using olderadults (ages 62 to 92 years) with mild to severesloping SNHL, Arehart et al94 measured work-ing memory and recognition of sentences innoise that had been processed with a hearing aidsimulator that implemented a form of NFCsimilar to that described by Simpson et al.34

The signal processing differed in that only themost intense peaks in the high-frequency spec-trum were compressed, instead of the entirehigh-frequency spectrum. NFC settings wereintentionally aggressive with start frequenciesas low as 1,000 Hz because the primary purposewas to obtain correlates for speech recognitionof distorted speech, not FL per se. As theamount of distortion increased, speech recog-nition was more adversely affected in listenerswith the poorest working memory abilitiescompared with those with better workingmemory abilities. In contrast, Ellis andMunro95 used similar test materials and thesame cognitive test as Arehart et al94 but did notfind a significant relationship between theworking memory of young normal-hearingadults (ages 18 to 50 years) and recognitionof speech with an NFC start frequency of 1,600Hz. Interestingly, they found a significantrelationship between working memory andrecognition of the unprocessed speech, whichindicates that the variation in working memoryabilities for this nonclinical population shouldhave been sufficiently large to capture an effectfor NFC processing had there truly been one.Further research is clearly needed to betterunderstand the role that cognition plays inprocessing frequency-lowered speech and howthis influences candidacy for this technology.

CONCLUSIONAfter reviewing all the studies mentioned in thisarticle, it should be clear that differences inoutcomes with FL vary along with the depen-dency of the test materials and individual listen-ers on information in the high-frequencyspectrum.Accordingly, when FL shows a benefitfor speech recognition it is primarily for fricative

consonants with restricted linguistic context.Not all of the individuals in the studies reviewedshowed improvement in speech recognition withFL. For most of these, speech recognition withFL was not statistically different from speechrecognition without FL, whichmight offer someassurance from a “do no harm” perspective.However, for some listeners, FL negativelyaffected speech recognition. In light of themodelpresented in Fig. 13, one might question wheth-er the settings in these cases actually optimizedthe total information in the signal. In addition,full benefit might not be realized until afterseveral months of experience with the technolo-gy, especially for speech in noise.

Although the focus of measuring benefithas been on speech recognition, one should beopen to the possibility that benefit from FLmight be less evident in terms of decreasedlistening effort, improved speech productionand vocal quality, improved localization andspatial unmasking, and so on, even when thereare no observable improvements in speechrecognition as measured in the laboratory.

Finally, the reader is reminded of the need to(1) understand the specific FL technology beforeimplementing it (i.e., what frequencies are low-ered, where are they lowered to, when are theylowered, and how are they lowered), (2) verifyelectroacoustically that FL moves additionalspeech information to a region of audibility, butdoes not unduly limit the audible bandwidth ofthe entire signal, and (3) validate outcomes toensure effectiveness for the individual.

ACKNOWLEDGMENTS

Varsha Hariram helped prepare the referencesand the information in Table 1, although anyerrors contained therein are solely the respon-sibility of the author.

REFERENCES

1. Hogan CA, Turner CW. High-frequency audibil-ity: benefits for hearing-impaired listeners. J AcoustSoc Am 1998;104:432–441

2. Ching TY, Dillon H, Byrne D. Speech recogni-tion of hearing-impaired listeners: predictionsfrom audibility and the limited role of high-frequency amplification. J Acoust Soc Am 1998;103:1128–1140


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

3. Turner CW, Cummings KJ. Speech audibility forlisteners with high-frequency hearing loss. Am JAudiol 1999;8:47–56

4. Baer T, Moore BCJ, Kluk K. Effects of low passfiltering on the intelligibility of speech in noise forpeople with and without dead regions at highfrequencies. J Acoust Soc Am 2002;112(3 Pt 1):1133–1144

5. Vickers DA, Moore BCJ, Baer T. Effects of low-pass filtering on the intelligibility of speech in quietfor people with and without dead regions at highfrequencies. J Acoust Soc Am2001;110:1164–1175

6. Boothroyd A, Medwetsky L. Spectral distributionof /s/ and the frequency response of hearing aids.Ear Hear 1992;13:150–157

7. Stelmachowicz PG, Pittman AL, Hoover BM,Lewis DE. Effect of stimulus bandwidth on theperception of /s/ in normal- and hearing-impairedchildren and adults. J Acoust Soc Am 2001;110:2183–2190

8. Fox RA, Nissen SL. Sex-related acoustic changesin voiceless English fricatives. J Speech Lang HearRes 2005;48:753–765

9. Mann VA, Repp BH. Influence of vocalic contexton perception of the /s/-/s/ distinction: spectralfactors. Percept Psychophys 1980;28:213–228

10. Whalen DH. Effects of vocalic formant transitionsand vowel quality on the English [s]-[s] boundary. JAcoust Soc Am 1981;69:275–282

11. Heinz JM, Stevens KN. On the properties ofvoiceless fricative consonants. J Acoust Soc Am1961;33:589–596

12. Nittrouer S, Studdert-Kennedy M. The role ofcoarticulatory effects in the perception of fricativesby children and adults. J Speech Hear Res1987;30:319–329

13. Zeng FG, Turner CW. Recognition of voicelessfricatives by normal and hearing-impaired subjects.J Speech Hear Res 1990;33:440–449

14. Nittrouer S. Age-related differences in perceptualeffects of formant transitions within syllables andacross syllable boundaries. J Phonetics 1992;20:351–382

15. Nittrouer S, Miller ME. Predicting developmentalshifts in perceptual weighting schemes. J AcoustSoc Am 1997;101:2253–2266

16. Nittrouer S, Miller ME. Developmental weightingshifts for noise components of fricative-vowelsyllables. J Acoust Soc Am 1997;102:572–580

17. Hedrick MS. Effect of acoustic cues on labelingfricatives and affricates. J Speech Lang Hear Res1997;40:925–938

18. Pittman AL, Stelmachowicz PG. Perception ofvoiceless fricatives by normal-hearing and hear-ing-impaired children and adults. J Speech LangHear Res 2000;43:1389–1401

19. Stelmachowicz PG, Lewis DE, Choi S, Hoover B.Effect of stimulus bandwidth on auditory skills in

normal-hearing and hearing-impaired children.Ear Hear 2007;28:483–494

20. Pittman AL. Short-term word-learning rate inchildren with normal hearing and children withhearing loss in limited and extended high-frequen-cy bandwidths. J Speech Lang Hear Res 2008;51:785–797

21. Hornsby BWY, Johnson EE, Picou E. Effects ofdegree and configuration of hearing loss on thecontribution of high- and low-frequency speechinformation to bilateral speech understanding. EarHear 2011;32:543–555

22. Stelmachowicz PG, Pittman AL, Hoover BM,Lewis DE. Aided perception of /s/ and /z/ byhearing-impaired children. Ear Hear 2002;23:316–324

23. Moeller MP, Hoover BM, Putman CA, et al.Vocalizations of infants with hearing loss comparedwith infants with normal hearing: Part I—phoneticdevelopment. Ear Hear 2007;28:605–627

24. Moeller MP, McCleary E, Putman C, Tyler-Krings A, Hoover B, Stelmachowicz P. Longitu-dinal development of phonology and morphologyin children with late-identified mild-moderatesensorineural hearing loss. Ear Hear 2010;31:625–635

25. Wolfe J, John A, Schafer EC, et al. Long-termeffects of non-linear frequency compression forchildren with moderate hearing loss. Int J Audiol2011;50:396–404

26. Rudmin F. The why and how of hearing /s/. VoltaReview 1983;85:263–269

27. Elfenbein JL, Hardin-Jones MA, Davis JM.Oral communication skills of children who arehard of hearing. J Speech Hear Res 1994;37:216–226

28. Simpson A. Frequency-lowering devices for man-aging high-frequency hearing loss: a review. TrendsAmplif 2009;13:87–106

29. Kuk F, Keenan D, Korhonen P, Lau CC. Efficacyof linear frequency transposition on consonantidentification in quiet and in noise. J Am AcadAudiol 2009;20:465–479

30. SermanM, Hanneman R, Kornagel U.White paper:Micon frequency compression. Siemens AG, 2012.Available at: http://hearing/siemens.com/Resources/literatures/Global/publications/2012%20-%20white-paper%20mocon%20frequcny%20compression.pdf?_blob=publicationfile. Accessed on March 19, 2013

31. Braida LD, Durlach NI, Lippmann RP, Hicks BL,Rabinowitz WM, Reed CM. Hearing aids—areview of past research on linear amplification,amplitude compression, and frequency lowering.ASHA Monogr 1979;19:1–114

32. Parent TC, Chmiel R, Jerger J. Comparison ofperformance with frequency transposition hearingaids and conventional hearing aids. J Am AcadAudiol 1998;9:67–77


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

33. McDermott HJ, Dorkos VP, Dean MR, ChingTYC. Improvements in speech perception with useof the AVR TranSonic frequency-transposinghearing aid. J Speech Lang Hear Res 1999;42:1323–1335

34. Simpson A, Hersbach AA, McDermott HJ. Im-provements in speech perception with an experi-mental nonlinear frequency compression hearingdevice. Int J Audiol 2005;44:281–292

35. Simpson A, Hersbach AA, McDermott HJ. Fre-quency-compression outcomes in listeners withsteeply sloping audiograms. Int J Audiol 2006;45:619–629

36. Turner CW, RobbMP. Audibility and recognitionof stop consonants in normal and hearing-impairedsubjects. J Acoust Soc Am 1987;81:1566–1573

37. Dubno JR,DirksDD,EllisonDE. Stop-consonantrecognition for normal-hearing listeners and lis-teners with high-frequency hearing loss. I: Thecontribution of selected frequency regions. J AcoustSoc Am 1989;85:347–354

38. Horwitz AR,Dubno JR, Ahlstrom JB. Recognitionof low-pass-filtered consonants in noise with nor-mal and impaired high-frequency hearing. J AcoustSoc Am 2002;111(1 Pt 1):409–416

39. Kuk F, Korhonen P, Peeters H, Keenan D, JessenA, Andersen H. Linear frequency transposition:extending the audibility of high frequency infor-mation. Hearing Review 2006;13:42–48

40. Glista D, Scollie S, BagattoM, Seewald R, Parsa V,Johnson A. Evaluation of nonlinear frequency com-pression: clinical outcomes. Int J Audiol 2009;48:632–644

41. Auriemmo J, Kuk F, Lau C, et al. Effect of linearfrequency transposition on speech recognition andproduction of school-age children. J Am AcadAudiol 2009;20:289–305

42. Alexander JM, Lewis DE, Kopun JG, McCreeryRW, Stelmachowicz PG. Effects of frequencylowering in wearable devices on fricative and affri-cate peroeption, 2008 International Hearing AidConference, August 13–16, 2008; Lake Tahoe, CA

43. Kuk F, Jessen A, Baekgaard L. Ensuring high-fidelity in hearing aid sound processing. HearingReview 2009;16:34–43

44. ANSI. ANSI S3.5–1997 (R2007), Methods forCalculation of the Speech Intelligibility Index.New York, NY: American National StandardsInstitute

45. McCreery RW, BrennanMA, Hoover B, Kopun J,Stelmachowicz PG. Maximizing audibility andspeech recognition with nonlinear frequency com-pression by estimating audible bandwidth. EarHear 2012;20:1–4

46. Alexander JM. Nonlinear frequency compression:balancing start frequency and compression ratio.Paper presented at: 39th Annual meeting of the

American Auditory Society; March 8–10, 2012;Scottsdale, AZ

47. Bohnert A, Nyffeler M, Keilmann A. Advantagesof a non-linear frequency compression algorithm innoise. Eur Arch Otorhinolaryngol 2010;267:1045–1053

48. Brennan M, McCreery R, Kopun J, Hoover B,Stelmachowicz PG. Signal processing preferencefor music and speech in adults and children withhearing loss, 39th Annual meeting of the AmericanAuditory Society, March 8–10, 2012; Scottsdale,AZ

49. Hillenbrand J, Getty LA, Clark MJ, Wheeler K.Acoustic characteristics of American English vow-els. J Acoust Soc Am 1995;97(5 Pt 1):3099–3111

50. Kuk F, Peeters H, Keenan D, Lau C. Use offrequency transposition on in thin-tube, open-earfittings. Hearing Journal 2007;60:59–63

51. Glista D, Scollie S, Sulkers J. Perceptual acclimati-zation post nonlinear frequency compression hear-ing aid fitting in older children. J SpeechLangHearRes 2012;55:1765–1787

52. Nyffeler M. Study finds that non-linear frequencycompression boosts speech intelligibility. HearingJournal 2008;61:22–26

53. Nittrouer S, Boothroyd A. Context effects inphoneme and word recognition by young childrenand older adults. J Acoust Soc Am 1990;87:2705–2715

54. Gravel JS, Fausel N, Liskow C, Chobot J. Child-ren’s speech recognition in noise using omni-direc-tional and dual-microphone hearing aid technology.Ear Hear 1999;20:1–11

55. Fallon M, Trehub SE, Schneider BA. Children’suse of semantic cues in degraded listening environ-ments. J Acoust Soc Am 2002;111(5 Pt 1):2242–2249

56. Hall JW III, Grose JH, Buss E, Dev MB. Spondeerecognition in a two-talker masker and a speech-shaped noise masker in adults and children. EarHear 2002;23:159–165

57. Bonino AY, Leibold LJ, Buss E. Release fromperceptual masking for children and adults: benefitof a carrier phrase. Ear Hear 2013;34:3–14

58. Stelmachowicz PG, Hoover BM, Lewis DE, Kor-tekaas RWL, Pittman AL. The relation betweenstimulus context, speech audibility, and perceptionfor normal-hearing and hearing-impaired children.J Speech Lang Hear Res 2000;43:902–914

59. Scollie SD. Children’s speech recognition scores:the Speech Intelligibility Index and proficiencyfactors for age and hearing level. Ear Hear 2008;29:543–556

60. Nishi K, Lewis DE, Hoover BM, Choi S, Stelma-chowicz PG. Children’s recognition of AmericanEnglish consonants in noise. J Acoust Soc Am2010;127:3177–3188


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

61. McCreery RW, Stelmachowicz PG. Audibility-based predictions of speech recognition for childrenand adults with normal hearing. J Acoust Soc Am2011;130:4070–4081

62. Hnath-Chisolm TE, Laipply E, Boothroyd A.Age-related changes on a children’s test of senso-ry-level speech perception capacity. J Speech LangHear Res 1998;41:94–106

63. Edwards J, Beckman ME, Munson B. The inter-action between vocabulary size and phonotacticprobability effects on children’s production accura-cy and fluency in nonword repetition. J SpeechLang Hear Res 2004;47:421–436

64. Kortekaas RWL, Stelmachowicz PG. Bandwidtheffects on children’s perception of the inflectionalmorpheme /s/: acoustical measurements, auditorydetection, and clarity rating. J Speech Lang HearRes 2000;43:645–660

65. Scollie S, Seewald R, Cornelisse L, et al. TheDesired Sensation Level multistage input/outputalgorithm. Trends Amplif 2005;9:159–197

66. Pittman AL, Stelmachowicz PG. Hearing loss inchildren and adults: audiometric configuration,asymmetry, and progression. Ear Hear 2003;24:198–205

67. Jerger S, Lai L, Marchman VA. Picture naming bychildren with hearing loss: I. Effect of semanticallyrelated auditory distractors. J Am Acad Audiol2002;13:463–477

68. Eisenberg LS. Current state of knowledge: speechrecognition and production in children with hear-ing impairment. Ear Hear 2007;28:766–772

69. Cox RM, Alexander GC. Maturation of hearingaid benefit: objective and subjective measurements.Ear Hear 1992;13:131–141

70. Gatehouse S. The time course and magnitude ofperceptual acclimatization to frequency responses:evidence from monaural fitting of hearing aids. JAcoust Soc Am 1992;92:1258–1268

71. Gatehouse S. Role of perceptual acclimatization inthe selection of frequency responses for hearingaids. J Am Acad Audiol 1993;4:296–306

72. Bentler RA, Niebuhr DP, Getta JP, Anderson CV.Longitudinal study of hearing aid effectiveness. I:Objective measures. J Speech Hear Res 1993;36:808–819

73. Turner CW, Humes LE, Bentler RA, Cox RM. Areview of past research on changes in hearing aidbenefit over time. Ear Hear 1996;17(3, Suppl):14S–25S

74. Cox RM, Alexander GC, Taylor IM, Gray GA.Benefit acclimatization in elderly hearing aid users.J Am Acad Audiol 1996;7:428–441

75. Horwitz AR, Turner CW. The time course ofhearing aid benefit. Ear Hear 1997;18:1–11

76. Saunders GH, Cienkowski KM. Acclimatizationto hearing aids. Ear Hear 1997;18:129–139

77. Bentler RA, Niebuhr DP, Getta JP, Anderson CV.Longitudinal study of hearing aid effectiveness. II:Subjective measures. J Speech Hear Res 1993;36:820–831

78. Wolfe J, John A, Schafer EC, Nyffeler M, BoretzkiM, Caraway T. Evaluation of nonlinear frequencycompression for school-age children with moderateto moderately severe hearing loss. J Am AcadAudiol 2010;21:618–628

79. Kuk F, Keenan D. Frequency transposition: train-ing is only half the story. Hearing Review 2010;17:38–46

80. Lunner T, Rudner M, Ronnberg J. Cognition andhearing aids. Scand J Psychol 2009;50:395–403

81. Gatehouse S, Naylor G, Elberling C. Linear andnonlinear hearing aid fittings—1. Patterns of ben-efit. Int J Audiol 2006;45:130–152

82. Lunner T, Sundewall-Thoren E. Interactions be-tween cognition, compression, and listening con-ditions: effects on speech-in-noise performance in atwo-channel hearing aid. J Am Acad Audiol 2007;18:604–617

83. Cox RM, Xu J. Short and long compression releasetimes: speech understanding, real-world preferen-ces, and association with cognitive ability. J AmAcad Audiol 2010;21:121–138

84. Foo C, Rudner M, Ronnberg J, Lunner T. Recog-nition of speech in noise with new hearing instru-ment compression release settings requires explicitcognitive storage and processing capacity. J AmAcad Audiol 2007;18:618–631

85. Masterson KM, Alexander JM. Factors influencingrelease from masking with fast vs. slow compres-sion. 38th Annual meeting of the American Audi-tory Society, March 10–12, 2011; Scottsdale, AZ

86. Sarampalis A, Kalluri S, Edwards B, Hafter E.Objective measures of listening effort: effects ofbackground noise and noise reduction. J SpeechLang Hear Res 2009;52:1230–1240

87. Pittman A. Children’s performance in complexlistening conditions: effects of hearing loss anddigital noise reduction. J Speech Lang Hear Res2011;54:1224–1239

88. Pittman A. Age-related benefits of digital noisereduction for short-term word learning in childrenwith hearing loss. J Speech Lang Hear Res 2011;54:1448–1463

89. Humes LE. Factors underlying the speech-recog-nition performance of elderly hearing-aid wearers. JAcoust Soc Am 2002;112(3 Pt 1):1112–1132

90. Humes LE. The contributions of audibility andcognitive factors to the benefit provided by ampli-fied speech to older adults. J Am Acad Audiol2007;18:590–603

91. Humes LE, Floyd SS. Measures of working memo-ry, sequence learning, and speech recognition in theelderly. J Speech Lang Hear Res 2005;48:224–235


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

92. Schneider BA. How age affects auditory-cognitiveinteractions in speech comprehension. Audiol Res2011;1(e10):34–39

93. Gosselin PA,Gagne J-P. Older adults expendmorelistening effort than young adults recognizingaudiovisual speech in noise. Int J Audiol 2011;50:786–792

94. Arehart KH, Souza P, Baca R, Kates JM.WorkingMemory, Age, and Hearing Loss: Susceptibility toHearing Aid Distortion. Ear Hear 2013;(Jan):3

95. Ellis RJ, Munro KJ. Does cognitive function pre-dict frequency compressed speech recognition in

listeners with normal hearing and normal cogni-tion? Int J Audiol 2013;52:14–22

96. Smith J, Dann M, Brown M. An evaluation offrequency transposition for hearing-impairedschool-age children. Deafness Educ Int 2009;11:62–82

97. Kopun J, McCreery R, Hoover B, Spalding J,Brennan M, Stelmachowicz P. Effects of Expo-sure on Speech Recognition with NonlinearFrequency Compression, 39th Annual meetingof the American Auditory Society, March 8–10,2012; Scottsdale, AZ


Dow

nloa

ded

by: P

urdu

e U

nive

rsity

Lib

rarie

s. C

opyr

ight

ed m

ater

ial.

Individual Variability in Recognition of Frequency-Lowered ...

Documents

Transcript of Individual Variability in Recognition of Frequency-Lowered ...