CHAPTER NINE

Introduction

The sophistication of hearing aid (HA) signal pro-cessing has increased rapidly in the last decade. Severalspecific advancements have been developed with thegoal of mitigating the negative perceptual and psycho-logical consequences of background noise for HA users.While attempts to reduce unwanted noise for HA usershave been made over the last three decades (see Bentlerand Chiou 2006 for a review), modern digital noise re-duction (DNR) algorithms attempt to limit backgroundnoise by initially classifying the input to the HA based onthe acoustic characteristics of the listening environ-ment. When noise is the primary signal detected by theHA, DNR algorithms reduce gain to improve listenercomfort. Ideally, DNR should maintain audibility for thespeech signal if speech and noise are both present in theenvironment. The sophistication of current DNR algo-rithms allows for these changes to be made independ-ently across multiple frequency bands simultaneously.Optimization across frequency bands minimizes distor-tion of the speech signal by only providing DNR in thefrequency bands where noise dominates the input signal(Hoetink , Korossy and Dreschler 2009). Although thesegoals may sound simple in theory, overlap between thefrequency spectra of speech and noise in realistic envi-ronments (Koopman, Franck and Dreschler 2001) hasthe potential to limit the algorithm from achieving these

goals. Specifically, the expected improvements in thesignal-to-noise ratio (SNR) and speech perception innoise may not be realized outside of laboratory settings.

Because DNR has been made a standard featureavailable in most digital hearing aids within the lastdecade, scrutiny regarding the efficacy and appropriate-ness of such algorithms for children who wear hearingaids has surfaced. Studies of DNR with adult listenershave begun to illustrate the potential advantages of DNRin terms of listener comfort in noise, showing that im-proved comfort can be achieved without negatively im-pacting speech understanding. Unfortunately, researchevaluating the efficacy of DNR for pediatric hearing aidusers has been limited. Lack of substantive evidence tosupport the use of DNR with children has led to recom-mendations that DNR be implemented with caution(Palmer and Grimes 2005). Clinicians must determine ifDNR should be activated for their pediatric hearing aidclients despite limited evidence to support their decisionto use this widely available feature. If an audiologist de-termines that DNR may be appropriate for a given child,it is important that the effects of the processing on thespeech signal be included in the verification process.Just as verification is important for other advanced sig-nal processing strategies, it is necessary with DNR soany potential changes in the speech signal be identifiedand optimal settings obtained.

The purpose of this article is to provide a brief re-view of how DNR is implemented in HAs. Previous stud-ies with adult listeners are highlighted in addition to amore detailed discussion regarding outcomes of the lim-ited number of recent DNR studies completed with chil-dren. A practical verification procedure for clinicians toevaluate DNR with their pediatric clients is presented.Strategies for optimizing DNR algorithms to minimizethe impact on speech audibility are discussed, as well as

153

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings?

Ryan W. McCreery, Samantha Gustafson and Patricia G. Stelmachowicz

Address correspondence to: Ryan W. McCreery, M.S., CCC-A,Boys Town National Research Hospital, 555 North 30th Street, Omaha, NE 68131, Email: ryan.mccreery@boystown.org.

A Sound Foundation Through Early Amplification154

considerations for future studies evaluating the efficacyand effectiveness of DNR for children.

Digital Noise Reduction in Hearing Aids

Noise reduction has long been a goal of HA manu-facturers due to the negative consequences of noise forboth listener comfort and speech understanding. Earlyattempts to reduce background noise in analog deviceswere based on limiting low-frequency gain at high inputlevels through either amplitude compression or adap-tive filtering. Improvements in speech recognition withthese systems were typically not realized since de-creased gain at low frequencies also decreased the audi-bility of the speech signal (Tyler and Kuk 1989). Withthe development of digital HA signal processing, the so-phistication of DNR algorithms has increased signifi-cantly; however, the goal of reducing interference fromnoise has remained unchanged. Many modern DNR al-gorithms use some form of amplitude modulation detec-tion to analyze the input to the HA (see Kates 2008 for areview). The differences in amplitude modulation be-tween speech and noise spectra allow the HA to estimatewhether the input is comprised of speech or noise andthen to adjust the amount of gain provided by the HA.Devices with multiple channels of signal processing canperform these analyses for discrete frequency regionsand reduce gain in only the channels where noise domi-nates the input. Such precision allows algorithms to pro-vide gain reduction in the frequency regions where thenoise is present, while attempting to maintain audibilityfor speech. The analysis of the signal characteristics bythe HA is an on-going process, allowing the algorithmsto adapt to changes in the acoustic environment.

In order to improve speech understanding in noisesignificantly through signal processing, algorithms wouldneed to effectively improve the SNR. DNR operates by re-ducing gain, which means that any improvements in SNRwould only be achieved by maintaining the amount of gainfor speech, while reducing gain for noise. Because of thesignificant spectral overlap between speech and noise ineveryday listening situations, current DNR algorithmsmust analyze and manipulate a signal that contains bothspeech and noise. Any changes in gain made by the algo-rithm are applied to the combined noisy speech signal. In-dependent manipulation of the speech and noise signalwould be required to change the SNR in each band; there-fore, it is unclear if significant improvement of the overallSNR can be achieved with current DNR technology un-less the noise is isolated to a limited frequency region.

Methods for quantifying the change in SNR improvementwith DNR processing have been proposed (Hagermanand Olofson 2004), but have not been extensively evalu-ated in formal studies of DNR.

Although many DNR systems use modulation detec-tion as the primary method of signal analysis, the imple-mentation of DNR algorithms varies substantially acrossmanufacturers and devices. Hoetink and colleagues(2009) systematically evaluated the characteristics ofDNR in twelve different HAs to quantify variations inhow these algorithms are implemented in different de-vices. Results from measurements using modulated andunmodulated ICRA (International Collegium for Reha-bilitative Audiology; Dreschler, Verschuure, Ludvigsenand Westermann 2001) noise stimuli indicated that DNRsystems in hearing aids vary based on a number of dif-ferent parameters including the amount of gain reduc-tion, number of channels where DNR is active, fre-quency range where noise reduction occurs, input levelwhere DNR is activated, and the wearer’s audiometricthresholds. Results from a comparison of the two typesof ICRA noise revealed that different DNR algorithmscan have distinct effects on the amount of gain forspeech signals in noise, varying from algorithms thatpreserve the speech signal with limited gain reductionto those which significantly reduce gain. The amount oftime required for onset and release of DNR algorithmsalso varies considerably across manufacturers. DNRsystems with faster attack and release times have beenfound to improve speech understanding and listener rat-ings of sound quality in real world environments(Woods, Nooraei, Galster and Edwards 2010). Presum-ably, algorithms with faster time constants have the abil-ity to adapt more rapidly to acoustic changes in speechor noise in the environment, which may result in a morerapid gain reduction with the onset of noise and less per-sistence of noise reduction with the onset of speech.

Because the characteristics of each manufacturer’sDNR systems are proprietary, variations in these char-acteristics may not be evident to clinicians without ex-tensive and systematic testing. These differences havethe potential to significantly impact the amount of gainthat is applied in different situations, directly impactingthe audibility of speech for the HA user. The wide rangeof implementations and proprietary nature of these algo-rithms support the need for an individualized verifica-tion approach to determine how DNR may impact the au-dibility of speech. However, the assessment of variousparameters of DNR for each manufacturer presents mul-tiple challenges for clinicians.

Studies of DNR with Adult Listeners

A growing body of literature assessing the effects ofDNR algorithms with adult listeners has started toemerge over the past 10 years because the availability ofthese algorithms in digital hearing aids has becomestandard. Studies of DNR with adult listeners have fo-cused on three main categories of outcome variables:speech recognition, listener ratings of comfort or pref-erence in noise, and ease of listening or listening effort.Although improvements in speech understanding innoise is a desirable outcome for individuals with hearingloss, the extant literature suggests that DNR algorithmsdo not significantly improve or degrade speech under-standing in noise for adult listeners (Alcantara, Moore,Kuhnel and Launer 2003; Boymans and Dreschler 2000;Ricketts and Hornsby 2005; Hu and Loizou 2007). Giventhat most approaches of DNR either decrease or main-tain amplification based on the presence, intensity andspectral characteristics of noise, the best potential out-come of this processing would be to maintain the audi-bility of the speech signal. With reduction in gain, thereis also potential to reduce overall audibility. Therefore,the limited improvements in speech understanding innoise that have been reported with DNR to date may notbe surprising.

Although DNR has not been demonstrated to de-crease speech understanding in noise, the positive im-pact of DNR on listener comfort in background noise, aswell as measures of ease of listening in adults have beenmuch more widely reported. Mueller, Weber andHornsby (2006) determined that adults with hearing lossdemonstrated increased Acceptable Noise Level (ANL)with DNR enabled. In previous studies with hearing-im-paired adults, the ability of listeners to tolerate a higherANL has been linked with increased likelihood of hear-ing aid use (Nábĕlek, Freyaldenhoven, Tampas, Burch-field and Muenchen 2006). In another study of DNR withadults, Ricketts and Hornsby (2005) used a paired com-parison task to measure listener preference for speech innoise at two different signal-to-noise ratios (SNRs).While results did not indicate an improvement in speechunderstanding for their participants, listeners demon-strated a very strong preference for speech in noiseprocessed using DNR over those signals without DNR.

Given the fact that listening in background noise isa significant complaint that has been reported for manyhearing aid users, improvements in listener preferenceor comfort in noise are encouraging, particularly if thealgorithms do not degrade speech understanding. More

recent studies have attempted to demonstrate the im-pact of DNR for reducing listening effort in adults. Suchstudies are based on the theory that listening in back-ground noise requires the listener to dedicate more cog-nitive processing resources to decoding the speech sig-nal relative to listening in quiet. The increased effort re-quired for listening in noise leaves the listener withfewer remaining cognitive resources to perform otherimportant processes, such as committing stimuli toshort-term memory. Therefore, interference from noisehas been demonstrated to result in decreased recall,even at SNRs where recognition remains intact (Conlin,Gathercole and Adams 2005). The method that is mostoften used to measure a listener’s allocation of cognitiveresources is a dual-task paradigm, where a listener is in-structed to perform two tasks. When the primary taskincreases in difficulty, performance on the secondarytask decreases as individuals allocate more cognitive re-sources to performing the primary task. Sarampalis andcolleagues (Sarampalis, Kalluri, Edwards and Hafter2009) used two different dual-task paradigms to meas-ure the influence of DNR on short-term memory and lis-tening effort. In one experiment, adult listeners were re-quired to repeat words or sentences as the primary taskand hold the words in memory for later recall. Addition-ally, speed of processing was estimated in another taskby measuring responses on a visual task completedwhile performing speech recognition in noise. DNR didnot have a significant effect on the speech recognition;however, free word recall and reaction time for the sec-ondary task improved at the lowest signal-to-noise ratioin each task when DNR was applied to the auditory stim-uli. Although the statistical effect sizes for each taskwere relatively small, these results suggest that positivebenefits of DNR may extend beyond the traditional out-come measure of improved speech recognition.

Evidence from DNR research with adults has impor-tant implications for clinical decisions regarding the ap-plication of this technology in hearing aid fittings. Al-though DNR does not affect speech understanding, bene-fits in terms of listener comfort, preference and ease oflistening have been demonstrated. Few studies of DNRwith adults have attempted to quantify the effects ofDNR on the speech and noise signals in an attempt topredict which algorithms have the greatest potential forperceptual improvement. The relationship betweenease of listening and speech recognition is also not wellunderstood. It is possible that improvements in listenercomfort or ease of listening may have the potential tooffset changes in audibility through reduction in gain

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 155

that may occur with DNR. This trade-off could result ina maintained level of speech understanding with muchless effort on the part of the listener. The relationshipbetween listening effort and speech understanding mayhave significant implications for using DNR with chil-dren, but requires additional investigation.

Despite the emergence of promising evidence tosupport the use of DNR with adults, it is important tonote that DNR signal processing strategies continue tobecome more sophisticated. Because the pace of innova-tion often exceeds the rate at which studies of these al-gorithms can be designed, completed and published, itwill be difficult for practicing clinicians to evaluate the ef-ficacy of these algorithms. Overall, DNR studies withadults have suggested that, while DNR algorithms do notsignificantly alter speech understanding, adult listenersdo report improvements in listening comfort, toleranceof background noise, and reduced listening effort withDNR. While studies with adult listeners are important forour understanding of how these algorithms may benefitadult hearing aid users, application of these results to apediatric population is tenuous because the listeningneeds of children are distinct from those of adults.

DNR for Children who Wear Hearing Aids

Data from adult studies related to DNR would sug-gest that such processing might be beneficial to chil-dren at least in terms of improving listening comfort innoise. Unfortunately, previous studies of children withhearing loss have demonstrated the limitations of at-tempting to predict performance of children based onadult behavioral data. Therefore, the potential benefitsand negative consequences of DNR for children must becarefully considered using data from studies with pedi-atric participants. Children with hearing loss requireamplification to promote development of speech and lan-guage; therefore, the impact of any signal processingstrategy on the audibility of speech must be consideredprior to implementation in pediatric hearing aid fittings.Because children require more audibility than adults toachieve levels of speech recognition similar to those ofadults (Stelmachowicz, Hoover, Lewis, Kortekaas andPittman 2000), any reduction in audibility for the speechsignal occurring as a consequence of an attempt to re-duce background noise could have a greater negativeimpact on children compared to adult listeners. Dataalso suggest that children are frequently listening inclassrooms and other environments where backgroundnoise is present at levels that could negatively impact

perception (Bradley and Sato, 2008). Neuman and col-leagues (Neuman, Wroblewski, Hajicek and Rubinstein2010) found that children show greater degradation ofspeech understanding in background noise and rever-beration than adults. Thus, the potential to reduce thenegative perceptual consequences of background noiseusing DNR may be even greater for children than foradult listeners.

While modern DNR algorithms are designed tolimit the amount of gain reduction when speech is pres-ent in the environment, reduced audibility is still a po-tential negative outcome. For example, Figure 1 showsthe reduction in gain for speech in noise when DNR isactivated in two devices that utilize spectral subtractionDNR. While there is no change in the amount of gain forthe HA in Panel A, the HA in Panel B shows a broadbandreduction in gain of approximately 6 dB when speechand noise are presented to the HA with DNR activated.Such a significant reduction in gain would limit the au-dibility for speech in noise, potentially limiting speechunderstanding in noisy situations. Unfortunately, thedifferences between the outputs of these two DNR algo-rithms for speech in background noise would not be ap-parent using most pediatric hearing aid verification pro-tocols for DNR.

To date, relatively few studies have evaluated the ef-fects of DNR on speech understanding or ease of listen-ing on children. Marcoux and colleagues (Marcoux,Yathiraj, Cote and Logan 2006) utilized a cross-languageparadigm where adult listeners with normal hearingwere exposed to a novel, non-native speech contrast thatis sufficiently difficult for adult English speakers to per-ceive. This method is frequently used in studies ofspeech sound acquisition. In their study, speech wasprocessed through a commercially-available hearingaid. The goal of the study was to determine if DNRwould interfere with the ability of listeners to perceiveand learn acoustic contrasts needed for speech and lan-guage development, which is a primary concern for anyhearing aid signal processing strategy that has the po-tential to be implemented with children. Results afterfour sessions of exposure to the contrast suggested thatadult subjects were able to learn the speech contrast re-gardless of whether DNR processing was utilized, sug-gesting that DNR did not interfere with the perceptionand acquisition of a novel phonetic cue. The authors citea number of specific limitations in generalizing thesefindings to pediatric hearing aid users, including mostimportantly the use of adult listeners. Although thesefindings may not predict specific outcomes in children

A Sound Foundation Through Early Amplification156

with hearing loss, the implementation of DNR used byMarcoux and colleagues did preserve the acoustic cuesin the speech signal sufficiently to allow listeners tolearn a novel speech contrast, suggesting that DNR canbe applied without negatively impacting cues importantfor learning phonetic contrasts.

One of the first attempts to evaluate DNR process-ing for school-age children with hearing loss reported inthe literature was completed by Auriemmo and col-leagues (2009) as part of a study evaluating multiple ad-vanced hearing aid features. Multiple dependent meas-ures were used to quantify the influence of DNR on lis-tening behavior in children, including speech recogni-tion, as well as subjective ratings of everyday listeningsituations by parents and the child HA users. Results re-vealed no differences in speech recognition betweenconditions with DNR and without DNR, consistent withprevious adult studies. Differences in subjective ratingsfor DNR were found only for questions related to soundsoriginating from behind the listener, which childrenrated as less bothersome with DNR. The main strengthsof this study were the use of both laboratory speech per-ception measures and children’s subjective ratings fromextended use in realistic situations. Additionally, be-cause the effectiveness of DNR is not dependent on spa-tial separation between the signal and noise as it is for

directional microphones, DNR may provide additionaladvantages for children in background noise beyondwhat could be realized with a directional microphonealone. Auriemmo and colleagues reported no additionalimprovement in speech recognition with DNR com-pared to the directional microphone condition in theirstudy. However, the laboratory speech recognitionmeasures used in this study did not include acousticconditions that would be expected to provide an advan-tage over a directional microphone, such as spatially dif-fuse noise or speech signals originating from the sidesor behind the listener. These conditions have beenshown to limit the advantage of directional microphonesin previous studies with school-age children (Ricketts,Galster and Tharpe 2007), but the relative advantage ofDNR under these conditions has not been investigated.

More recently, additional evidence to support theuse of DNR with school-age children was reported byStelmachowicz and colleagues (2010). Children withmild-to-moderate hearing loss between 5 and 10 years ofage were asked to repeat nonsense vowel-consonant-vowel (VCV) syllables, monosyllables, and sentences atthree different SNRs. The speech stimuli wereprocessed using a commercially-available HA with amodified spectral subtraction DNR algorithm imple-mented over sixteen frequency bands. HA gain and out-

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 157

Figure 1. Hearing aid output level (dB SPL) in a 2cc coupler for a 60 dB speech signal in steady-state noise at a +3 dB SNR plotted as a functionof frequency (Hz) for two hearing aids. The solid line in each panel is the hearing aid response with DNR off, while the dashed line is the responsewith DNR on. The difference between the two responses for HA 1 (left panel) represents a reduction in gain for speech in noise with DNR on.Changes in gain with DNR on are minimal for HA 2 (right panel).

3

put characteristics were optimized using Desired Sensa-tion Level (DSL) targets for average speech for eachchild’s audiogram. On average, speech recognition innoise did not change when DNR was activated for any ofthe stimulus types or SNRs. Despite the fact that therewas no observed effect of DNR overall, significant indi-vidual variability in speech recognition was observed,particularly for children in the 5 to 7 year-old age range.Although speech recognition was highly variable acrosssubjects, none of the children in the study showed a con-sistent improvement or degradation in speech under-standing across conditions when DNR was activated.

Another study of DNR with children conducted byBentler, Kirby and Stiles (2010) attempted to assess thepotential impact of DNR on speech recognition, novelword learning, and subjective ratings of sound quality innormal hearing school-age children. While the resultsfor speech recognition demonstrated no effect of theDNR processing, novel word learning rates and chil-dren’s subjective ratings of sound quality of speech innoise were both improved in the DNR conditions com-pared to the conditions without DNR. Results from thisinvestigation support the previous studies’ findings oflimited impact on speech understanding with the addi-tional benefits of improved sound quality ratings andword learning rates in children. The researchers specu-lated that the improvements observed in novel wordlearning rates with DNR were related to a decrease inthe cognitive processing load afforded by the DNR sig-nal processing.

Although these preliminary DNR findings supportthe implementation of this feature with school-age chil-dren, the limitations of these studies should be con-sidered when determining the appropriateness of DNR forpediatric hearing aid users. Specifically, despite the factthat DNR is implemented differently across manufactur-ers, the studies by Auriemmo, et al. (2009) and Stelma-chowicz, et al. (2010) each evaluated only a single DNRalgorithm, limiting the potential generalization of thesefindings to other DNR algorithms. Additionally, compar-ative data for different algorithms on the same task to as-sess various aspects of DNR algorithms and how theymay impact that algorithm’s effectiveness with childrenare not possible when only a single algorithm is as-sessed. Although studies by both Auriemmo, et al. andBentler, et al. (2010) included children’s ratings of com-fort, additional listener ratings of comfort and soundquality in children have not been widely reported in pe-diatric studies of DNR. In both cases, improvements inchildren’s ratings of sound quality occurred without im-

provements in speech understanding, suggesting thatchildren may be sensitive to improvements in signalquality that are not significant enough to improve per-ception.

The ability of researchers to generate informed hy-potheses about how children perform with DNR hasbeen hindered by the difficulty quantifying the effect ofthese algorithms on the speech signal. As discussed pre-viously, improvements in SNR with DNR are a desiredoutcome that would be required in order to improvespeech recognition. Quantifying specific changes in au-dibility related to DNR would allow for hypothesesabout the impact of these algorithms on speech under-standing. For example, an algorithm that reduced thesignal quality above 3000 Hz could have a negative im-pact on fricative perception. In order to determine ifDNR improves the SNR after the signal has beenprocessed by the hearing aid, the speech and noise sig-nals would have to be evaluated separately, as the com-bined speech and noise output may not reflect alter-ations of important speech cues that may have occurredduring signal processing. Analysis of the effects of DNRon the speech and noise signals independently may al-low estimation of any changes in SNR that might not beobservable from the combined speech and noise signal.

In an attempt to address some of the limitationsfrom previous studies of DNR and children, Gustafsonand colleagues (Gustafson, McCreery, Hoover, Kopunand Stelmachowicz in preparation) conducted an inves-tigation of the impact of two DNR algorithms on speechrecognition, listening effort, and ratings of sound clarityfor a group of 7 to 12 year-old children with normal hear-ing. Multiple candidate DNR algorithms in commer-cially available devices from different manufacturerswere considered for the study. Based on their signal pro-cessing characteristics, two devices were selected. Onemajor limitation of previous DNR studies is the lack ofattempts to quantify the impact of the processing on thespeech signal. Hagerman and Olofson (2004) developeda procedure known as the inversion method to quantifythe effects of digital signal processing strategies such asDNR on the integrity of the speech signal when meas-ured in noise. The inversion method requires two sam-ples of speech in background noise that are identical ex-cept that the phase of the noise is inverted in one sam-ple. When the two samples are added together, the noisesignals cancel and the resulting waveform contains onlythe speech signal. When the two samples are sub-tracted, the speech signals cancel, leaving only the noisesignal. From these two extracted signals, the effects of

A Sound Foundation Through Early Amplification158

signal processing on both the speech and noise signalscan be estimated, including changes in the SNR bymeasuring relative changes in the level of the two sig-nals. The inversion method has also been previouslyused to evaluate the effects of amplitude compressionon the relative levels of speech and noise (Souza, Jen-stad and Boike 2006). The two algorithms selected byGustafson and colleagues had different effects on theSNR following processing with DNR. As measured bythe inversion method, HA1 had minimal impact on theSNR while HA2 resulted in an improvement in the SNRof 7 dB. The hypotheses of the study were that the HAproviding the largest SNR improvement would show thegreatest improvement in speech recognition, reductionin verbal response time, and improvements in children’sratings of clarity.

Another significant limitation of previous studieswas the use of either closed-set stimuli or stimuli withhigh linguistic redundancy when estimating perform-ance changes with DNR. While providing stimuli withlinguistic context can help to support children in diffi-cult listening environments and provides a valid estima-tion of what children are able to do in realistic situationswith contextual cues, the availability of context could po-tentially overshadow the effects related to subtlechanges in signal processing. To limit the available se-mantic cues during the speech recognition task, non-word consonant-vowel-consonant (CVC) stimuli wereutilized to require children to rely more heavily on theacoustic-phonetic representation and use bottom-up de-coding skills rather than top-down linguistic knowledge.Children must be able to perceive these acoustic-phoneticcues as they develop speech and language.

Results from the study showed a slightly differentpattern than was predicted from the inversion method.Although HA2 resulted in a 7 dB improvement in SNR,phoneme recognition improved by only 6.5% comparedto HA1. While this difference in performance was statis-tically significant, previous studies of phoneme recogni-tion for normal-hearing children in this age range havedemonstrated approximately a 20% improvement inphoneme recognition with a 7 dB improvement in SNR(McCreery, Spratford, Lewis, Hoover and Stelmachow-icz 2010). Additionally, verbal response time and ratingsof sound clarity were improved with both algorithms,even without a significant improvement in the SNR inHA1 as predicted by the inversion method. In an effortto explain the differences in speech recognition be-tween the two algorithms, further analysis of the ex-tracted speech signal from the inversion method was

completed using magnitude squared coherence. Coher-ence compares the spectra of two signals to determinethe degree of similarity as a function of frequency andvaries between one, suggesting that the signals are spec-trally identical, and zero, suggesting that the two spectraare completely dissimilar. Coherence measures havepreviously been used to evaluate equivalent input noisein hearing aids (Lewis, Goodman and Bentler 2010). Co-herence measures comparing the extracted speech sig-nals with DNR active versus inactive for HA1 and HA2revealed that the high-frequency region of the speechspectrum was better preserved by HA2 when DNR wasactivated. Decreased perception of phonemes with en-ergy in the high frequencies such as fricatives would beanticipated for DNR processors that did not maintain thelevel of the speech spectrum above 3 kHz. Future stud-ies could potentially use coherence measures to predictperceptual performance for algorithms that significantlyalter the speech spectrum.

Overall, preliminary research regarding the efficacyof DNR signal processing for children who wear hearingaids has begun to emerge. Multiple studies have sup-ported findings from the adult literature that DNR doesnot appear to have an impact on school-age children’sspeech understanding in noise. Additionally, a few stud-ies have also demonstrated that children’s ratings of clar-ity and sound quality of background noise are also im-proved with DNR. Studies have suggested that DNR mayalso improve novel-word learning (Bentler et al. 2010)and verbal response time (Gustafson et al. in prepara-tion). Both of these outcome variables have been attrib-uted to decreased cognitive processing load for childrenlistening in background noise. Furthermore, some re-searchers have speculated that the presence of back-ground noise can interfere with the rehearsal of auditorystimuli (Sarampalis et al. 2009), an important part of theprocess for committing auditory stimuli to memory.

Despite significant progress evaluating DNR strate-gies for children over the past few years, a significantnumber of questions regarding the efficacy of this tech-nology with children remain unresolved. Currentlyavailable studies of DNR with children have included ei-ther children with normal hearing or children with mild-to-moderate hearing losses as participants. Furtherstudies should be completed for children with greaterdegrees of loss, as well as evaluating a wider range ofvariables in children with all degrees of hearing loss.Children with severe-to-profound hearing loss may be atthe greatest risk for experiencing reduced audibilitywith DNR strategies simply because they have more lim-

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 159

ited audibility based on their available dynamic range.Furthermore, most of the studies of DNR with childrenhave been completed in the school-age population. Al-though our knowledge of DNR in this age range is im-portant, studies with infants and younger childrenwould provide clinicians with the evidence and confi-dence to utilize this technology at an earlier age. How-ever, the current speech recognition and word learningparadigms would likely have to be adapted significantly,or alternative tasks would have to be developed beforestudies of DNR could be completed in younger children.To date, studies evaluating how experience and trainingwith DNR signal processing could influence children’sperformance with this technology have also not been es-tablished. Finally, additional studies validating the inver-sion method and coherence measures for evaluating thepotential impact of DNR on the speech signal and audi-bility are needed before firm conclusions can be madefrom these measurements.

Verification of DNR

The primary purpose of providing amplification toinfants and young children is to improve outcomes forspeech and language development and provide childrenwith the access to their environment. Maintaining audi-bility of the speech signal and controlling the level ofmore intense sounds to prevent discomfort both pro-vide an important foundation for meeting these goals.Therefore, verification procedures for pediatric hearingaid fittings should at least demonstrate that speech is au-dible and that loudness is adequately controlled to pre-vent discomfort. With the development of advancedacoustic signal processing features such as DNR, theneed to verify the impact of these technologies on audi-bility has become a critical part of the process. Any ad-vanced feature that may affect the audibility of speechrequires verification to quantify the frequency regionswhere changes in audibility occur and the degree towhich audibility is altered. As previously stated, the pa-rameters of DNR algorithms vary considerably bothacross product lines within the same manufacturer andacross different hearing aid companies. The proprietarynature of signal processing strategies means that clini-cians are often unable to predict the impact of DNR onaudibility. Additionally, some manufacturer’s DNR sys-tems vary in the amount of gain reduction provided bythe algorithm based on the client’s audiogram used toprogram the hearing aid, providing less reduction ingain as the degree of hearing loss increases in an at-

tempt to preserve audibility (Hoetink et al. 2009). Thecombination of how DNR processing varies due to indi-vidual differences between clients and manufacturersholds clinicians responsible for verifying these featureswhen implemented with a child.

Three different methods of verification for DNR areavailable to clinicians based on the available verificationequipment. Verification of DNR can occur either using aprobe microphone to measure the output of the hearingaid in the client’s ear or in a 2cc coupler. Although probemicrophone measures are recommended whenever pos-sible, performing the additional steps required for DNRverification using probe microphone measurements islikely to be difficult in infants and young children. There-fore, the most practical method of DNR verification is of-ten completed in a 2cc coupler. If the results of DNR veri-fication are to be compared to the client’s thresholds, areal-ear-to-coupler-difference (RECD) measurementshould be applied to improve estimates of audibility. Oneadvantage to performing DNR verification in a 2cc cou-pler is that the verification can occur prior to the fittingappointment, leaving more time to provide informationalcounseling to children and their parents. For the pur-pose of the current article, the verification procedureswill be presented as if the process takes place in a 2cccoupler, keeping in mind that these same measures canbe completed using real-ear probe microphone meas-ures in cooperative clients.

Three main types of test stimuli can be used to eval-uate the effects of DNR: 1) steady-state broadband noiseor speech-shaped noise, 2) recordings of environmentalsteady-state noise signals, and 3) speech signals withsteady-state background noise. Test signals must be cali-brated to provide a controlled estimate of how DNRfunctions at realistic input levels. The availability ofthese signals will depend on the verification equipmentavailable. Although all three types of test signals can beused for verification of DNR in hearing aids, the informa-tion that can be inferred from each test signal is notequivalent. Steady-state broadband and speech-shapednoises, in addition to recordings of steady-state environ-mental signals, can be used to provide information aboutthe amount of gain reduction provided by the DNR algo-rithm, frequencies where gain reduction occurs, andDNR time constants. However, only speech signals withsteady-state background noise can provide an estimateof how the DNR algorithm will respond when speechand noise are present simultaneously. The amount ofgain reduction observed for steady-state noise or record-ings of environmental noise should not be assumed to

A Sound Foundation Through Early Amplification160

predict the amount of audibility that would be lost forspeech. Rather, using steady-state noise signals pro-vides clinicians with an estimate of the maximum effectof DNR that may occur when only noise is present in theenvironment.

If steady-state noise test signals are to be used forthe DNR verification process, the HA should be at-tached to the 2cc coupler and placed in the test chamber.The HA should be attached to the programming soft-ware to allow selection of features during the verifica-tion process. Prior to evaluating DNR, directional micro-phones should be set to a fixed omnidirectional settingin the programming software, ensuring that interactionsbetween DNR and directional microphone processingdo not affect the results. First, the HA response to thesteady-state signal without DNR activated should be ob-tained, if possible. Several models of HAs do not allowDNR to be deactivated for verification. In these cases,the default DNR setting can be used as a baseline. The

steady-state noise signal should be presented to thehearing aid at a level that would be consistent with envi-ronmental noise levels experienced by the client, be-tween 55 – 75 dB SPL. Measurements at multiple inputlevels can help clinicians to determine if the DNR algo-rithm has level dependency (i.e., different characteris-tics for different input levels). Once a baseline measureof the HA response without DNR is obtained, the pro-gramming software should be used to activate the DNRsetting that will be used with the client. The steady-statenoise is then presented to the HA in the coupler again tomeasure the device’s response with DNR activated. Forboth measures, the signal should be presented to theHA for a minimum of 30 seconds to allow the DNR algo-rithm to reach its maximum effect and ensure that thealgorithm remains engaged. Comparison of the re-sponse obtained with and without DNR activated will al-low the clinician to make judgments about the impactthat the algorithm may have on audibility when the lis-

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 161

Figure 2. An example of verification of DNR using a steady-state signal using the Audioscan Verifit®. Hearing aid output level (dB SPL) in a 2cccoupler for a 60 dB air conditioner noise plotted as a function of frequency (Hz) with DNR off (Test 1) and DNR on (Test 2). A broadband reductionin gain of approximately 12 dB occurs when a steady-state noise signal is presented to the hearing aid.

3

tener is in background noise. Changes that occur withsteady-state signals should not be assumed to reflect theresponse of the HA when speech is present. Figure 2shows the results of a verification procedure for DNRusing steady-state signals.

In order to estimate the effect of DNR on speech inbackground noise, the clinician must use a verificationprocedure that includes speech and steady-state back-ground noise at calibrated levels. Although not specific-ally developed for DNR verification, the Audioscan Ver-ifit® Directional Test Mode allows the user to presentsteady-state noise and speech simultaneously at an ad-justable SNR in the test box. With the hearing aid at-tached to the 2cc coupler and programmed to an omni-directional microphone setting, sequential measure-ments of the HA response with DNR activated and deac-tivated can be completed with both speech and steady-

state noise presented to the hearing aid. A positive SNRshould be selected for the stimulus presentation to en-sure that the speech is the primary signal in the test box.Ideally, the level and frequency characteristics of the re-sponse from the HA with DNR activated should be verysimilar to the response without DNR, reflecting that au-dibility is preserved when the listener is in an environ-ment with both speech and noise present. Any differ-ences that are observed between the lines correspon-ding with the left and right speakers in the Verifit® testbox are related to the fact that the hearing aid responsediffers for the stimuli being presented from eachspeaker. If the HA is set to a fixed omnidirectional re-sponse, differences between the left and right speakerresponse should not occur. Differences between the leftand right speaker responses with an omnidirectionalsetting may indicate a microphone problem or that the de-

A Sound Foundation Through Early Amplification162

Figure 3. An example of verification of DNR using speech plus steady-state noise using the directional test mode on the Audioscan Verifit®. Hearingaid output level (dB SPL) in a 2cc coupler for a 65 dB speech signal with a 62 dB steady-state noise plotted as a function of frequency (Hz) withDNR off (Test 1) and DNR on (Test 2). The difference between the two responses reflects a reduction in gain when speech and noise are presentedto the hearing aid.

3

vice has adaptively changed into a directional response.Figure 3 plots the verification results for speech andsteady-state noise obtained for two different hearing aids.

Unfortunately, using speech signals in addition tosteady-state noise is not without limitations. The HA re-sponse obtained using this method represents a com-bined input of speech and noise to the hearing aid; there-fore, relative changes between the levels of the speechand noise that could be observed using other measure-ment techniques, such as the inversion method or co-herence, are not apparent to the clinician. This methodalso does not predict how specific DNR algorithms willperform in realistic environments where the amplitudemodulation spectrum of the background noise is moresimilar to speech, such as in background noise com-prised of multiple talkers. Fortunately, most DNR algo-rithms do not activate unless the background noise hasamplitude modulation characteristics that differ fromspeech. However, additional research is needed to un-derstand how these algorithms behave when the as-sumptions about the modulation spectra of backgroundnoise are violated.

Clinical Recommendations

The title of this chapter poses an important clinicalquestion that should be considered by clinicians who fitamplification for infants, children and adolescents. Un-fortunately, the current state of evidence regardingDNR with children and the limitations of our verificationprocedures do not allow for a straightforward answer tothe question. However, the limited body of researchdoes provide some consistent findings that can help cli-nicians to determine the best approach to use with theirpediatric clients. Based on the results from several stud-ies completed with school-age children, DNR does notappear to have a negative impact on speech understand-ing, while providing improved or equivocal ratings ofcomfort and sound quality. Although the evidence tosupport the use of DNR to reduce listening effort is onlyin the most fundamental stages and has only been eval-uated with normal-hearing children, there appears to besome indication that DNR may help to improve ease oflistening, although much more work with children withhearing loss is necessary before firm conclusions can bereached. Based on the current data, DNR would be ap-propriate for school-age children as long as verificationmethods can demonstrate that audibility is not compro-mised when speech and noise are presented to the de-vice simultaneously.

DNR should be applied with caution for youngerchildren and infants, primarily because there have notbeen studies to show that speech recognition is pre-served within this critical time frame for speech and lan-guage acquisition. School-age children, even those withhearing loss, have developed a significant amount ofspeech and language knowledge that can be used to sup-port top-down processing if distortion of the speech sig-nal occurs during processing by the HA. However, in-fants and young children are still developing these foun-dational skills. The decision to implement any signalprocessing feature that could alter this process shouldnot be taken lightly or without strong evidence to sup-port it. Future studies of DNR should attempt to adaptexperimental paradigms to answer some of these criticalquestions for younger groups of children.

Finally, DNR should not be viewed as the only solu-tion that audiologists have at their disposal to reduce thenegative impacts of background noise on children. Theevidence regarding directional microphones and fre-quency-modulation (FM) systems suggests that both ofthose technologies may be appropriate solutions forchildren who wear hearing aids, depending on the lis-tening situation. If background noise is a concern, theaudiologist should consider all potential solutions, char-acteristics of the child’s likely listening situations, aswell as the effectiveness of each of these tools whenmaking decisions about appropriate implementation oftechnology for a specific child. DNR does have limitedadvantages over directional microphones and FM sys-tems in specific situations. Unlike directional micro-phones, DNR does not require spatial separation of thebackground noise and signal of interest in order to be ef-fective. DNR can also be implemented in situations withmultiple talkers of interest, a situation where most cur-rent FM technology would be impractical. Audiologistsshould also consider counseling teachers, parents andchildren about strategies to reduce background noise inclassrooms and in the home, as well as positioning in the en-vironment to maximize audibility of the signal of interest.

Conclusion

Minimizing the negative perceptual and psycho-logical consequences of background noise is an impor-tant goal for audiologists who provide amplification forchildren. DNR is a widely-available signal processingstrategy that adjusts the gain of the hearing aid in an at-tempt to minimize noise and maintain the integrity of thespeech signal. Preliminary research evaluating the effi-

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 163

cacy of DNR for children who wear hearing aids indi-cates that, for school-age children, DNR can improve lis-tener ratings of sound quality and reduce listening ef-fort without negatively affecting speech understanding.If DNR is determined to be appropriate, clinicians mustverify the impact of this signal processing strategy onthe audibility of speech, preferably using a speech signalin steady-state background noise. Additional evidence isneeded to support the use of DNR for younger children,as well as children with severe-to-profound hearing loss,since DNR research has not yet been completed withthese groups.

Acknowledgements

Special thanks to Brenda Hoover for reviewing aprevious version of this manuscript. Research reportedin this chapter was supported by the National Institutefor Deafness and Communication Disorders grants RO1 DC4300 10, T35 DC8757, P30 DC4662, and F31 DC010505.

References

Alcantara, J.L., Moore, B.C., Kuhnel,V., and Launer, S.2003. Evaluation of the noise reduction system in acommercial digital hearing aid. International Journalof Audiology 42(1): 34–42.

Auriemmo, J., Kuk, F., Lau, C., Dornan, B.K., Sweeton,S., Marshall, S., and Stenger, P. 2009. Efficacy of anadaptive directional microphone and a noise reduc-tion system for school-aged children. Journal of Edu-cational Audiology 15: 15–27.

Bentler, B., Kirby, B., and Stiles, D. 2010. Impact of digi-tal noise reduction on pediatric performance. Postersession presented at the annual conference of theAmerican Academy of Audiology, San Diego, CA.

Bentler, R., and Chiou, L.K. 2006. Digital noise reduction:An overview. Trends in Amplification 10(2): 67–82.

Boymans, M., and Dreschler, W.A. 2000. Field trialsusing a digital hearing aid with active noise reduc-tion and dual-microphone directionality. Audiology39(5): 260–8.

Bradley, J.S., and Sato, H. 2008. The intelligibility ofspeech in elementary school classrooms. Journal ofthe Acoustical Society of America 123(4): 2078–86.

Conlin, J.A., Gathercole, S.E., and Adams, J.W. 2005.Children’s working memory: Investigating perfor-mance limitations in complex span tasks. Journal ofExperimental Child Psychology 90(4): 303–17.

Dreschler, W.A., Verschuure, H., Ludvigsen, C., and We-stermann, S. 2001. ICRA noises: Artificial noise sig-nals with speech-like spectral and temporal proper-ties for hearing instrument assessment. Internatio-nal Collegium for Rehabilitative Audiology. Audio-logy 40(3): 148–57.

Gustafson, S., McCreery, R., Hoover, B., Kopun,J.G., and Stelmachowicz, P. in preparation. Easeof listening and speech recognition for normalhearing children with the use of digital noise reduction.

Hagerman, B., and Olofson, A. 2004. A method to mea-sure the effect of noise reduction algorithms usingsimultaneuous speech and noise. Acta Acustica 90:356–361.

Hoetink, A.E., Korossy, L., and Dreschler, W.A. 2009.Classification of steady state gain reduction produ-ced by amplitude modulation based noise reductionin digital hearing aids. International Journal of Au-diology 48(7): 444–55.

Hu, Y., and Loizou, P.C. 2007. A comparative intelligibi-lity study of single-microphone noise reduction algorithms. Journal of the Acoustical Society of America 122(3): 1777.

Kates, J.M. 2008. Digital hearing aids. San Diego, CA:Plural Publishing, Inc.

Koopman, J., Franck, B.A., and Dreschler, W.A. 2001.Toward a representative set of «real-life» noises. Au-diology 40(2): 78–91.

Lewis, J.D., Goodman, S.S., and Bentler, R.A. 2010. Mea-surement of hearing aid internal noise. Journal of theAcoustical Society of America 127(4): 2521–8.

Marcoux, A.M., Yathiraj, A., Cote, I., and Logan, J. 2006.The effect of a hearing aid noise reduction algo-rithm on the acquisition of novel speech contrasts.International Journal of Audiology 45(12): 707–14.

McCreery, R., Ito, R., Spratford, M., Lewis, D., Hoover,B., and Stelmachowicz, P.G. 2010. Performance-in-tensity functions for normal-hearing adults and chil-dren using computer-aided speech perception assessment. Ear and Hearing 31(1): 95–101.

Mueller, H.G., Weber, J., and Hornsby, B.W. 2006. Theeffects of digital noise reduction on the acceptanceof background noise. Trends in Amplification 10(2):83–93.

Nábĕlek, A.K., Freyaldenhoven, M.C., Tampas, J.W.,Burchfield, S.B., and Muenchen, R.A. 2006. Accept-able noise level as a predictor of hearing aid use.Journal of the American Academy of Audiology 17(9):626–39.

A Sound Foundation Through Early Amplification164

Neuman, A.C., Wroblewski, M., Hajicek, J., and Rubin-stein, A. 2010. Combined effects of noise and rever-beration on speech recognition performance of nor-mal-hearing children and adults. Ear and Hearing31(3): 336–44.

Palmer, C.V., and Grimes, A.M. 2005. Effectiveness of sig-nal processing strategies for the pediatric popula-tion: A systematic review of the evidence. Journal ofthe American Academy of Audiology 16(7): 505–14.

Ricketts, T., Galster, J., and Tharpe, A.M. 2007. Directio-nal benefit in simulated classroom environments.American Journal of Audiology 16(2): 130–44.

Ricketts, T.A., and Hornsby, B.W. 2005. Sound qualitymeasures for speech in noise through a commercialhearing aid implementing digital noise reduction.Journal of the American Academy of Audiology 16(5):270–7.

Sarampalis A, Kalluri S, Edwards B, Hafter E. 2009. Ob-jective measures of listening effort: Effects of back-ground noise and noise reduction. Journal of Speech,Language, and Hearing Research 52(5): 1230–40.

Souza, P.E., Jenstad, L.M., and Boike, K.T. 2006. Measuringthe acoustic effects of compression amplification onspeech in noise. Journal of the Acoustical Society ofAmerica 119(1): 41–4.

Stelmachowicz, P., Lewis, D., Hoover, B., Nishi, K.,McCreery, R., and Woods, W. 2010. Effects of digitalnoise reduction on speech perception for childrenwith hearing loss. Ear and Hearing 31 (3): 345–355.

Stelmachowicz, P.G., Hoover, B.M., Lewis, D.E., Korte-kaas, R.W., and Pittman, A.L. 2000. The relation bet-ween stimulus context, speech audibility, and per-ception for normal-hearing and hearing-impairedchildren. Journal of Speech, Language, and HearingResearch 43(4): 902–14.

Tyler, R.S., and Kuk, F.K. 1989. The effects of «noise sup-pression» hearing aids on consonant recognition inspeech-babble and low-frequency noise. Ear andHearing 10(4): 243–9.

Woods, W., Nooraei, N., Galster, J., and Edwards, B.2010. Real-world listening preference for an opti-mized digital noise reduction algorithm. Hearing Review 17(9): 38–43.

Should Digital Noise Reduction be Activated in Pediatric Hearing Aid Fittings? 165