directional benefit in the presence of speech and ... · the behind-the-ear version of the phonak...

5

J Am Acad Audiol 18:5–16 (2007)

*Department of Hearing and Speech Science, Vanderbilt Bill Wilkerson Center, Vanderbilt University

Benjamin Hornsby, Dan Maddox Hearing Aid Research Laboratory, Vanderbilt Bill Wilkerson Center, Room 8310 MedicalCenter East, South Tower, 1215 21st Ave. South, Nashville, TN 37232-8242; E-mail: [email protected]; Phone: 615-936-5132; Fax: 615-936-5013

This research was supported in part by Phonak, Inc., and the Dan Maddox Hearing Aid Research Foundation.

Directional Benefit in the Presence of Speechand Speechlike Maskers

Benjamin W.Y. Hornsby*

Todd A. Ricketts*

Abstract

Recent research suggests that omnidirectional hearing aids are relativelyineffective at improving speech understanding in everyday conversationalspeech settings when the background noise contains both energetic andinformational masking components. Energetic masking refers to situationswhere the peripheral (or neural) activity of the target is less than that of themasker, thus making the target inaudible. In contrast, informational maskingeffects, in this paper, refer to additional masking effects that are not energeticin nature. The current study evaluated the benefits of directional technologyin the presence of background noises that contained both energetic andinformational masking components. Aided speech recognition (in bothomnidirectional and directional modes) was assessed in the presence of threetypes of maskers (forward and reversed speech and speech-modulated noise)that varied in the amount of informational masking they were expected toproduce. Study results showed significant directional benefit in all conditions.This finding suggests that in everyday conversational speech environments,directional technology is equally efficacious regardless of the magnitude ofinformational masking present in the background noise. In addition, studyfindings suggest that the semantic information present in the masking speechmay play only a limited role in contributing to informational masking in everydayenvironments.

Key Words: Directional benefit, energetic masking, informational masking,hearing aids, hearing loss, speech intelligibility

Abbreviations: HINT = Hearing in Noise Test; REIG = real ear insertion gain;RT = Reverberation Time; SNHL = sensorineural hearing loss; SNR = signal-to-noise ratio

Sumario

Investigaciones recientes sugieren que los auxiliares auditivos omnidireccionalesson relativamente ineficientes en cuanto a mejorar la comprensión del lenguajeen condiciones cotidianas de conversación, cuando el ruido de fondo contienecomponentes de enmascaramiento tanto energéticos como de información.El enmascaramiento energético se refiere a situaciones donde la actividadperiférica (o neural) de la señal meta es menor que aquella del enmascarador,haciendo la señal inaudible. En contraste, los efectos de enmascaramientode información, se refieren a efectos de enmascaramiento adicional, que noson de naturaleza energética. El estudio actual evaluó los beneficios de latecnología direccional en presencia de ruidos de fondo que contenían tantocomponentes de enmascaramiento energético como de información. Se

Journal of the American Academy of Audiology/Volume 18, Number 1, 2007

6

One of the most common complaints ofpersons with hearing loss is difficultyunderstanding speech, particularly in

the presence of a background noise (e.g.,Kochkin, 2000). Substantial research hasdemonstrated that providing a well fitomnidirectional hearing aid can improve speechunderstanding and reduce the communicationdifficulties of persons with hearing loss (e.g.,Mulrow et al, 1990; Humes et al, 1997; Larsonet al, 2000, 2002). Despite these findings, manypersons with hearing loss continue to reportsignificant communication difficulties ineveryday environments (Kochkin, 2000). Thedifficulty understanding speech in backgroundnoise is due largely to the masking effects ofthe background noise. In common everydaysettings, background noise may originate froma number of different sources and may includecompeting talkers (speech) and/or nonspeechsignals. When the background noise consistsof speech, the masking effects can be due to both“energetic” and “informational” components.

The effects of “energetic” maskers havebeen well studied and are assumed tooriginate in the auditory periphery (i.e.,cochlea or proximal portions of the auditorynerve). Energetic masking occurs when theexcitation or neural response in a givenfrequency range, due to the target, is lessthan that produced by the background noise

(e.g., Moore et al, 1997). The effects of energeticmasking on threshold and speech understandingare, on average, quite predictable, at least forpersons with normal hearing (e.g., French andSteinberg, 1947; ANSI S3.5, 1997).

In contrast, “informational masking”may be described as masking that is not“energetic” in nature, thus implying more“central” masking effects (e.g., Durlach etal, 2003). When the target and backgroundnoise consists of speech, informationalmasking may occur when there are highlevels of uncertainty regarding thecharacteristics of the target stimulus and/ormasker (Brungart and Simpson, 2004;Freyman et al, 2004). Informational maskingmay also occur when there are similarities inthe temporal and/or semantic structure ofthe background competition and the speechtarget (e.g., Brungart, 2001; Brungart et al,2001). In real-world settings, however, manyfactors come into play that can reduce theeffects of informational masking on speechunderstanding. For example, informationalmasking is reduced when the target speechand maskers are spatially separated, as iscommon in everyday settings (Hawley et al,1999, 2004; Arbogast et al, 2002, 2005).Likewise, differences in vocal characteristicsand the context of the target and maskingspeech, as well as the presence of visual cues,

evaluó el reconocimiento amplificado del lenguaje (tanto en el modoomnidireccional como direccional) en presencia de tres tipos de ruidosenmascarantes (lenguaje anterógrado y reverso, y ruido de lenguaje modulado),con variaciones en la cantidad de enmascaramiento de información quedebían producir. Los resultados del estudio mostraron un beneficio direccionalsignificativo en todas las condiciones. Este hallazgo sugiere que en ambientescotidianos de conversación, la tecnología direccional es igualmente eficiente,independientemente de la magnitud del enmascaramiento de informaciónpresente en el ruido de fondo. Además, los hallazgos del estudio sugieren quela información semántica presente en el lenguaje enmascarante puede jugarsolamente un papel limitado, en ambientes cotidianos, en su contribución comoenmascaramiento de información.

Palabras Clave: Beneficio direccional, enmascaramiento energético,enmascaramiento de información, auxiliares auditivos, pérdida auditiva,inteligibilidad del lenguaje

Abreviaturas: HINT = Prueba de Audición en Ruido; REIG = ganancia deinserción de oído real; RT = Tiempo de Reverberación; SNHL = hipoacusiasensorineural; SNR = tasa señal-ruido

Directional Benefit in Speech Maskers/Hornsby and Ricketts

7

would work together to reduce target/maskeruncertainty and similarities in the temporaland semantic structure between the targetand masking speech, thus reducinginformational masking (Brungart, 2001;Brungart et al, 2001; Helfer and Freyman,2005).

Although the magnitude of informationalmasking effects may be reduced in real-worldenvironments (e.g., settings where thebackground noise consists of spatiallyseparated individual talkers) compared tosome experimental conditions, recentresearch suggests that residual effects mayremain substantial (Hornsby et al, 2006).Hornsby et al measured the signal-to-noiseratio, or SNR, needed for 50% sentencerecognition, for persons with hearing loss inboth the unaided and aided conditions, inthe presence of multiple (two, four, or seven)individual talkers or multiple speech-modulated noises that were spatiallydistributed around the listener. Informationalmasking effects were estimated by comparingthresholds in the speech maskers tothresholds obtained using speech-modulatednoises as maskers. The masking noises wereshaped to match the long-term spectrum ofthe speech maskers and were modulatedwith the envelopes of the individual talkers.The magnitude of informational maskingvaried (between approximately 0.0 and 2.5dB) based on the number of maskers andwhether participants were listening unaidedor aided.

In addition, the study results suggestedthat omnidirectional hearing aids wererelatively ineffective in improving speechunderstanding when the masking noisescontained an informational maskingcomponent (i.e., speech-masker conditions).Specifically, when the background noiseconsisted of actual talkers, improvements inspeech understanding in the aided condition(quantified as the improvement in SNR) weresmall (range of 0.7–1.4 dB), and aidedperformance was not significantly differentthan unaided performance. In contrast, whenthe background noise consisted of modulatedspeech noises, the use of omnidirectional aidsresulted in significant (range of 1.5–2.9 dB)improvements in speech understandingcompared to unaided performance.

The finding that omnidirectional hearingaids are limited in their ability to improvespeech understanding deficits associated with

informational masking in everyday speech,although discouraging, is not totallyunexpected. Hearing aids are designed,primarily, to restore audibility and are thusnot expected to heavily impact the centralfactors (e.g., stimulus-masker uncertainty/similarity) that appear to be largelyresponsible for informational masking effectsthat occur in speech settings. This findingdoes, however, highlight the limitations ofomnidirectional hearing aids in real-worldnoisy environments, particularly those thatcontain speech signals as the primary masker.Given these limitations, there is a clear needfor additional research investigating theutility of additional technological andcounseling strategies that may help reducethe negative effects of informational maskerson hearing aid wearers.

One option for improving speechunderstanding in the presence of speechmaskers is the use of directional technologyin hearing aids. The use of directionaltechnology in hearing aids has been shownto improve speech understanding in a varietyof noise backgrounds, at least in laboratorysettings, compared to performance with well-fit omnidirectional hearing aids (see Rickettsand Dittberner, 2002, for a review). Despitethis success in laboratory environments, thebenefits of directional technology in real-world settings appear to be more modest(Cord et al, 2004) and may be related to thefact that the background noises present inreal-world settings contain both informationaland energetic masking components. Thebenefits of directional technology are dueprimarily to their ability to improve SNR,which for most noises is monotonically relatedto speech understanding. For highlyinformational maskers, however, the functionrelating SNR to speech understanding mayvary substantially and may not always bemonotonic (e.g., Dirks and Bower, 1969;Brungart, 2001). Thus, the benefits ofdirectional technology may be limited orhighly variable in the presence of maskerscontaining both informational and energeticmasking components. The current studyextends our previous work (Hornsby et al,2006) by investigating the benefits in speechunderstanding provided by directionaltechnology in the presence of speech andspeechlike background noises containing bothenergetic and informational maskingcomponents.


8

PROCEDURES

Participants

Participants consisted of 14 adults (sevenmale and seven female) with symmetrical(≤20 dB difference between ears at anyaudiometric test frequency), mild-to-moderately severe flat or slopingsensorineural hearing losses (SNHL). SNHLwas defined as having no significant air-bonegap (<10 dB) at any frequency tested and nohistory of conductive pathology. Allparticipants’ hearing thresholds were between25 and 70 dBHL at 500Hz and between 50and 85 dBHL at 3000 Hz, bilaterally (seeFigure 1). Participants with hearing lossranged in age from 23 to 83 years (medianage: 74 years).

Test Setting

The test environment was the same asthat used in our previous study (Hornsby etal, 2006). Briefly, a 3.2-meter square (2 meters

high) sound-treated room, modified withreflective panels, served as the testenvironment. Frequency-specific reverberationtimes (RT60; time required for 60 dB decayafter signal offset) were measured at theposition of the listener’s head (without thelistener present) using frequency-modulatedtones. Measured RT60 values, at octavefrequencies, were 485 msec (250 Hz), 440msec (500 Hz), 400 msec (1000 Hz), 310 msec(2000 Hz), and 220 msec (4000 Hz).

Hearing Aids

The behind-the-ear version of the PhonakClaro™ (211dAZ) was used as the testinstrument. This is a digital, 20-channel, low-threshold, fast-acting compression hearingaid. This hearing aid’s digital noise reductioncircuitry, referred to by the manufacturer as“fine scale noise cancellation,” was disabledfor all testing. The devices were fit bilaterallyin both omnidirectional and adaptivedirectional modes using the manufacturer’sfast-acting compression algorithm, which theyrefer to as “digital perception processing”(fast-acting DPP™). Using this algorithm,nominal compression thresholds acrosssubjects ranged from approximately 30 to 50dB SPL, and nominal attack and releasetimes were 5 and 90 milliseconds respectively.Further details related to compressionprocessing were not deemed of interest sincepast research has shown that directionalbenefit is generally unaffected by the presenceof low-threshold compression and thecompression parameters selected (Ricketts etal, 2001).

Given the relatively discrete competingnoise source positions used in this experiment(i.e., single interferers from multipleloudspeakers), the adaptive, rather thanfixed, directional mode was chosen for use inthis study. It is important to note that“automatic” switching between directionaland omnidirectional modes was not used inthis study. That is, the aids were programmedto function in either omnidirectional oradaptive directional mode with the specificmode chosen by the experimenter (i.e., the aidcould not “automatically” switch betweenprograms). In adaptive directional mode, thetwo omnidirectional microphones bothconstantly function in a directional array.The directional sensitivity pattern is then

Figure 1. Average audiograms of study participantswith hearing loss. Open and filled symbols are for theleft and right ears, respectively. The range of hear-ing losses included in this study is shown by dashedlines.


9

altered by automatically varying the physicaldirectional properties (by varying internaldelay) until an attenuation pattern resultingin the lowest output is obtained. This processof shifting directional sensitivity patternsrequires only several milliseconds to complete.The adaptive directional circuitry is limitedso that directional parameters that result innulls in the front hemisphere are avoided. Inthis way important sounds that arrive fromthe front hemisphere are not inadvertently,and undesirably, attenuated (Checkley andKuehnel, 2000).

In theory, the adaptive mode should allowthe device to optimize the directionalcharacteristics for a given speakerconfiguration and therefore allow forassessment of optimal, but realistic,directional benefit in the presence of discrete,stationary, competing maskers.

Participants with hearing loss were fitbilaterally. The hearing aid fittings, based onthe NAL-RP prescriptive method (Byrne andDillon, 1986), were verified using thecomposite noise test signal (0° azimuth) onthe Frye Systems Fonix 6500 real earanalyzer. For a 65 dB input, averagemeasured real ear insertion gain (REIG)values were within 4 dB of prescribed gainvalues at test frequencies of 250 to 4000 Hzand within 6 dB at 6000 Hz. The averagedifference between omnidirectional anddirectional REIG was less than approximately2 dB at frequencies between 250 and 6000 Hz(see Figure 2).

Speech Understanding Test Materialsand Masking Stimuli

Speech understanding in noise wasassessed using a modified version of theHearing in Noise Test (HINT; Nilsson et al,1994). The modifications to the HINTprocedure were all related to the presentationand type of competing noise used as describedbelow. The HINT is an adaptive procedureused to determine the SNR necessary toachieve 50% correct sentence recognition.Using this test the sentence level is adaptivelyvaried in the presence of a constant levelbackground noise to determine threshold.During speech testing the background noisewas turned on approximately one secondprior to the presentation of the first sentenceand was on continuously during testing (i.e.,

there were no silent periods betweensentences). Each experimental condition wasevaluated using two ten-sentence lists.Experimental condition and list order wererandomly assigned to each subject using aLatin square design. In addition, theparticipants were tested in bothomnidirectional and directional modes withthe presentation order counterbalanced.

Figure 2. Average measured and target real earinsertion gain (REIG in dB) for the right and left earsin omnidirectional (circles) and directional (trian-gles) modes. Error bars represent one standard devi-ation around the mean. For clarity, symbols areslightly offset on the frequency axis.


10

Speech materials were presented through asingle loudspeaker (Tannoy System 600)located approximately 1.2 meters from thesubject (0° azimuth) and positioned atapproximately ear level.

Speech recognition was assessed in afour- and seven-masker configuration usingactual talkers (forward speech), reversedspeech, and modulated speech-shaped noisesas maskers. Recent work suggests that theinformational masking effects of individualtalkers peaks as the number of individualtalkers approaches eight (Simpson and Cooke,2005). Using the four- and seven-maskerconfigurations allows us to maximizeinformational masking effects whilemaintaining face validity in the experimentalconfiguration (e.g., speech maskerssurrounding a listener).

In each masker configuration, speechunderstanding was assessed using maskersthat varied in the amount of informationalmasking they were expected to produce. Theforward speech maskers consisted of maletalkers reading a specific topic and wereassumed to produce the highest amount ofinformational masking. The male talkersread test passages from the Connected SpeechTest (CST; Cox et al, 1987, 1988). Each CSTpassage consists of approximately tensentences describing a specific topic (e.g.,lemons). The passages were derived from achildren’s educational reading source (seeCox et al, 1987, for details). The CST passageswere chosen as masking materials becausethey provide a contextually rich dialogue,rather than single unrelated sentences, thatmay result in informational masking moreconsistent with that present in everydaycommunication settings. Recordings of theindividual talkers reading the CST passageswere made in an anechoic chamber and storedon a computer hard disk. All talkers werenative speakers of American English. Offlinedigital filtering, using the FIR2 function anda 1000th order FIR filter implemented inMatlab™, was used to match the long-termrms spectra of each CST passage recorded byeach talker to that of the long-term rmsspectra of the HINT materials.

The modulated noise maskers used inthis study were derived from uncorrelatedsegments of Gaussian noise that were firstspectrally shaped, using the FIR2 functionand a 1000th order FIR filter implementedin Matlab, to match the long-term rms

spectrum of the HINT materials. The shapednoises were then modulated using theenvelopes of the same single talkers readingthe CST passages. The envelopes of thesingle-talker maskers were derived in Matlabby implementing half-wave rectification of agiven single-talker passage, followed by low-pass filtering of the half-wave rectified signal,using a sixth-order butterworth filter with a30 Hz low pass cutoff frequency. The envelopewas then applied to the shaped noiseproviding a primarily energetic masker thatretained the long-term spectral and temporalpatterns of the single-talker maskers. Thelong-term spectrum of all individual-talkerand modulated noise maskers closelyapproximated the long-term spectrum of theHINT sentences. The mean error, across one-third octave bands from 160 to 8000 Hz,between individual talkers, modulated noisesand the HINT spectrum was 0.19 and -0.22dB, respectively, with standard deviationsless than 1 dB. The maximum error, in agiven one-third octave band, for any individualtalker or modulated noise was ~3 dB.

In addition, a third masker typeconsisting of reversed speech was added to theexperimental conditions. Specifically, theforward speech maskers were reversed inthe time domain, resulting in an unintelligiblemasking noise with similar spectral andtemporal properties as the forward speech. Itwas expected that reversing the speech wouldcreate a masker with similar energeticmasking properties compared to the speechbut without semantic information that maycontribute to target and masker similarity,thus reducing informational masking.

We estimated informational maskingeffects by comparing recognition performancein the forward speech-masker condition (moreinformational masking) to performance inthe reversed speech and speech-modulatednoise masker conditions (less informationalmasking). In summary, sentence recognitionwas assessed in a total of 12 conditions (twoaid types [omni and directional] x two noiseconfigurations [four and seven maskers] xthree masker types [forward and reversedspeech and speech-modulated noise]).

Each masker was presented from aseparate loudspeaker that was spatiallyseparated from the target speech, which wasalways located at a 0° azimuth. Althoughspatially separating the speech and noiseprovides a more realistic test configuration,


11

it should be noted that this configurationwas expected to reduce the amount ofinformational masking compared to thatmeasured in some previous laboratory studies(Hawley et al, 1999, 2004; Arbogast et al,2002, 2005). In the four-masker configuration,speakers were located symmetrically atazimuths of 45°, 135°, 225°, and 315°. In theseven-masker configuration, three additionalspeakers were placed at azimuths of 90°,180°, and 270°. All maskers were presentedfrom Definitive Technology BP-2X bipolarloudspeakers placed 1.2 meters from thelistener. Informal testing confirmed that,although difficult, listeners with normalhearing could attend to the speech ofindividual speech maskers even when allseven were presented simultaneously.

In each test condition the combined levelof the maskers (at the position of the listener’shead with the listener absent), was a constant65 dBA. The level of each individual maskerwas equated prior to adjusting the overalllevel of the combined maskers to 65 dBA.Calibration prior to each test session wasperformed to assure an overall level of 65dBA (±1 dB; Larson Davis 814 sound levelmeter set to slow averaging) in allexperimental conditions.

RESULTS

The current study examined the effectsof type and number of maskers on the aidedspeech understanding of persons with hearingloss using both omnidirectional and directionalhearing aid modes. Figure 3 shows HINTthresholds obtained in each masker type asa function of number of maskers in bothomnidirectional and directional modes. Plottedin this fashion, a more negative thresholdrepresents better understanding in noise.

Performance differences across conditionswere examined using a three-factor, repeated-measures analysis of variance (ANOVA). Thewithin subjects independent variables weremicrophone mode (omnidirectional ordirectional), masker type (forward speech,reversed speech, or speech-modulated noise),and number of maskers (four or seven); thedependent variable was the HINT score ineach masker condition. A 0.05 level ofsignificance was used in all analysesdescribed here. Summary results of theprimary analyses are shown in Table 1.

As expected, a significant main effect ofmicrophone mode was observed with betterperformance occurring when listening withthe microphones set to directional mode.When switching from omnidirectional todirectional modes, performance improved,on average, between 3.3 dB and 4.6 dB acrossthe test conditions. This directional benefitvaried widely across individuals ranging from

Figure 3. Thresholds (in dB SNR) required for 50%sentence recognition for aided participants with hear-ing loss as a function of number of maskers (four orseven). Circles, triangles, and squares represent per-formance in background noises of forward speech,reversed speech, and modulated noise, respectively.Performance in omnidirectional and directional modesis shown by the open and filled symbols, respectively.Error bars represent one standard deviation.

Table 1. Results from a Three-Factor, RepeatedMeasures ANOVA

Overall ANOVA Results

Effect Df F p-level

Microphone 1,13 580.8 <0.001

Masker Type 2,26 38.8 <0.001

Num_Maskers 1,13 4.57 0.052

Microphone x Masker Type 2,26 1.47 0.248

Microphone x Num_Maskers 2,26 0.45 0.517

Masker Type x Num_Maskers 2,26 1.97 0.160

Microphone x Masker Type x Num_Maskers 2,26 0.87 0.433

Note: Significant effects with p-levels less than 0.05 areshown in boldface. Microphone: omnidirectional or directional;Masker Type: forward speech, reversed speech, or noise masker;Num_Maskers: number of maskers (i.e., four or seven).

a low of 0 dB to a maximum of 7.7 dB. Nosignificant interaction effects betweenmicrophone mode and any other factor wereobserved, suggesting that, on average,significant directional benefit was presentand similar in magnitude across all testconditions. Figure 4 shows directional benefit(omnidirectional-directional HINT thresholds)for the test conditions evaluated in this study.

In addition to the significant main effectof microphone mode, a significant main effectof masker type was also present. There were,however, no significant interaction effects.The lack of a significant interaction suggeststhe effect of masker type was similar in bothomnidirectional and directional modes and inboth the four- and seven-masker conditions.Follow-up testing was performed, using aseries of three-factor ANOVAs, to compareperformance differences between maskertypes. These planned comparisons showedthat performance in the modulated noisebackground was significantly better (df1,13; p< 0.001) than when the background noiseconsisted of forward or reversed speech. Thisadditional masking, above that observed withthe modulated noise, suggests that both theforward and reversed speech maskers

contained an additional informationalmasking component that affected speechunderstanding.

In addition, there was no significantdifference (df1,13; p = 0.078) in performancebetween the forward and reversed speechbackground noise conditions. In other words,reversing the speech maskers provided nomeasurable release from informationalmasking in any of our test conditions. Figure5 shows the magnitude of informationalmasking effects in the current study (thedifference in HINT thresholds measured inspeech and speech-modulated noise maskers)for the study participants in bothomnidirectional and directional modes, as afunction of number of maskers.

As seen in Figure 5, the magnitude ofinformational masking effects in the currentstudy varied slightly (but not significantly)across microphone mode, type of masker(forward or reversed speech), and number ofmaskers. The lack of a significant interactioneffect here may be due in part to relativelylow statistical power. The low power, however,is due largely to the fact that relatively smalldifferences are observed across conditionsrelative to the variability across study


12

Figure 4. Directional benefit (omnidirectional thresholds–directional thresholds for 50% sentencerecognition) in dB SNR as a function of number ofmaskers (four or seven) and type of masking noise(forward speech, circles; reversed speech, triangles;and speech modulated noise, squares). Error bars represent one standard deviation. Positive valuesreflect better performance in directional mode.

Figure 5. Informational masking effects, defined asthe difference between the HINT thresholds in the for-ward or reversed speech masker minus the HINTthresholds in the modulated noise masker, as a func-tion of number of maskers. Informational maskingeffects in omnidirectional and directional modes areshown by the open and filled symbols respectively.Results for forward (circles) and reversed (triangles)speech maskers are shown as separate symbols. Errorbars represent one standard deviation.


13

participants. Informational masking effects(additional change in SNR required for 50%sentence recognition over that due to amodulated noise) ranged from approximately1.0 to 2.6 dB across conditions.

DISCUSSION

Masker Type and Directional Benefit

The data shown in Figure 5 show thatinformational masking effects were similarin magnitude to those observed in the aided(omnidirectional) conditions of our previousstudy, which ranged from approximately 1.5to 2.5 dB (Hornsby et al, 2006). It should benoted, however, that these values arerelatively small compared to some studiesdesigned to maximize informational maskingeffects.

For example, Arbogast et al (2002, 2005)highlighted informational masking whileminimizing the effects of energetic maskingby creating sine wave speech targets andmaskers. The sine wave speech was createdby modulating pure tones of a given frequencywith the amplitude envelope of actual speechfiltered at that center frequency. The authorsminimized energetic masking by placing thesine waves used to generate the maskingspeech at center frequencies well removedfrom the center frequencies used to create thetarget speech. The result was intelligibletarget and masking speech that had minimalspectral overlap and thus minimal energeticmasking. Using this method, Arbogast et al(2005) reported informational masking effects(the difference between speech recognitionthresholds for sine wave speech obtainedusing modulated noise versus sine wavesentences as background maskers) ofapproximately 7.0 to 12.0 dB in their unaidedparticipants with hearing loss. Thus,informational masking effects, under certainconditions, can be quite large (Hawley et al,1999, 2004; Arbogast et al, 2002, 2005).

The results from the current study andthose of Hornsby et al (2006), however,suggest that in everyday conversationalspeech settings (i.e., where the target speechis a single talker and the masking speech ismade up of multiple, different, spatiallyseparated talkers), informational maskingeffects, while present, may be relatively small.

A primary focus of the current study was toexplore the benefits of directional processingin the presence of background noises thatwere designed to vary in the amount ofenergetic and informational masking theywere expected to produce. Despite thepresence of a significant informationalmasking component, our study resultsrevealed significant directional benefit thatwas independent of masker type. That is,the improvement in SNR provided by thedirectional technology resulted in comparableimprovements in speech understandingregardless of whether the background noiseconsisted of actual speech, reversed speech,or speech-modulated noise. In addition, themagnitude of directional benefit (3.3 dB to 4.6dB) in the current study was comparable tothat seen in other studies where thebackground noise surrounded the listener(Ricketts and Dittberner, 2002). Thesefindings are encouraging given previousresults showing that omnidirectional hearingaids were relatively ineffective at improvingspeech understanding when the backgroundnoise consisted of spatially separatedindividual talkers as in the current study(Hornsby et al, 2006).

The finding of improved speechunderstanding with the use of directionaltechnology in the presence of speech maskersis not totally unexpected given our attemptat simulating somewhat real-worldconditions. In most cases, when measuringspeech understanding in noise, a systematicincrease in SNR (which is comparable toswitching from omnidirectional to directionalmode) results in a monotonic improvement inspeech understanding, although the slopes ofthe performance functions may vary based onthe target speech and masking materialsused (e.g., Miller, 1947).

This monotonic decrease in performanceas the SNR worsens can occur even insituations where informational masking isthought to play the prominent role (Brungartet al, 2001; Arbogast et al, 2002, 2005). Insome cases, however, “plateaus” in thefunction relating speech recognition to SNR(i.e., a nonmonotonic function) have beenobserved, specifically when the test SNR isin the range of 0 to -10 dB (Egan et al, 1954;Dirks and Bower, 1969; Brungart, 2001). Insome cases at these SNRs, performanceremains stable despite a change in SNR.This “plateau” appears to be limited, however,

to situations where substantial informationalmasking is present, such as when the maskeris a single talker speaking with the sametemporal characteristics as the target speech(Dirks and Bowers, 1969). In addition, the“plateau” often disappears and a monotonicfunction is again observed when more thana single talker is used as the masking noise(Brungart et al, 2001). Consequently, in real-world situations for which directionaltechnology is capable of improving the SNR,improvements in speech understandingshould be possible regardless of the type ofbackground noise.

Informational Masking Effects inReversed Speech

Another finding of interest in the currentstudy was the relative lack of benefit obtainedwhen reversing the forward speech maskers (seeFigure 5). Our study results showed nosignificant difference in performance betweenthe forward and reversed speech maskers in anycondition. One interpretation of these resultsis that the semantic content of the forwardspeech masker resulted in no additionalinformational masking. This is consistent withprevious research showing no difference inspeech understanding when using forward andreversed speech maskers (Miller, 1947; Dirksand Bower, 1969; Duquesnoy, 1983; Festen andPlomp, 1990; Larsby and Arlinger, 1994).

Other studies, however, have shown ameasurable improvement in speechunderstanding with the reversal of a forwardspeech masker (Sperry et al, 1997; Freyman etal, 2001; Hawley et al, 2004; Summers andMolis, 2004). This effect tends to be largest,however, in conditions where substantialinformational masking is thought to exist. Forexample, release from masking due to timereversal of a forward speech masker has beenobserved in conditions where there is no spatialseparation between the target and maskingspeech and/or in conditions where substantialsimilarities exist in the spectro-temporal and/orsemantic characteristics of the target andmasking speech (e.g., Freyman et al, 2001;Hawley et al, 2004). Freyman et al (2001)observed a substantially larger improvementin speech understanding due to reversal of atwo-talker speech masker when the targetspeech and maskers both came from a 0°azimuth (high informational maskingcondition). The release from masking due totime reversal was substantially smaller when

the target speech and maskers were perceived(via use of the precedence effect) as spatiallyseparated (resulting in less informationalmasking).

By creating a time-reversed copy of theforward speech masker, the underlyingassumptions are that the temporal and spectralcharacteristics of the forward speech masker,and thus its energetic masking properties, areessentially unchanged while the informationalmasking component due to the semanticinformation in the masker is removed.However, a contrasting view was suggestedby Rhebergen et al (2005). These authorssuggested that the acoustic characteristics ofreversed speech result in significantly higheramounts of energetic masking than wouldoccur when forward speech was used as amasker. Specifically, they suggest the envelopeof forward speech is dominated by the relativelyabrupt onsets and gradual decays of plosivesounds. When the speech is reversed, theplosives provide an abrupt offset that maymake the reversed speech a more effectiveforward masker; thus, we would expect poorerthresholds in the presence of a reversed speechmasker if no additional factors were involved.

To examine this hypothesis, Rhebergenet al measured performance in the presence ofa forward and reversed foreign languagemasker (Swedish) and reported poorerperformance using the reversed foreignlanguage masker. In this situation there wasno usable semantic information in eithermasker, suggesting that the poorer performancewas due to greater energetic masking effectsin the reversed speech. It should be noted,however, that Dirks and Bower (1969)performed a similar experiment using a Latinmasker and found no difference in performancein forward and reversed modes. Thus, thevariation in forward masking effects in reversedand forward speech may vary as a function oflanguage.

In the current study, although no differencebetween forward and reversed speech wasobserved, performance in the reversed speechbackground noise was significantly poorer thanin the modulated noise. This could be due, assuggested by Rhebergen et al (2005), to anincrease in energetic masking properties ofthe reversed speech (due to increased forwardmasking), compared to the speech-modulatednoise, that was offset by a similar amount (butin the opposite direction) by a release frominformational masking due to the time reversalof the speech maskers.


14


15

An alternative interpretation, if oneassumes that no differences exist in the forwardmasking properties of forward and reversedAmerican English speech, is that thesimilarities in the temporal and spectralcharacteristics of the target and masking (bothforward and reversed) speech are the primaryfactors responsible for the “informational”masking observed in the current study, notdifficulty disentangling the semanticinformation in the target from that in themasker. Dirks and Bower (1969) highlightedhow similarities in the temporal structure ofthe target and masking speech could alter themasking characteristics of a background noise.While measuring synthetic sentenceidentification as a function of SNR in differentbackground noises, they observed a “plateau”in performance as SNR decreased in certaintest conditions. Specifically, speechunderstanding remained stable in cases wherethe same speaker delivered both the targetand masking stimuli at approximately thesame level (e.g., between 0 to -10 dB SNR).

This “plateau” effect is associated withinformational masking, in that energeticmasking effects are known to change withSNR, yet no appreciable change inperformance is observed (Brungart, 2001).Dirks and Bower measured sentencerecognition using both intelligible (English)and unintelligible (Latin) masking stimuli inboth forward and reversed modes. In theseconditions a “plateau” was most apparentwhen either the English or Latin maskerswere presented in the forward mode. In anadditional condition, a single subject wastrained to identify reversed syntheticsentences as the target speech and listenedto this reversed target speech in both forwardand reversed (English and Latin) maskingconditions. In this condition, the “plateau”was most apparent when the maskers werealso played in reversed mode. Together thesefindings suggest that it is the similarities inthe temporal properties of the target andmasking speech, not the semantic informationin the masker, which were responsible forthis plateau in performance.

In the current study, there aresubstantial similarities in the short- andlong-term temporal structure of the targetand masking speech while substantialdifferences in semantic information exist.This is similar to situations that occur ineveryday environments when listening to a

single talker in a background noise ofconversational speech from multiple talkers.In these cases, it is possible that thesimilarities in temporal structure of the targetand masking stimuli, rather than thesemantic properties of the masker, are largelyresponsible for any “informational masking”effects. Further research in this area isneeded to identify the roles that these andother factors play in any informationalmasking effects that may be present ineveryday conversational settings.

SUMMARY

In the current study, the use of directionaltechnology provided a significant

improvement in speech recognition forpersons with hearing loss. This benefit waspresent, and comparable in magnitude,regardless of whether the background noisewas primarily energetic (speech-modulatednoise) or contained both energetic andinformational masking components (forwardand reversed speech). These findings offeradditional support for the use of directionaltechnology in everyday settings where thepresence of background noise, whether speechor nonspeech, limits speech understanding.

REFERENCES

ANSI (American National Standards Institute). (1997)ANSI S3.5-1997 American National StandardMethods for the Calculation of the Speech IntelligibilityIndex. New York: American National StandardsInstitute.

Arbogast TL, Mason CR, Kidd G. (2002) The effect ofspatial separation on informational and energeticmasking of speech. J Acoust Soc Am 112(5):2086–2098.

Arbogast TL, Mason CR, Kidd G. (2005) The effect ofspatial separation on informational masking of speechin normal-hearing and hearing-impaired listeners. JAcoust Soc Am 117(4):2169–2180.

Brungart DS. (2001) Informational and energeticmasking effects in the perception of two simultane-ous talkers. J Acoust Soc Am 109(3):1101–1109.

Brungart DS, Simpson BD. (2004) Within-ear andacross-ear interference in a dichotic cocktail partylistening task: effects of masker uncertainty. J AcoustSoc Am 115(1):301–310.

Brungart DS, Simpson BD, Ericson MA, Scott KR.(2001) Informational and energetic masking effectsin the perception of multiple simultaneous talkers.J Acoust Soc Am 110(5):2527–2538.

Byrne D, Dillon H. (1986) The National AcousticLaboratories’ (NAL) new procedure for selecting the

gain and frequency response of a hearing aid. EarHear 7(4):257–265.

Checkley P, Kuehnel V. (2000) Advantages of an adap-tive multi-microphone system. Hear Rev 8(9):58–60,74.

Cord MT, Surr RK, Walden BE, Dyrlund O. (2004)Relationship between laboratory measures of direc-tional advantage and everyday success withdirectional microphone hearing aids. J Am AcadAudiol 15(5):353–364.

Cox RM, Alexander GC, Gilmore C. (1987)Development of the Connected Speech Test (CST).Ear Hear 8(5, Suppl.):119S–126S.

Cox RM, Alexander GC, Gilmore C, Pusakulich KM.(1988) Use of the Connected Speech Test (CST) withhearing-impaired listeners. Ear Hear 9(4):198–207.

Dirks D, Bower D. (1969) Masking effects of speechcompeting messages. J Speech Hear Res 12(2):229–245.

Duquesnoy AJ. (1983) Effect of a single interferingnoise or speech source upon the binaural sentenceintelligibility of aged persons. J Acoust Soc Am74(3):739–743.

Durlach NI, Mason CR, Kidd G, Arbogast TL, ColburnHS, Shinn-Cunningham BG. (2003) Note on infor-mational masking. J Acoust Soc Am 113(6):2984–2987.

Egan J, Carterette EC, Thwing E. (1954) Some fac-tors affecting multi-channel listening. J Acoust SocAm 26(5):774–782.

Festen JM, Plomp R. (1990) Effects of fluctuatingnoise and interfering speech on the speech-receptionthreshold for impaired and normal hearing. AcoustSoc Am 88(4):1725–1736.

French NR, Steinberg JC. (1947) Factors governingthe intelligibility of speech sounds. J Acoust Soc Am19:90–119.

Freyman RL, Balakrishnan U, Helfer KS. (2001)Spatial release from informational masking in speechrecognition. J Acoust Soc Am 109(5, pt. 1):2112–2122.

Freyman RL, Balakrishnan U, Helfer KS. (2004)Effect of number of masking talkers and auditorypriming on informational masking in speech recog-nition. J Acoust Soc Am 115(5, pt. 1):2246–2256.

Hawley ML, Litovsky RY, Colburn HS. (1999) Speechintelligibility and localization in a multi-source envi-ronment. J Acoust Soc Am 105(6):3436–3448.

Hawley ML, Litovsky RY, Culling JF. (2004) The ben-efit of binaural hearing in a cocktail party: effect oflocation and type of interferer. J Acoust Soc Am115(2):833–843.

Helfer KS, Freyman RL. (2005) The role of visualspeech cues in reducing energetic and informationalmasking. J Acoust Soc Am 117(2):842–849.

Hornsby B, Ricketts T, Johnson E. (2006) The effectsof speech and speech-like maskers on unaided andaided speech recognition in persons with hearing loss.J Am Acad Audiol 17(6):435–450.

Humes LE, Christensen LA, Bess FH, Hedley-Williams A. (1997) A comparison of the benefit

provided by well-fit linear hearing aids and instru-ments with automatic reductions of low-frequencygain. J Speech Lang Hear Res 40(3):666–685.

Kochkin S. (2000) MarkeTrak V: “Why my hearingaids are in the drawer”: the consumer’s perspective.Hear J 53(2):34, 36, 39–41.

Larsby B, Arlinger S. (1994) Speech recognition andjust-follow-conversation tasks for normal-hearing andhearing-impaired listeners with different maskers.Audiology 33(3):165–176.

Larson VD, Williams DW, Henderson WG, LuethkeLE, Beck LB, Noffsinger D, Wilson RH, et al. (2000)Efficacy of 3 commonly used hearing aid circuits: acrossover trial. NIDCD/VA Hearing Aid Clinical TrialGroup. J Am Med Assoc 284(14):1806–1813.

Larson VD, Williams DW, Henderson WG, LuethkeLE, Beck LB, Noffsinger D, Bratt GW, et al. (2002) Amulti-center, double blind clinical trial comparingbenefit from three commonly used hearing aid cir-cuits. Ear Hear 23(4):269–276.

Miller GA. (1947) The masking of speech. PsycholBull 44:105–129.

Moore BCJ, Glasberg BR, Baer T. (1997) A model forthe prediction of thresholds, loudness, and partialloudness. J Audio Eng Soc 45(4):224–240.

Mulrow CD, Aguilar C, Endicott JE, Tuley MR, VelezR, Charlip WS, Rhodes MC, et al. (1990) Quality-of-life changes and hearing impairment. A randomizedtrial. Ann Intern Med 113(3):188–194.

Nilsson M, Soli S, Sullivan JA. (1994) Developmentof the Hearing In Noise Test for the measurement ofspeech reception thresholds in quiet and in noise. JAcoust Soc Am 95:1085–1099.

Rhebergen KS, Versfeld NJ, Dreschler WA. (2005)Release from informational masking by time rever-sal of native and non-native interfering speech. JAcoust Soc Am 118(3):1274–1277.

Ricketts T, Dittberner AB. (2002) Directional ampli-fication for improved signal-to-noise ratio: strategies,measurements, and limitations. In: Valente M, ed.Hearing Aids: Standards, Options, and Limitations.2nd ed. New York: Thieme, 274–346.

Ricketts T, Lindley G, Henry P. (2001) Impact of com-pression and hearing aid style on directional hearingaid benefit and performance. Ear Hear 22(4):348–361.

Simpson S, Cooke M. (2005) Consonant identifica-tion in N-talker babble is a nonmonotonic function ofN (L). J Acoust Soc Am 118(5):2775–2778.

Sperry JL, Wiley TL, Chial MR. (1997) Word recog-nition performance in various background competitors.J Am Acad Audiol 8(2):71–80.

Summers V, Molis MR. (2004) Speech recognition influctuating and continuous maskers: effects of hear-ing loss and presentation level. J Speech Lang HearRes 47(2):245–256.


16

directional benefit in the presence of speech and ... · the behind-the-ear version of the phonak...

Documents