Johnson et al: The Effect of Digital Phase Cancellation Feedback ...€¦ · the feedback pathway...

J Am Acad Audiol 18:404–416 (2007)

404

*Department of Hearing and Speech Sciences, Dan Maddox Hearing Aid Research Laboratory, Vanderbilt University,Nashville, TN

Earl E. Johnson, M.S., Department of Hearing and Speech Sciences, Vanderbilt University, 1215 21st Ave S., Room 8310,Nashville, TN 37232; Phone: 615-936-5087; E-mail: [email protected]

This work was supported by funding from GN ReSound and the Vanderbilt University Dan Maddox Hearing Aid ResearchLaboratory.

The Effect of Digital Phase Cancellation FeedbackReduction Systems on Amplified Sound Quality

Earl E. Johnson*Todd A. Ricketts*Benjamin W.Y. Hornsby*

Abstract

The effect of feedback reduction (FBR) systems on sound quality recorded fromtwo commercially available hearing aids was evaluated using paired comparisonjudgments by 16 participants with mild to severe sloping hearing loss. Thesecomparisons were made with the FBR systems on and off without audiblefeedback and while attempting to control for differences in gain and clinical fittingfactors. Wilcoxon signed rank test analyses showed that the participants wereunable to differentiate between signals that had been recorded with the FBRsystems on and off within the same hearing aid. However, significant between-instrument differences in sound quality were identified. The results support theactivation of the FFT-phase cancellation FBR systems evaluated herein withoutconcern for a noticeable degradation of sound quality.

Key Words: Feedback reduction system, hearing aid, paired comparison, soundquality

Abbreviations: FBR = feedback reduction; FFT = fast Fourier transform; LMS= least mean square; NAL-NL1 = National Acoustic Laboratories Non-linear 1;RMS = root mean square

Sumario

Se evaluó el efecto de los sistemas de reducción de la retroalimentaciñn(FBR) sobre la calidad del sonido registrado por medio de dos auxiliariesauditivos disponibles comercialmente, utilizando juicios de comparación enpares, de 16 participantes con una hipocusia leve a moderada de pendientepronunciada. Estas comparaciones se realizaron con los sistemas FBRapagados y encendidos, sin retroalimentación audible y tratando de controlarlas diferencias en cuanto a ganancia y a factores clínicos de adaptación. Losanálisis de pruebas de rango certificados por Wilcoxon mostraron que losparticipantes no podian diferenciar entre señales que habian sido registradascon los sistemas de FBR encendidos o apagados dentro del mismo auxiliarauditivo. Los resultados apoyan la activación de los sistemas de FBR decancelación de fase FFT, aqui evaluados, sin preocupaciones por unadegradación identificable en la calidad del sonido.

Palabras Clave: Sistema de reducción de la retroalimentación, auxiliarauditivo, comparación en pares, calidad de sonido

Abreviaturas : FBR = reducción en la retroalimentación; FFT = transformaciónrápida de Fourier; LMS = cuadrado medio mínimo; NAL-NL1 = LaboratoriosNacionales de Acústica No lineal 1; RMS = media de la raíz al cuadrado

EEffffeecctt ooff FFBBRR SSyysstteemmss/Johnson et al

405

The presence of audible, acoustic feed-back is a major concern as annoyancecaused by feedback can drastically

reduce patient satisfaction (Kochkin, 2003).In fact, 24% of hearing aid wearers report-ed dissatisfaction with a hearing aid due toaudible noise generated by the hearing aidor “whistling,” a common complaint associ-ated with feedback (Kochkin, 1997). Ahearing aid will start to feedback once thegain of the hearing aid has been increasedat a frequency where the amplitude of thefeedback loop response is greater than 0 dBat a phase multiple of 360 degrees(Hellgren et al, 1999), where a feedbackloop response refers to the frequency,amplitude, and phase characteristics of thepathway from the receiver to the micro-phone. In addition, aspects of both hearingaids and hearing loss exacerbate acousticfeedback problems.

Used in almost all current hearing aids(Dillon, 2001), the aim of low thresholdcompression (often referred to as "widedynamic range compression," or "WDRC")systems is to maximize audibility of speechby placing the acoustic speech cues within areduced dynamic range. Thus, compressionsystems provide increasing gain fordecreasing inputs down to the compressionthreshold. As a result of increased gain forlow inputs, feedback is more likely to occurfor low intensity inputs in compression sys-tems. Furthermore, approximately 90% ofhearing losses in adults are downward slop-ing from 500 Hz to 4 kHz (Dillon, 2001),and prescriptive targets for sloping hearinglosses accordingly apply more gain for high-er frequencies than lower frequencies. Inaddition, the amplitude of the feedback loopresponse is greatest between 2 and 5 kHz,and therefore, feedback is most likely tooccur in this frequency range (Kates, 1999).

CCOOMMBBAATTIINNGG TTHHEE PPRROOBBLLEEMM OOFF FFEEEEDDBBAACCKK

Given how common and deleterious thepresence of hearing aid feedback is, it

is not surprising that a number of methodshave been proposed to reduce the likelihoodof feedback occurring. As described below,some of these methods are quite effective atreducing feedback levels; however, concernsrelated to potential side effects remain.Specifically, it is unknown whether the pos-

itive impact of reduced feedback might beoffset by concomitant reductions in soundquality related to this processing.

SSoolluuttiioonnss ffoorr FFeeeeddbbaacckk

One traditional method for reducing thelikelihood of feedback is to increase themicrophone and receiver separation byincreasing the size of the hearing aid prod-uct worn by the user. Through the use oflarger hearing aids, the feedback loopresponse must be more intense due to theincreased distance separation from thereceiver and microphone to yield audibleoscillations. Decreasing the residual diame-ter of the vent or the selection of a fullyoccluding vent are other techniques thatwill decrease the likelihood of audible feed-back (Kuk, 1994). However, one negativeside effect of reducing the vent size is theincreasing presence of the occlusion effect,which results in reduced sound quality forthe hearing aid wearer’s own voice(Wimmer, 1986). Given the trading rela-tionship between the occlusion effect andacoustic feedback as a function of vent size,it is of considerable interest to reduce theocclusion effect while limiting feedback.

A number of algorithms have been intro-duced using advanced digital signal pro-cessing in modern hearing aids in anattempt to reduce audible feedback, whilealso allowing for greater amounts of gainwith larger vent sizes that reduce problemswith the occlusion effect. Most of the FBR(feedback reduction) systems in new hear-ing aid models are classified as a continu-ous method feedback reduction system(Chung, 2005). A continuous method meansthat the system constantly monitors for thepresence of feedback and updates the can-cellation parameters without interruptingoutput of the hearing aid processed signal.This type of system contrasts earlier imple-mentations of noncontinuous FBR systemsthat introduced an audible probe tone toestimate cancellation parameters to reducedetected feedback. Two implementations ofthe continuous and noncontinuous methodsof FBR systems are notch filtering andphase cancellation.

While not assessed in the current study,the operation of adaptive notch filter sys-tems may be described simply as those thatreduce gain within a frequency band where

feedback is occurring (Agnew, 1996). Thenotch filter in these systems was designedso that it may exist within one channel overa small frequency range or within concur-rent frequency bands over a broader fre-quency range. In addition, some systemshave been designed to incorporate two ormore narrow notches simultaneously. Thistype of system is now less commonly usedby hearing aid manufacturers as a means oflong-term digital feedback reduction incomparison to the next type of FBR systemdiscussed.

The FFT (fast Fourier transform)-phasecancellation method of FBR system imple-mentation attempts to reduce feedback bygenerating a signal of equivalent amplitude180° out of phase with the estimated feed-back pathway frequency response. Thisestimated response is then processed inparallel with incoming signals. The contin-uous digital, FFT-phase cancellationmethod continually adjusts adaptive filterweights while processing signals and track-ing changes in the feedback loop responseutilizing a least mean squares (LMS) adap-tation (Widrow et al, 1976) or other similartechniques. After several recursive itera-tions of estimating the feedback pathway,optimal filter weights best reflecting theactual feedback pathway, as determined bythe LMS differences between the estimatedand actual pathway, are used (Maxwell andZurek, 1995). Feedback is alleviated by fil-tering the hearing aid output with thisapproximation of the feedback path trans-fer function, which may be established dur-ing the initial processes in the hearing aidfitting session (static pathway) and/orwhile monitoring the feedback pathwayduring actual hearing aid use (dynamicpathway). The details related to derivingthe feedback pathway transfer function arehearing aid specific. Accordingly, perform-ance of FBR systems might be expected tovary between two hearing aid productswith FFT-phase cancellation FBR process-ing implemented differently.

EEvvaalluuaattiioonn ooff EElleeccttrrooaaccoouussttiicc FFBBRRSSyysstteemmss

The performance or effectiveness of anFBR system is often only evaluated bymeasuring additional gain before feedback.This is often termed “gain margin” or “gain

headroom.” Depending on the effectivenessof the FFT-phase-cancellation-based FBRsystem, gain headroom can be as large as20 dB before onset of feedback without lossof any of the high-frequency response; how-ever, when measured in real ears, it is gen-erally between 8 and 15 dB (see Kates,1999, for review).

In contrast to simply measuring gainheadroom, Maxwell and Zurek (1995) sug-gested that both gain headroom and inad-vertent side effects of processing be meas-ured when assessing FBR systems afterfinding reduced sound quality for signalsprocessed with feedback reduction tech-niques. These authors suggested that dueto processing of all methods of digital FBRalgorithms, signals of interest can be cor-rupted or possess other annoying features.That is, all signals subjected to FBR systemprocessing have potential negative charac-teristics compared to unprocessed signals,(e.g., spectral dips, phase distortions, orother distortions). However, since 1995there has been much advancement in digi-tal processing related to feedback reductionprocessing, and these artifacts may be lessnotable than before. For example, a studyof a laboratory-based hearing aid feedbackreduction system utilizing a continuousphase cancellation method yielded no sig-nificant effect on the pleasantness andintelligibility of speech using speech-qualityratings (Greenberg et al, 2000). However,no known published study has evaluatedthe effect of continuous FBR systems onperceived sound quality by listeners withhearing impairments with commerciallyavailable, digital hearing aids.

In these commercial systems, the alloca-tion of processing power is spread acrossmultiple digital signal processing tasks inthe hearing aid (e.g., noise reduction, datalogging, etc.) and may limit estimationaccuracy of the feedback loop responseand/or production of the feedback cancella-tion signal resulting in audible artifacts.Additionally, prior sound-quality evalua-tion of FBR systems has mostly concentrat-ed on speech inputs without much attentionto other important signals such as music.Both speech and more tonal music signalsare of interest, as they represent commonlyheard sounds in the daily lives of personswith hearing loss; however, it is unclear ifFBR systems will have differing effects on

JJoouurrnnaall ooff tthhee AAmmeerriiccaann AAccaaddeemmyy ooff AAuuddiioollooggyy/Volume 18, Number 5, 2007

406

sound quality of these two signal types. Forexample, signals such as the musical notesfrom a flute may be inadvertently classifiedas feedback due to their tonal nature, andunintentionally subjected to FBR systemprocessing.

The purpose of this experiment was toexamine the effect of two commerciallyimplemented continuous, FFT-phase can-cellation FBR systems on perceived soundquality by listeners with sensorineuralhearing impairments. FBR systems similarto those selected are increasingly becomingstandard in digital hearing aids (Chung,2005). Paired comparisons of overall soundquality were made on speech and music sig-nals recorded at the output of the hearingaids with FBR activated and deactivated.Recorded signals were used to allow formatching of frequency response and ampli-tude level both across and within instru-ments.

MMEETTHHOODD

PPaarrttiicciippaannttss

A total of 16 participants that ranged inage from 58 to 84 years (mean = 75) madesound-quality comparisons monaurally ofhearing aid-processed speech and musicrecordings. Pure-tone air-conduction hear-ing thresholds were measured on all partic-

ipants indicating sloping hearing losses inthe borderline normal/mild to severe rangefrom 250 to 8000 Hz. Comparisons withprevious clinical records reaffirmed thisfinding and ensured that hearing losseswere sensorineural in nature. The averageaudiogram for the participants’ ear used inthe study is shown in Figure 1. This widerange of hearing loss was included to repre-sent the range of hearing losses that maybe fit with a BTE hearing aid that utilizesFBR processing.

SSiiggnnaall PPrreeppaarraattiioonn

HHeeaarriinngg AAiiddss

Two commercially available BTE hearingaids were selected for this study. Each hear-ing aid was coupled to a single participant’sear using a skeleton earmold with a 3 mmvent (Westone style 12). This earmold wasselected to increase the likelihood of feed-back requiring engagement of the hearingaid FBR systems. The first hearing aid(HA1) utilized an FFT-phase canceller forreducing feedback. The second hearing aid(HA2) was a product that combined bothFFT-phase cancellation and adaptive notchfilters for the reduction of feedback; howev-er, the FFT-phase cancellation portion ofthe system was responsible for the long-term cancellation feedback in the static


407

FFiigguurree 11.. Average audiogram for the test ears used from the study participants with 95% confidenceinterval bars.

feedback pathway assessed in this experi-ment as it should have engaged after 5–10sec of use given this system’s feedbackreduction time constants.

The feedback system for HA1 utilized thecontinuous method of feedback reductionand has been described in two stages: feed-back detection and suppression (Chung,2005). First, the feedback path was esti-mated by internally generating a noiseburst and analyzed for level and frequencyat the microphone input. Quantitatively,the noise burst was a known signal by thehearing aid system, and thus, the feedbackpathway loop response is estimated byexamining the difference of the knownnoise burst and the composition of the noiseburst at the input of the microphone. A sig-nal generated 180° out of phase with theestimated feedback pathway of equivalentamplitude is then processed in parallelwith incoming signals of interest to some-what negate the feedback pathway.Secondly, the system adaptively monitoredincoming signals for feedback. Adaptationtimes for the feedback system in this sys-tem were proprietary (Chung, 2005). Thisspecific FBR system was constrained, thusmeaning it does not adapt to signals thatdeviate significantly from the initial esti-mated feedback pathway (Kates, 1999).That is, this system was deliberately designedto not cancel environmental tone-like or

transient signals that were not feedback.HA2 was a continuous method FBR sys-

tem that has been described in three steps.Like HA1, a noise was internally generatedand analyzed at the microphone input toset the baseline for the slow-acting compo-nent of the feedback reduction system. Asignal 180° out of phase with the estimatedfeedback pathway was then processed inparallel with incoming signals for suppres-sion of static pathway feedback. A series oftonal-like signals were generated for set-ting the baseline for the fast acting compo-nent that monitored feedback in thedynamic pathway. Adaptation times havebeen reported at less than 1 sec for the fast-acting component and 5 to 10 sec for theslow-acting component (Chung, 2005).

SSttiimmuullii PPrreeppaarraattiioonn

Two different test signals were recordedas processed by the hearing aids with oneparticipant. The test signals were two sen-tences (approximately 7 sec in duration)from the lemon passage of the SpeechIntelligibility Rating Test (Cox andMcDaniel, 1989) and 7 sec of Mozart’s FluteConcerto no. 2 in D. These signals are here-in referred to, respectively, as speech andmusic. The spectral characteristics of thesignals are shown in Figure 2. In the figure,the tonality of the music signal is apparent


408

FFiigguurree 22.. Long-term average frequency spectrum of the music signal (excerpt from Mozart’s Flute Con-certo no. 2 in D) and speech signal (sentence excerpt from the Lemon passage of the Speech Intelligi-bility Rating Test).

in contrast to the speech signal. The signalswere presented at a level of 50 dBA in thesound field to the participant who wore thehearing aids programmed to insertion gainNAL-NL1 (National Acoustic LaboratoriesNon-linear 1) prescriptive targets for a 50dB input. Hearing thresholds and insertiongain targets for this participant are shownin Figure 3. The speech and music stimuliwere output from an Echo Darla soundcardon a Pentium 4 computer routed through aGrason-Stadler GSI-61 audiometer to pro-vide amplification and presented in thesound field from a point-source, forward-firing loudspeaker (Tannoy System 600TM)located at a 0° azimuth 1 m from the par-ticipant. In addition to a hearing aid, theparticipant wore an earhook, on which thereference and measurement microphone forthe Fonix 6500-CX system was placed, aswell as a low-noise recording microphone(Etymotic Research, ER-7). The amplifiedtest signals were recorded in the ear canalusing the ER-7 microphone and a secondEcho Darla soundcard.

Due to a default manufacturer setting,HA2 initially limited its gain to 6 dB belowhearing aid detected feedback providingone reason why HA2 provided less gainheadroom than HA1. Thus, HA1 was firstprogrammed to match HA2’s maximumoutput before feedback in both the FBR sys-tem off and on recordings of the speech andmusic signals in the static feedback path-way. Thus, for both hearing aids the gainlevels at the time of recordings were well

below a level to cause feedback. Frequencymatching across aids was verified with theFrye 6500CX creating conditions with sim-ilar gain headroom for the two aids. Theserecordings are referred to herein as “Lower”conditions. For the Speech Lower andMusic Lower gain condition recordings, thegain headroom for both HA1 and HA2 was2.4 dB and 2.23 dB, respectively. Given achange from the manufacturer default of 6dB to 1 dB for the feedback margin setting,maximum gain headroom for HA2 would beexpected to be 5 dB greater than the gainheadroom utilized in these experimentalconditions.

Additional recordings of the signals weremade with HA1 at the hearing aid outputlevel deemed (1 dB) below audible feedback.For these recordings, only HA1 was pro-grammed to maximum gain without audi-ble feedback to create “Higher” gain experi-mental conditions. This second conditionwas evaluated so that the effect on soundquality of the FBR system in HA1 could beexamined in an experimental conditionthat required substantial effectiveness ofthe system unlike the lower gain condi-tions. In the Speech Higher and MusicHigher conditions, gain headroom corre-sponded to 15.75 dB and 13.1 dB. For boththe Lower and Higher gain conditions,increases in gain for the FBR-on conditionswere performed while maintaining the gen-eral frequency response shape of the pre-scribed NAL-NL1 frequency response.

It is important to consider that variations


409

FFiigguurree 33.. Audiogram in dB HL for the participant and insertion gain targets in dB for the hearing aidused when recording experimental stimuli.

in frequency shaping and gain during therecording of a signal and at the time of sig-nal presentation may supersede small dis-tortions of the FBR processed signals and,thus, dominate the rating of sound quality.Consequently, it was important to equatehearing aid conditions (e.g., FBR-off versusFBR-on) for both amplitude and frequencyresponse to determine the effect of FBR sys-tem processing in isolation. Thus, followingthe recording process, all speech and musicfiles were shaped offline using AdobeAudition 1.5 FFT frequency shaping tomatch the frequency response prescribed bythe NAL-NL1 prescriptive method (Byrneet al, 2001) for the participant with thepoorest hearing thresholds at octave fre-quencies 250–8000 Hz (Figure 4). Theseshaped stimuli were presented to all studyparticipants at the same output levels.While this gain and shaping providedgreater than normal audibility for mostparticipants, shaping to this participant’sprescriptive targets was chosen to furtherincrease the audibility of any potential neg-ative attributes in the FBR processed sig-nals for study participants. Thus, the audi-bility of these attributes was at least equiv-alent to or greater than recordings thatcould have been made at individual specificNAL-NL1 prescriptive target that areknown not to provide complete audibility.

The shaped speech and music files wereindividually presented via an ER-2 ear-phone to verify correct shaping in aZwislocki coupler connected to a LarsonDavis 814 sound-level meter. An analysis ofoutput sound pressure level through 6 kHzshowed good agreement between the pre-scribed output and shaped output of thespeech and music files with an average RMS(root mean square) difference of 2.19 dB.

PPrroocceedduurreess

Participants completed round robinpaired comparisons of the recorded teststimuli monaurally. The paired comparisonmethod was chosen over other methods(e.g., category rating and magnitude esti-mation) of accessing sound quality as it is amore sensitive and effective technique fordifferentiating small differences in soundquality (Eisenberg et al, 1997), as wehypothesized that, if present, any differ-ences in sound quality as a byproduct ofFBR processing would be small. Certainly,more realistic of real-world use, the hearingaids could have been fit individually to thepatients for a period of time to comparesound-quality differences. However, thismethod is grossly less adequate for differen-tiating small reliable differences in soundquality than the paired comparison method.


410

FFiigguurree 44.. Audiogram in dB HL for the participant and insertion gain targets in dB used when shapingexperimental stimuli to the frequency response presented to all participants.

The paired comparison method requiresthat the two items of interest remain inmemory for the length of time required tomake direct comparison between the items(Thurstone, 1927a, 1927b; Edwards, 1957).The length of the stimuli (i.e., 7 sec) waschosen over longer duration stimuli (e.g., 1min or longer) as shorter signals, of a fewseconds in duration, are more appropriatethan longer signal segments for paired com-parison testing as the listener needs torecall the sound quality of both signals inshort-term memory when making a prefer-ence judgment. In addition, short signalsegments are preferred for timeliness asjudgments need to be made multiple timesfor each experimental condition to obtain areliable preference (Punch and Parker,1981).

Of the 16 participants, ten made sound-quality assessments with their right earwhile six used their left ear. Ten pairedsound-quality comparisons of the record-ings included the FBR-on and FBR-off con-ditions within each hearing aid, as well asFBR-on across both hearing aids and FBR-offacross both hearing aids. Note that not allpossible sound-quality comparisons wereincluded as comparisons of FBR-off versusFBR-on across aids were not of interest andnot examined (e.g., HA1 FBR-off versusHA2 FBR-on, etc.). All comparisons werepresented a total of eight separate times toestablish a more reliable account of thesound-quality differences in the pair(Punch and Parker, 1981). Thus, each par-ticipant made 80 total comparisons. Theactual presentation order of the pairs wasrandomized within each experimental con-dition block (Speech Lower, Music Lower,Speech Higher, and Music Higher). Thepresentation order of the experimental con-ditions was counterbalanced within a sub-ject and across subjects.

Each of the participants made an overallsound-quality preference judgment for eachpair of shaped recordings (e.g., HA1 FBR-onversus HA1 FBR-off, etc.). A strength ofpreference rating scale was added to accen-tuate the magnitude of the difference in theoverall sound-quality preference for eachpair (Keidser et al, 1995; Ricketts andHornsby, 2005). The strength of preferencerating scale used “Much Better,”“Moderately Better,” and “Slightly Better.”Number values should not represent this

scale as the distance between numbers maynot reflect the amount of differencebetween ordinal rankings (Speaks, 1992;Howell, 2002). Thus, numerals of 1, 2, and3 corresponding respectively to “LittleBetter,” “Moderately Better,” and “MuchBetter” were used to reflect ordinal dis-tances for the strength of preference ratingscale. For each comparison, the preferredsignal according to its strength of prefer-ence was assigned a numeral, and the non-preferred signal was assigned a zero.

RReessuullttss

Nonparametric statistics (i.e., sign testand Wilcoxon signed rank test) were chosenover more commonly used parametric sta-tistics (e.g., matched pairs t-test) due to theordinal versus interval nature of the data.The overall preference judgment resultswithout the rating scale were analyzedwith the sign test and refer to the frequen-cy count of one processed signal preferredover another processed signal in each com-parison. The same comparisons with thestrength of preference rating were also ana-lyzed with Wilcoxon signed rank test. Asexpected, use of the strength of preferencerating increased the likelihood of observinga significant difference in paired compar-isons in almost all cases. In cases where thelikelihood was not increased using thestrength of preference ratings, significantdifferences were not shown using the over-all preference rating either. Results arepresented based on the analyses with theoverall preference and strength of prefer-ence ratings.

Planned, a priori comparisons of maininterest paired the same hearing aid in itsFBR-on and -off state. These pairingsaddressed the study’s main purpose of deter-mining an FBR system’s effect on soundquality. For these six comparisons, an alphalevel of significance was defined at .05.None of these experimental condition pair-ings yielded a significant sound qualitypreference (see Table 1 for details).Additionally, the lack of observed differ-ences using both the speech and music sig-nals suggests no effect of the signal typesexamined with these systems either.

In addition to within-hearing-aid differ-ences, it was also of interest to determine ifdifferences existed between hearing aids


411

when the FBR system was deactivated andactivated (Table 2). This comparison wascompleted to confirm matching of the twohearings aids. Although the two hearingaids could not be matched exactly, the rateof recurrence for sound-quality differencesbetween the two was unexpected. Post hoccomparisons across hearing aids based onan alpha level as defined by .05/4 or 0.0125to control for an increase in familywise errorrate demonstrated significant differences inthree out of four paired comparisons.

When significant differences were noted,HA1 was less preferred than HA2.Representative of this trend, in the Music

Lower gain condition, HA1 was less pre-ferred than HA2 (p < .001). A paired com-parison of the HA1 FBR-off versus HA1FBR-off recordings for a measure of presen-tation order rating bias was also analyzed.Thus, the HA1 FBR-off recording was pre-sented in both the first and second intervalsto determine if participants chose a giveninterval more often. The comparison yieldeda nonsignificant difference in perceivedsound quality (p = .256), suggesting that thelisteners’ preferences were not related to theordering of the paired comparisons.

To represent the data visually to thereader, albeit in an interval scale, Figure 5


412

Table1. FBR-Off versus FBR-On Comparisons for All Experimental Conditions Sign test Wilcoxon signedα = .05 rank test α = .05

Speech Lower ConditionHA1 FBR-on HA1 FBR-off Z = -.265, p = .791 Z = -1.863, p = .062HA2 FBR-on HA2 FBR-off Z = -1.503, p = .133 Z =-1.762, p = .078

Sign test Wilcoxon signedα = .05 rank test α = .05

Speech Higher ConditionHA1 FBR-on HA1 FBR-off Z = -.088, p = .930 Z = -.291, p = .771


Music Lower ConditionHA1 FBR-on HA1 FBR-off Z = -.972, p = .331 Z = -964, p = .335HA2 FBR-on HA2 FBR-off Z = -.088, p = .930 Z = -.201, p = .841


Music Higher ConditionHA1 FBR-on HA1 FBR-off Z = -.088, p = .930 Z = -.201, p = .841

Note: Statistical values for the Sign test relate to the frequency count of the preferences and the Wilcoxon signed rank test relate to the strength of the comparison preferences for all participants. Bolded α valuesrepresent the significant alpha level used to determine significant preferences of sound quality. No asterisk ispresent to indicate a preference as the comparisons did not yield a significant finding.

Table 2. Hearing Aid Comparisons in Lower Gain ConditionsSign test Wilcoxon signedα = .0125 rank test α = .0125

Speech Lower ConditionHA2 FBR-on HA1 FBR-on Z = -1.149, p = .251 Z = -.943, p = .346HA2 FBR-off HA1 FBR-off Z = -2.563, p = .010* Z = -3.697, p < .001*

Control comparisonHA1 FBR-off HA 1 FBR-off Z = -1.326, p = .185 Z = -1.137, p = .256


Music Lower ConditionHA2 FBR-on HA1 FBR-on Z = -3.977, p < .001* Z = -5.129, p < .001*HA2 FBR-off HA1 FBR-off Z = -5.038, p < .001* Z = -6.024, p < .001*

Note: Statistical values for the Sign test relate to the frequency count of the preferences and the Wilcoxon signedrank test relate to the strength of the comparison preferences for all participants. Bolded α values represent thesignificant alpha level used to determine significant preferences of sound quality. The asterisks indicate pairingsin which a significant preference for a hearing aid was present.

demonstrates the effect of activating theFBR system, showing the difference in themean rating of the FBR-off and -on com-parison preferences for each of the twohearing aids. Recall that only HA1 wasexamined in the Speech Higher and MusicHigher conditions. Figure 5 is presentedsuch that a positive difference indicatesmore preference for FBR-on processing anda negative difference indicates more prefer-ence for FBR-off processing. Figure 6 is pre-sented such that a positive difference indi-cates more preference for HA2 and a nega-tive difference indicates more preferencefor HA1 with the exception of the controlcomparison. In creating these figures, thepreferred signal chosen by the participantwas assigned a value of 1 to 3 correspon-ding to the strength of preference ratingscale and a negative 1 to 3 for the nonpre-ferred signal. Thus, for each experimentalcondition, an average value of near zero forthe eight repeated preference judgments byall participants presumes no preference. Itis of note that when the data was viewed asinterval in nature, matched pairs t-tests onthe data showed the same pattern of non-significant and significant findings as theWilcoxon signed rank tests. Thus, an aster-isk, if present in Figures 5 and 6, indicatesa paired comparison for which a significantsound-quality preference was established.

DDiissccuussssiioonn

Recall that the primary purpose of thisinvestigation was to determine if FBR sys-tem processing significantly affected theperceived sound quality of a signal whencompared to a signal not subjected to FBRsystem processing. Activation of the FFT-phase cancellation FBR systems used in thetwo test hearing aids did not lead to signif-icant changes in sound quality for any ofthe six tested conditions that paired thesame hearing aid with the FBR system onand off. Two of these six comparisons werewith HA1’s FBR system on while providingsubstantial gain headroom of 13.1 dB and15.75 dB at the time of recording in theMusic Higher and Speech Higher experi-mental condition, respectively. Thus, clini-cally, the benefit of additional gain affordedto listeners with hearing impairments bythis FBR system is not expected to be offsetby degraded sound quality, at least for theimplementation of FBR system processingused in the test devices. These results arein good agreement with previous work on alaboratory-based hearing aid system with acontinuous FBR system, which also showedno effect on intelligibility or pleasantness ofsound using speech-quality ratings(Greenberg et al, 2000).

That said, a closer visual examination of


413

FFiigguurree 55.. Mean strength of preference rating difference in FBR-off versus -on conditions for all experimen-tal conditions across all subjects. Note that HA1 had greater gain headroom capabilities than HA2, and onlyHA1 was assessed in the “Higher” gain headroom conditions.

Figure 5 showed a slight trend of the FBR-onrecordings being preferred less than theFBR-off recordings across all conditions,even though no difference reached statisti-cal significance. More statistical power inthe experimental design can alwaysincrease the likelihood of identifying smallbut real differences, but this small trend, ifreal, may not be of clinical significance.That is, patients would be unlikely to noticedecrements in sound quality of this magni-tude during the everyday use of their hear-ing aids given that no preference was estab-lished in this study using repeated meas-ures with sequential comparisons.Moreover, if real, the trend may beexplained by increased equivalent inputnoise occurring as a byproduct of the gainheadroom provided by the FBR system andnot a processing artifact of the FBR pro-cessing. An analysis of the overall RMSlevel of equivalent-input noise for the FBR-on and -off recordings showed a differenceof 2.3 dB and 7 dB for the lower and highergain conditions, respectively, at the time ofrecording. Recall that at the time of record-ing the amplitude of the FBR system offand on recordings varied by the amount ofgain headroom but that the FBR system offand on recordings were later equated foroverall RMS level when shaped to the NAL-NL1 prescriptive targets for the partici-pant’s hearing thresholds with the worst

hearing sensitivity. Thus, the increased gainheadroom could have slightly negativelybiased the signals recorded with the FBRsystem on due to increased equivalent-inputnoise of the hearing aid at the time ofrecording. It is of note, however, that pairedcomparisons did not reach statistical signif-icance even with this potential bias.

While not the focus of this investigation,the observed differences in preferenceacross the different hearing aids in FBR-offversus FBR-off and FBR-on versus FBR-onrecordings were of note. These differenceswere not expected, particularly given thematched gain settings and the fact that thetwo tested hearing aids utilized the samemethod of long-term feedback reduction.These differences between aids, however,cannot be solely attributed to the effects ofthe FBR systems as other differences ineach hearing aid’s unique digital signal pro-cessing were still present, and these differ-ences existed with the FBR system deacti-vated. Other aspects of the hearing aids’signal processing were present, includingcompression time constants, differences incompression implementation, and otherunknown differences in signal processing,which may have altered moment-to-moment gain in the amplitude varyingspeech and music signals. Equivalent-inputnoise was not different between the twoaids by more than a few tenths of a dB and


414

FFiigguurree 66.. Mean strength of preference rating difference in FBR-off versus FBR-off and FBR-on versusFBR-on conditions between the two test hearing aids across all subjects.

was not higher for the nonpreferred aid;thus, it is not the expected cause of sound-quality differences between aids either.Finally, the sound-quality differencesbetween aids may not be present whenthese products are fit clinically to individ-ual specific prescriptive targets as thesesound-quality results were from recordingsthat had been shaped to the NAL-NL1 pre-scriptive targets for the participant withthe worst hearing thresholds. Collectively,these issues may explain a part of theobserved differences across aids and com-plicate the exact causes of differencesbetween HA1 and HA2. These differencesare not considered important given thefocus of this study but are included anywayto highlight the sensitivity of the methodol-ogy used for identification of differences insound quality. Given the sensitivity of thepaired comparison technique used, thesedata are viewed as strong evidence that theFBR processing used in the two hearingaids evaluated does not effect sound quality.

SSUUMMMMAARRYY

Within the methods employed to evalu-ate the two hearing aid FBR systems,

no significant effect on sound quality wasobserved as a result of activating the FBRsystem in either device for short segmentsof a speech and music signal. This was eventhe case in the presence of 13.1 dB and15.75 dB of additional gain for both themusic and speech signal, respectively, withHA1. The method of paired comparisonsutilizing a strength of preference rating todemonstrate significant sound-quality dif-ferences in the experiment was showneffective as differences were consistentlyshown in comparisons between HA1 andHA2. Thus, clinically, the benefit of addi-tional gain afforded to listeners with hear-ing impairments by the continuous phasecancellation FBR systems examined is notoffset by degraded sound quality.


415

RREEFFEERREENNCCEESS

Agnew J. (1996) Acoustic feedback and other audibleartifacts in hearing aids. Trends Amplif 1:45–82.

Byrne D, Dillon H, Ching T, Katsch R, Keidser G.(2001) NAL-NL1 procedure for fitting nonlinearhearing aids: characteristics and comparisons withother procedures. J Am Acad Audiol 12:37–51.

Chung K. (2005) Challenges and recent develop-ments in hearing aids: part II. Feedback and theocclusion effect reduction strategies, laser shellmanufacturing processes and other signal process-ing technologies. Trends Amplif 8:125–164.

Cox RM, McDaniel DM. (1989) Development of theSpeech Intelligibility Rating (SIR) test for hearingaid comparisons. J Speech Hear Res 32:347–352.

Dillon H. (2001) Hearing Aids. Turramurra,Australia: Boomerang Press.

Edwards AL. (1957) Techniques of Attitude ScaleConstruction. New York: Appleton-Century-Crofts.

Eisenberg LS, Dirks DD, Gornbein JA. (1997)Subjective judgments of speech clarity measured bypaired comparisons and category rating. Ear Hear18:294–306.

Greenberg JM, Zurek PM, Brantley M. (2000)Evaluation of feedback-reduction algorithms forhearing aids. J Acoust Soc Am 108:2366–2376.

Hellgren J, Lunner T, Arlinger S. (1999) Systemidentification in hearing aids. J Acoust Soc Am105:3481–3496.

Howell DC. (2002) Statistical Methods forPsychology. 5th ed. Belmont, CA: Duxbury Press.

Kates JM. (1999) Constrained adaptation for feed-back cancellation in hearing aids. J Acoust Soc Am106:1010–1019.

Kiedser G, Dillion H, Byrne D. (1995) Candidates formultiple frequency response characteristics. EarHear 16:562–574.

Kochkin S. (1997) Subjective measures of satisfac-tion and benefit: establishing norms. Semin Hear18:37–48.

Kochkin S. (2003) MarkeTrak VI: isolating theimpact of the volume control on customer satisfac-tion. Hear Rev 10:26–35.

Kuk F. (1994) Maximum usable real-ear insertiongain with ten earmold designs. J Am Acad Audiol5:44–51.

Maxwell JA, Zurek PM. (1995) Reducing acousticfeedback in hearing aids. IEEE Trans Speech AudioProc 3:304–313.

Punch JL, Parker CA. (1981) Pairwise listener pref-erences in hearing aid evaluation. J Speech HearRes 24:366–374.

Ricketts TA, Hornsby BWY. (2005) Sound qualitymeasures for speech in noise through a commercialhearing aid implementing “digital noise reduction.”J Am Acad Audiol 16:270–277.

Speaks C. (1992) Introduction to Sound: Acousticsfor the Hearing and Speech Sciences. San Diego:Singular Publishing Group.

Thurstone LL. (1927a) A law of comparative judg-ment. Psychol Rev 34:273–286.

Thurstone LL. (1927b) Psychophysical analysis. AmJ Psychol 38:369–389.

Widrow B, McCool JM, Larimore JM, Johnson CR.(1976) Stationary and nonstationary learning char-acteristics of the LMS adaptive filter. Proc IEEE64:1151–1162.

Wimmer V. (1986) The occlusion effect from ear-molds. Hear Instrum 37:19, 56.


416

Johnson et al: The Effect of Digital Phase Cancellation Feedback ...€¦ · the feedback pathway...

Documents

Transcript of Johnson et al: The Effect of Digital Phase Cancellation Feedback ...€¦ · the feedback pathway...