Embed Size (px)
Citation preview
/
\NZ VeteransAdministration
Journal of Rehabilitation Researchand Development, Vol . 23 No. 1 1986pages 1-15 SPECIAL ARTICLE
~
°
~
°
^
Performance Requirementsfor Hearing Aids
HARVEY 0LLOw.B.F ., PkD . aJOHN MACRAE, B .A. (Hoes)
National Acoustic Laboratories, Australia5 Hickson RoadSydney, 2000Australia
Abstract — The requirements for hearing aid frequencyresponses, SSPL characteristics, compression limiting,directional response, and internal noise are examined.On the basis of the assumptions made, aids would bestserve the hearing impaired if they had flat very-low-frequency gains (and various mid- and high-frequencygains), S$PL responses adjustable in shape and level,directional responses, and internal noise levels below arecommended criterion level.
Introduction
The work reported in this paper had as its aim thedetermination of specifications for hearing aids to beissued by the National Acoustic Laboratories (NAL) ofAustralia. It examines the performance range needed in afamily of hearing aids if a reasonable proportion of thehearing impaired population is to be adequately fitted.The performance range recommended is not far removedfrom that which can be achieved using existing compo-nents in a behind-the-ear hearing aid case . Since there isno unanimous agreement about the basis on which pa-rameters such as gain, frequency response, maximumoutput, type of compression, and directionality of micro-phone should be selected, it is obviously impossible tobe dogmatic about what constitutes the ideal range ofhearing aids . The recernmendations made here arise, inpart, from the fitting procedures in current use at theNational Acoustic Laboogoh*o<NAU . in part from spe-cially oonductedoxpmrimon1o .andinpart from consider-ation of basic theoretical principles.
Specifically, the required range of frequency responses
is determined by applying the threshold-based selectionprocedure published by Byrne and Tonisson in 1976 (1) to
a representative sample of audiograms. The saturated
sound pressure level (SSPL) requirements were obtained
hvassVnR\Dg that the SSPL should not exooed1U*o!iont'u~!udneao discomfort level (LDL), and by per\orningl~}L
--~ap0rkl0n g of kills paper were
prepared while this author was atthe Center for Research in Speech and Hearing Sciences, CityUniversity of New York, 33 West 42nd Street, New York, NewYork 10036 .
measurements with a hearing-aid-type transducer andcoupling system on a representative sample of hearingimpaired people . It is pointed out that the resulting com-plex SSPL requirements can be met with a single-channelhearing aid if the compression circuitry is appropriatelyarranged. Recommendations about the din*o1ional char-acteristics of aid microphones are based largely onmeasurements made on a KEMAR anthropometric acous-tic manikin .whUerouommonda\ionaabou\themaximumtolerable internal noise of hearing aids are based on aspecially performed experiment which determined themaximum tolerable signal-to-noise ratio for narrow- andwide-band random noise added to continuous discourse.
Gain and Frequency Response
Probably the most important specifications for a hear-ing aid are its gain and its frequency response . Althoughthe two terms really relate to the same thing, i .e ., theresponse curve showing gainat each frequency, it iocon-venient to separate the curve into two components . "Fre-quency response" is usually taken to mean the shape ofthe response curve, and "gain" is used to refer to theoverall degree of amplification . Gain is usually quantifiedby the maximum gain, or by the average gain at specifiedfmquanoiea, or the gain at 1 kHz or 2 kHz. Both gain andfrequency response are presently selected at NAL on thebasis of the procedure described by Byrne and Tonisson(1), which requires only threshold loss as its input data . Itis thus possible, using this procedure, to predict whatrange of responses will be required if audiograms of thepopulation to be fitted are available.
The data used \odetermine thogain andfrequency re-sponse nequ!vementoconoie1edcf pure tone audiogramsobtained from the case reports, selected at random, of229 children, 249 war veterans, and 219 old age pensionersfitted with hearing aids.
For each audiogram, the required open access 2-cccoupler (HA, configuration, ANSI S3 .7-1973) gain at eachoctave frequency from 250 Hz to 8000 Hz was calculatedby means of a slightly modified version of the August 1976revision of the method described by Byrne and Tonissonin (1) . In conductive and mixed-loss cases, one-quarter ofthe air-bone gap was added to the gain that would havebeen required had the loss been purely sensorineural.
In cases where the threshold exceeded the limit of theaudiometer, the hearing threshold level (HTL) was assumedto be 5 dB greater than the audiometer limit.
Cumulative frequency distributions ol the 'requited gainat each frequency are given in Figure 1 . The required gainsvary from 0d8(u\250Mzand50VH z) vp\o90d8(at\xVz
and 2 kHz). The cases requiring these 'jery large gains
had significant conductive components in the hearing
u
2
DILLON and MACRAE : Performance Requirements
. wisni
ImIaMmErmDamolvim111111111111,111n11IM111MIT
MiWnlinIVIIIIIIIII/11111115611 II
EMI 1115° IIIIVAIIMIII11•11111111PAIIIIIINSIM1111111111111111111111/811•11111110MIPM1111=11111111111111111=11111111111rIM2111111101 .11111•1111=1111111=1FIKMIMEE
11WEIIIIIIIIINIMIMMI1111111•111111111IM1111111MMIE111111111
11111111IMEWINMI11•11MMEIInIII
IMIMINMRAMMNNMMNMMNMIMMMNMNNMNMMNMNIMIMNMI0
10
20
30
40
50
60
70
80
90
GAIN (dB)
MIN WI111110M.11.111
1111111•1MIEWENNIM111/111111111
1BMMMIIIIIIIIA111•1111IM111111111=1111.11.1111IMMI
AdliElIM1111111111IIIE.rfiili AMIl
rk\''' \p*
•111.111,11•11•111al= AM AIVAE11111WAN11=1111all1111W AMIN III MEIN1111FAIIIAM1111111111PIMINIIIIIIIIIIINIIIIII
=El IMIIIIIIMII=1111MI1111111111.1111111=11111111111MEM11111111011I11111101MIMIll 11111EIMINIMI1111111111111111111I
-20
-10
0
10
20
30
40
50SLOPE (dB/Oct)
99 999 8
995
99
98
>-• 80J
70<00 600CC 50
95
• 90
2
1
05
02
5
FIGURE 1Cumulative frequency distributionsof required open access 2-cc coupletgain for the entire sample at frequen-cies 250 Hz to 8 kHz .
FIGURE 2Cumulative frequency distributions
of required slopes at octaves 250 Hz/500 Hz, 500 Hz/1 kHz, 1 kHz/2 kHz,
2 kHz/4 kHz and 4 kHz/8 kHz for theNAL client population.
99.9998
99 .5
99
98
95
- 90
>—-
80
7003
6003
50
0_
40
>uJ 30
1.'.rc-
20
E 10
5
2
0. 5
02
3
Journal of Rehabilitation Research and Development Vol . 23 No . 1 1986
loss. Frequency response requirements were assessedmainly by examining the range of slopes required in eachoctave between adjacent audiometric frequencies (Fig . 2).
The 500 Hz to 1 kHz octave obviously requires the greatestslope, on average . The 250 Hz to 500 Hz octave is note-worthy in that the cumulative curve does not span a largerange; that is, 95 percent of the population can be fitted by aslope between -8 and +12 dB/octave with the median(50percent) case requiring only +2dB/ootava . (A nega-tive slope indicates a response whose gain decreases asfrequency rises.) Also noteworthy is the large span re-quired in each of the three octaves above 1 kHz.
The cumulative slope curves were broken down invari
ous ways . Cumulative slope distributions are shownseparately for each population type in Figures 3(a) to 3(e).The population needs are different only in the 2-kHz to4-kHz and the 4-kHz to 8-kHz octaves . In the first of these,the curve for the child population is displaced to the leftof the other two, indicating that, on average, children'shearing losses do not deteriorate from 2 kHz up to 4 kHzas much as do adult losses . (This is not too surprising,given the noise-induced and/or presbycusic origins ofmost adult hearing losses, especially among the olderpopulation served by NAL .) Between 4 kHz and 8 kHz thepensioner population requires a greater positive slope thanthe other two, again conforming to expectations based on
NINIMIMNIMEFIANINEINMEMIWtAmmmommmmamnim
MIMEranignelailMsisimssaurmEam
99-9
998
99 .5
99
m
95
90
FIGURE 3a, 3bCumulative frequency distributions ofrequired slopes for the three sub-popula-tions ag (a) 250 Hz/500 Hz and (b) 500 Hz/1 kHz.
Please turn the page for Figures 3c, 3d,and 3e representing frequencies to 8 kHzas follows : (c) 1 x*ozxHz, (d)exHzNkHz,and (e) 4 kHz/8 kHz : Pensioners (solid),Veterans (duexou) . Children (dot-dashed).
-10
10
20
30
40
50
SLOPE (dB/Oct)
999
99 8
995
99
90
m
m
w
w
30
m al EMI10 IM"
11111=1111z, asmomw
02-
-10
o
m
20
30
40
50
SLOPE (dB/Oct.)
°
"9 5mm"8 all=•11111111111MIM1111MME11"
momminammmummammmoreammus
99 11111111•11.11111allIlIM1111INIMMI111119e Ea kHz / 2 kHz 10.1 Maim
11111•11111•1 MIIIIM111-- -
9' WEE','o' =111111111=11=11.1WPAIIIII.1•11111111111111111•1•111=11
<
20
mImmommmammlimmmmmms
1 ° am arrAammmm=1111111111112111
° 2 ammommumumimmummmommommummmmmuNnoms-20
-10
0
10
20
30
40
50
SLOPE IdB/Oct
99 9
998 111166"MMMMI.E995
99
E11111•111.1•1111WIMENIIIIIIIIIII
"
2 kHz/4 kHz rIlIAMI111111111EMI
8° amsmalmomrimmomomim.ammmmmmmIrmummmomm.
'° immmizimmmmmMN= FA•11EIIIIIIIII.MMI
° 5° 2 ammommmomommmmmmmmummmmNummowlsommmummmmonmiummum
-35
-30
-20
10
SLOPE IdB/Ocf
20 30
35
999
998
995 EN99
954 kHz / 8 kHz MIMEN nMIIIIE
9° 111111WINIIIIIIIIIE11121MIUM1111100 MINN:n° mamIlmolmummlminn%mmmoam-- =11111111111111111111MIIIWAWAIII•MIINMNMNIIIIIIIIIIII II
cc'
4° Immmmmmorm. =minanualmmimmm20
s 1111111.1.11111
_33 35-30 _20 -10
SLOPE IdB/Oct
O 10 20 30
FIGURE 3c, 3d, and 3e
Cumulative frequency distributions of re-
quired slopes for the three sub-populations
at (c) 1 kHz/2 kHz, (d) 2 kHz to 4 kHz, and (e)
4 kHz to 8 kHz : Pensioners (solid), Veterans
(dashed), and Children (dot-dashed) .
°
-
~
~
Journal of Rehabilitation Research and Development Vol . 23Nu1 1888
the relative nature of presbycusic and noise-induced hear-
ing The second breakdown of the curves investigated thedifferent requirements for high- and low-gain aids . Re-sponses requiring more than 52 dB maximum gain at anyfrequency were classified as high-gain aids : the basis forthis classification is that this is the 50 percent point onthe maximum gain cumulative distribution curve. It wasfound that low- and high-gain aids require a similar rangeof response slopes within each octave . However, distribu-tion curves for the low-gain aids were slightly steeper in
°
each octave, indicating that high-gain aids need to havesomewhat more flexible response shaping.
The last breakdown examined the dependencies be-tween the slope in one octave and slope in adjacent octavesfor individual response curves . For example, if a particu-larly steeply sloping response is required in the 1 kHz to2 kHz region, then is it possible to narrow the range ofslopes required in adjacent octaves? Unfortunately, littleor no correlation was found between the slopes requiredin adjacent octaves . This means that a universal hearingaid would need one tone oon1rol adjustment for everyoctave over which the response is to be fitted . Further-
+ more, if the slope within each octave is quantized into6-dB steps, the total number of response options can becalculated to be 34,020 . Of course, that could be greatlyreduced by deciding to fit less than 100 percent of cases,or by planning to fit accurately only to 4 kHz instead of to0or8 kHz—for example, fitting 95 percent of the popula-tion° to 4 kHz requires "only" 1176 different responseshapes . The authors believe that the 6-dB step size usedabove is a reasonable one in that a useful range of slopes(such as 24 dB/octave) can be achieved in a practicalnumber of switch steps . We know of no research data thatallows us to predict how far from optimum a response canbe before the aid wearer is significantly adversely affected.
While the cumulative slope curves are extremely useful,they do not enable the shape of any individual responsecurves to be seen . For this reason, a clustering analysiswas performed to produce a more easily handled numberof representative response curves . So that differences inthe shapes of the curves could be readily observed, theresponses were normalized before clustering sothat eachhad 0 dB gain at 500 Hz.
Results for the three subpopulations, each grouped into10 clusters, are shown in Figure 4 . One striking feature ofthese responses is the shape in the 250 Hz to 500 Hzoctave . There is a clear requirement for most aids to havea flat or near-flat response in this region . The requirementis poorly met by most existing aids, which make use of
FIGURE 4Ten clusters for each vubpopux*Uon .wimgamsat o00*znormal-
ized mode .
6
DILLON and MACRAE: Performance Requirements
tone control circuits incorporating series capacitors .Thiscauses a response in which the generally steeper slopesabove 500 Hz are continued down indefinitely . It is pos-sible that including a low-frequency shelf in the responseof aids may resolve the present paradoxical situationwhereby decreasing the low-frequency cut-off of an aid'seoponaoinornonontheque!ityofnapnoduoUon--butde-creases intelligibility : see Punch, 1881 (2) . That is, vvi1halow-frequency shelf, it is likely that amplification can beextended down to very low frequencies at a level sufficientto be audible but insufficient to cause upward spread ofmasking . The requirement for near-equal gain at 250 Hzand 500 Hz is not peculiar to the Byrne and Tonisson selec-tion prooeduve ' i1vvi!!naau!t both from MCL-based proce-dures and from dynamic range bisection procedures,because neither hearing losses nor MCLs tend to varymuch between 250 Hz and 500 Hz for the majority of clients . bFurthermove, it doesn't matter whether or not the proce-dure includes a correction for the speech spectrum shape,because the average speech spectrum is also flat between250 Hz and 500 Hz, as is pointed out by Byrne, 1977 (3).
Indirect evidence of the need for an extended !ovvfre-quency mopon0000maafnomapoeohdiooriminationteotoconducted by Plant as reported in 1984(4) . In those studies,aided severely-hearing-impaired subjects showed poordiscrimination between nasal sounds and voiced stops,and poor discrimination of vowel duration (i .e ., long versusshort vowels) whenever the first formant was lower than400 Hz. Both of these types of errors are readily explainedif the low frequency components of the speech signal areinaudible or sufficiently low in level to be masked byhigher frequency components. Theimportance of!mwfre-quency oomponentoinaentencadiaoriminutionhmobeanshown by Speaks (5) . This also is understandable in termsof the suprasegmental information known to be carriedby low frequency components.
Overall, we feel fairly confident that an extension of thelow frequency response of hearing aids by use of a lowfrequency shelf will be benefioial to many hearing aidwearers . Even with present aids, low-gain aids oftenaohievea low frequency shelf at 0 dB gain when worn,because of transmission of low frequency sound in throughthe vent or around a loosely fitting mold . Unfortunately,this mechanism is of no use to those who require a lowfrequency gain of more than 0 dB.
In concluding this section, we would like to putalimita-tion on the data presented . The required responses andrange of slopes are clearly only as accurate as the Byrneand Tonisson selection procedure . The procedure oftenneeds to be modified when applied to profoundly im-paired individuals . Many such individuals are fitted withan aid with a fairly flat response up to about 1 kHz andmay not be able to make much use of amplification abovethat frequency . This type of response curve is not repre-sented in the present data . Validation studies (for lessthan profoundly impaired individuals) on the procedureare currently being undertaken by Byrne . Acomp!eieana!'
b MCL abbreviates "most comfortable loudness ."
ysis is not yet available, but a preliminary analysis of thedata indicates that some change to the procedure is likely.While overall required gain will probably continue to beestimated to vary at about half the rate of the hearing loss,the required range of slopes does not appear to be asgreat as that predicted by the present procedure . In par-ticular, the steep slope which the procedure invariablypredicts in the 500 Hz to 1 kHz octave is likely to be lessfor many clients.
Saturated Sound Pressure Level
In comparison with the selection of gain and frequencyresponse, the selection of the optimum maximum poweroutput (MPO) or saturated sound pressure level (SSPL) ofa hearing aid for each individual client has received verylittle attention, either in the research literature or in hear-ing aid fitting practice. Yet it is quite clear that the SSPLcurve of an aid is very important . One has only to subtractthe gain of an aid from its SSPL to realize that it does nottake a very intense sound in the environment to drive someaids into saturation . Under this condition it seems pos-sible that the shape and level of the SSPL curve could bean even more important attribute of the aid than the gaincurve.
If the SSPL of an aid is excessively high for a particularclient, then one of several things can happen . First, theoutput level accompanying relatively intense input soundsmay be sufficiently high to cause the aid wearer discom-fort or even pain . Second, an excessively high SSPL curvemay cause damage to the person's remaining hearing.There have been documented cases where this hasoccurred, despite the fact that the person experienced nodiscomfort from the aid : see Hawkins, 1982 (6).
If the SSPL has been set too low, then the usable dynamicrange of hearing for that person has been unnecessarilyreduced. This will result in the removal of some amplitudecues from the speech signal, with possible loss of informa-tion and a decrease in the variety of sound levels to whichthe aid wearer is exposed. Further, if the aid does not in-corporate compression limiting, then distortion will occurevery time the aid saturates (which will be more often thannecessary) . For an individuel with a very restricted dy-namic range, a low SSPL setting may even cause speechin particular frequency regions to be always inaudible,irrespective of the input level or the volume-control position.
If one temporarily excludes the hearing damage con-straint, then how should SSPL be selected? In the absenceof any contrary evidence, we will make the assumptionthat at each frequency, SSPL should not exceed loudnessdiscomfort level (LDL) at that same frequency . It seemslikely that for aids with "peaky" SSPL curves, the peakSSPLvvi!! be able to exceed the LDL at that frequency,since for reasonably broadband input signals, outputlevels equal to the peak SSPL will never be reached . Forthe purposes of this report, it will be assumed that loud-ness discomfort intheprimary constraint onSSPLselection.
LDL's were obtained from 71 sensorineurally impairedchildren and from 45 adults with mixed losses : see Dillon
7
Journal of Rehabilitation Research and Development Vol . 23 No . 1 1986
!-
,
&Macrae, 1984 (7), for details of the LDL measurementprocedure used.
The data will be presented here in two ways . First, therelationship between LDL (and thus SSPL) and HTL willbe examined by pooling the data across subjects at eachfrequency. Second, the relationship between LDL and fre-quency will be examined for each subject.
Figure 5 shows the to1ol range of SSPL-gain combina-donoroquined a12kHz . SSPL has been equated to LDL,and gain has been calculated from HTL using the Byrneand Tonisson selection procedure . Cases have been in-cluded only if the gain at 2 kHz was equal to the peak gainrequired for that subject. (This means that none of the gainrange required to cover all the subjects' needs can beobtained by manipulation of the tone control, and so mustbe covered by some combination of fitter-adjusted gainand SSPL controls, user-operated volume control, and thefitting of different models .) Although some data points fallin the upper left-hand corner, it does not seem necessaryor even desirable to use very high power on a low-gain aid,even if the client can tolerate the output levels . We have(fairly arbitrarily) drawn a line across the upper left-handcorner using the criterion that, at each frequency, speechthat peaks 12 dB above the longterm average spectrum ofa 70-dB-overall signal should just fail to clip or limit in theaid . We are thus left with the six-sided region shown inFigure 5 as the solid line.
150
140
One may query whether aids in the bottom right handcorner of the gain-MPO plot need be fitted . That is, is ahigh-gain, low-SSPL aid ever required? We feel that theanswer is an unequivocal yes! Examination of the dataobtained from the hearing impaired children revealed that33 percent of the measured LDL-threshold pairs indicateda dynamic range of less than 15 dB. These people withvery restricted dynamic ranges, and usually with largelosses, require an aid that will be in saturation much ofthe time. The alternatives are to provide them with an aidthat guarantees that the signal at that frequency is ofteninaudible—or else is usually uncomfortable or even pain-ful . While it may be argued that people with large sensori-neural losses have reduced frequency and temporalresolution, and cannot extract all the information presentin a given frequency region, it does not seem fairto assumethey can extract no information . On the other hand, it canbe safely assumed that either an inaudible signal, or anaid rejected due to loudness discomfort, will lead to alack of aided benefit.
The goal of ensuring an audible but comfortable signalin all frequency regions can be further helped by examin-ing thoohapaofindividua!LDLvemua'fmquenoyoon1oum.Similarities and differences in the shapes of the contours'can best be appreciated if the LDLs are all normalized toone value at a particular frequency . Figure 6 shows suchdata for both the child and adult samples . For either set
FIGURE 5Required SSPL (assumed equal toclient LDL) versus required gain(calculated from the pure tonethreshold using the Byrne & Tonis-son selection formula) for combin-ed ommanuauunumaat cxHz.
X x
XX
X
30 40
50
70
Maximum Coupler Gain (dB)
8
DILLON and MACRAE: Performance Requirements
of data, there is a 40 dB spread in the curves by 2 kHz . IfSSPL is to be equated to LDL, then obviously the shapeof the SSPL curve has to be capable of being shaped bythe same amount . When one considers the large range ofLDL shapes, and the impracticability of fitting them withasingle SSPLshape, there appears to be a strong needfor controllable-shape SSPL curves. This need will bestrongest for medium- and high-power aids, since thepeople wearing these aids will normally have the mostreduced dynamic ranges.
The need for some type of dynamic range reductionsystem for many clients is unquestioned, as some of thedata in the previous section have shown, While peak clip-ping oanaohiove1hadeoirodmduoUon .it does so at theexpense of increased distortion and a consequent loss in
FIGURE 6LDL versus frequency for (a) child dataand (b) adult data . All LDL values have
been normalized to 0 dB at 250 Hz toillustrate the range of curve shapes
encountered .
intelligibility . Compression nyohamn, by virtue of gainreduction for higher level sounds, achieve the range re-duction with little perceptually relevant distortion andlittle consequent loss in intelligibility . The lack of un-desirable side effects is obtained only if the parametersof the compression system are appropriately chosen.
The most critical parameter is the recovery time, whichon the evidence available, should be somewhere in therange of 60-to-120 ms and probably nearer the bottomthan the top : see Walker and Dillon, 1982 (8) . Releasetime is defined as the time taken for the output to settleto within ±2 dB of its final value after a 25-dB step decreasein the input level . The lower limit is set by the need to pre-vent the gain from varying excessively during each voicingperiod and thus generating distortion products .The upper
30
20 -
10 f
aa-o 0
-10 ~
0
500
1000
2000 4000 6000
Frequency (Hz)
Frequency (Hz)
~
9Journal of Rehabilitation Research and Development Vol . 23 No . 1 1986
limit is set by the need to prevent gain reductions, whichoccur during intense vowels, from persisting during follow-ing weak consonants, thus reducing their intelligibility.
Selection of the optimum attack time seems less critical.Values too small will cause unnecessary gain reductionduring extremely brief noise impulses, which will thencause the signal immediately following to be less audible.Values too large will not provide sufficient protectionagainst high-level sounds with sudden onsets, unless theaid incorporates an additional limiting device . Attack time,defined similarly to release time but for a 25-dB increasein input level, should probably be in the range of 2-to-5 ms,but the evidence is not available to positively excludevalues anywhere between 0 .5 ms and 20 ms.
Compression systems can be categorized into threeuseful types : (i) compression limiting, (ii) whole-rangesyllabic compression, and (iii) slow-acting automaticvolume control . The first two have attack and recoverytimes short enough to allow the gain to vary significantlyduring a single syllable . The last two operate in the com-pression mode during moderate level or even low levelsounds, while the first one operates in the compressionmode only when the sound level is so intense that a gainreduction is necessary to prevent uncomfortably or danger-ously loud sounds from occurring . There is no audiologicalevidence that any one of the three is superior to the othersin terms of speech intelligibility or quality—there is alsono logical reason why all three types shouldn't be com-bined in a single aid. However, for various practical reasons,we recommend that compression limiting be the systemchosen. Its advantage over the slow-acting system is thatit inherently provides protection against brief intensesounds, thus removing the need to provide additionalcircuitry to perform that essential function . Its advantageover the whole-range compressor is that a fitting proce-dure is available . That is, the compression limiter is usedto select the aid SSPL which, in turn, is equated to theclient's LDL.
It should not be assumed that the more complex full-range compression systems will eventually prove to besuperior . Both types of compression remove some of thespeech cues that are associated with the waveform enve-lope. While these cues are not an important source ofinformation for normal hearers (compressed speech ishighly intelligible), they do provide important cues for themore severely impaired, as was reported by Risberg &Lubker in 1978 (9) . Whole-range compression has theadvantage that there is a large input range over which theintensity variations are only reduced rather than removed,but has the disadvantage that there is almost no inputlevel at which they are left untouched . Compression limit-ing has the advantage that low- and moderate-level inputsare left uncompressed, but the intensity variations areremoved almost completely from higher level inputs.Compression limiting also has the advantage that, becausethere is no gain increase for low level signals, the ampli-tude modulation of background noise during pauses inthe signal of interest will not be a problem. This effect be-comes more noticeable as the signal-to-noise ratio
FIGURE 7(a) Block diagram of an aid incorporating compression showingall possible locations of the controls; (b) simplified equivalent of(a) with blocks H1A and H1B combined.
deteriorates.Viewed somewhat differently, whole-range compression
systems, including slow-acting volume controls, may bebeneficial by keeping all speech at a near-optimum loud-ness level so that the speech can be kept near the peakof the hearing impaired person's performance-intensityfunction . However, all environmental sounds above the(low) compression threshold are brought up, or down, tothis level, possibly creating fatigue and a feeling of detach-ment from the true "noisiness" of the environment . Onlythe appropriate research will enable the superior systemto be chosen.
Ideally, the 1-0 curve should be completely flat in thelimiting region. The compression ratio should certainly begreater than about 5 so that outputs significantly abovethe chosen SSPL value are never obtained.
The properties of circuits employing compressionamplifiers are critically dependent on the relative loca-tions of the compressor amplifier, the point at which thevoltage is sensed to provide the compressor control volt-age, and of the rest of the amplifier subcircuits . Themonitored signal can, in principle, be fed either backwardor forward to the variable-gain amplifier . We will consideronly feedback arrangements because the only compres-sion amplifiers available for hearing aid use are configuredto work in that manner. A completely general circuit ar-rangement thus appears as in Figure 7(a) . However, sub-circuits have the same effect whether they are placed inblocks H 1A or H1B. Asimplifiedblockdiagram is thus shownin Figure 7(b) . The block H 1 includes the microphone andall amplifiers, filters, controls, etc ., before the feedbackpoint, while the block H 2 includes all amplifiers, filters,controls, etc . (as well as the earphone, tubing, and ear-mold) after the feedback point . It is then straightforwardto show that the gain of the aid in the linear region is equalto H 1 H 2 and that the SSPL is equal to kH2IG. The compres-sion threshold referred to the input is simply equal to theSSPL divided by the gain in the linear region : the compres-
H1BH1A 00/P
(ai
H 00/P
0I/P
(b)
10
DILLON and MACRAE : Performance Requirements
sion threshold referred to the input is thus not an inde-pendent design choice . This approach bypasses theargument about whetherthe compression threshold shouldbe frequency dependent, and what the nature of thatdependency should be.
From the above discussion, it is clear that any compo-nent in block H 1 affects the gain, any component in block Gaffects the SSPL, and any component in block H 2 affectsboth gain and SSPL. From the fitter's point of view, it ismost convenient to have noninteracting controls, andtherefore the tone controls and fitter-adjusted gain controlshould, preferably, be in block H 1 . Similarly, control ofSSPL shape and of overall level should be in block H 2 .
The optimum location of the volume control has beenargued in some depth elsewhere : see Dillon and Walker,1983 (10) . There it was argued that, for compression-limit-ing devices, output-controlled compression (from the pointof view of volume control operation) is the necessary con-figuration . That is, the volume control should precede thefeedback point and so must also be in block H 1 .
Directional Response Characteristics
The problems caused to the hearing impaired by back-ground noise have long been noted . Directional micro-phones appear to be the only available device completelyworn by the aid user which can offer a partial alleviationof this problem. Hearing aids incorporating directionalmicrophones have now been available for more than 10years. Despite their objectively measured ability to im-prove the signal-to-noise ratio (SNR) when the signal andnoise are spatially separated, the majority of aids soldstill do not incorporate directional microphones . A reviewby Mueller published in 1981 (11) gives a good summaryof behavioral tests comparing directional and nondirec-tional aids . The majority of studies have shown that, innoisy situations, directional hearing aids offer superiorspeech intelligibility to the wearer. The remainder of thestudies have indicated no difference between the two aidtypes. The directional properties of hearing aids seem todisappear when the aids are measured electroacousticallyin reverberant environments, because the sound energy isincident from all directions, irrespective of the location ofthe signal source. However, in reverberant situations, in-telligibility is increased if the ratio of direct-to-reverberantsound can be increased. Provided the aid wearer isoriented toward the signal of interest, the directionalmicrophone will always emphasize the early-arriving directsound compared with the later-arriving reverberant sound.Directional hearing aids can thus be expected to offersuperior intelligibility even in reverberant environments,although the superiority may be less marked in highlyreverberant environments.
Obviously, not all directional aids exhibit the samedegree of directionality . Equally obviously, their benefitis greater, the greater their directionality . One commonbut fairly poor measure of directionality is the ratio of thefront sensitivity to back sensitivity . This is not a suitablemeasure because an angle of 180 degrees (from the front)may happen to coincide with a null in the polar response
curve or it may fall on the peak of a minor lobe . Measuressuch as the unidirectional index, directivity index, or"front to total random" ratio are better because they arenot sensitive to the exact location of the response nullsor minima.
Each of the methods of quantifying directivity involvesobtaining a ratio by comparing sensitivity in some speci-fied direction with the sensitivity averaged over anotherrange of directions. In this paper, these ratios will be ex-pressed in dB . The additions involved in taking theaverages are performed on a power law basis . The varia-tion of sensitivity with direction is obtained from a polardiagram which shows the sensitivity for every angle ofthe incoming wave, usually with the sound source speci-fied as lying in the same horizontal plane as the micro-phone. The definition of several useful measures ofsensitivity are as follows (with 0 degrees representing thedirection in front of the observer, and 90 degrees thedirection to the side of the observer on which the hearingaid is mounted):
Directivity index: sensitivity at 0 degrees compared withsensitivity averaged over all directions.
Unidirectional index: sensitivity averaged over the frontalsemicircle compared with sensitivity averaged over therear semicircle.
Maximum to average : sensitivity in the most sensitivedirection compared with sensitivity averaged over alldirections.
Front to total random : sensitivity averaged over frontsemicircle compared with sensitivity averaged over alldirections.
Front '/a to rear 3/a : sensitivity averaged over 0 degreesto 90 degrees compared with sensitivity averaged over90 degrees to 360 degrees.
These methods can be used to quantify the relativeadvantages of the three main contenders for "directional"hearing aids . The first contender is a nondirectional hear-ing aid placed on a head with the directionality impartedby head-baffle and head-shadow effects . Unfortunately,at 2 kHz, the sensitivity maximum occurs at about 70 de-grees from the front, and the frontal sensitivity is about5 dB down from this maximum . Hillman (12) pointed outin 1981 that this makes it difficult for the aid wearer tosimultaneously achieve the best signal-to-noise ratio andview the lips of the person speaking in a conversation.
The second contender is a conventional directionalhearing aid with parameters selected to give good on-the-head performance.
The third contender is some arrangement that makesuse of the person's own directional aid, the pinna . This isdone in in-the-canal aids (and to a lesser extent in in-the-ear aids) by locating a nondirectional microphone nearthe ear canal entrance . However, it is also possible totake advantage of the pinna with behind-the-ear aids ifthe microphone is set in the earmold rather than in theaid case. For all of these methods, the directionality isfrequency-dependent because, for a given microphone, itdepends on the relationship between signal wavelength,
11
Journal
and Development Vol . 23 No . 1x985
~
°
in!e1 port spacing, head size, and pinna size . Figure 8shows the directivity index for each of the three optionsat the octave frequencies from 250 Hz to 4 kHz. Data forthe unaided ear were obtained from Shaw, 1974 (13), andrefer to pressure at the eardrum . Up to 6 kHz, eardrumpressure and ear-canal entrance pressure both displaythe same directional dependency . Data for the two "hear-ing aids" were obtained by mounting a Knowles EB 1864microphone in a hearing aid case (with an effective portspacing of 20 mm) which was in turn mounted on KEMAR'shead.
Directionality for the non-aided ear and the non-direc-tional aidamaimi!arupto2kHzbeoauoebo1haneequaUyaffected by the head . By 4 kHz, the pinna dimensions aresignificant and the unaided ear receives a boost for fron-tally inoiden\ooundoandana1tenuationforooundmfnomthe rear, while the non-directional aid situated above thepinna receives no such benefit.
The directional aid is clearly superior to both theothers up to 2 kHz . By 4 kHz its directionality is no betterthan that of the non-directional aid (due largely to the low-pass filter in the rear port used to create the internal timedelay, and partly to the fact that the port spacing is anappreciable part of a wavelength).
Overall, these measurements show that up to at least2 kHz the directional aid can provide a SNR about 4-to-S dBbetter than either a non-directional aid or in-the-ear micro-phone placement . By 4 kHz, in-the-ear microphone place-ment iaeupo,ior' a!thuughihemau!ioahownn*fe/un!ytoplacement at the earcanal entrance . The benefit obvious-ly bmcomoo!oaoaomonaandmomofthanonoheiah!!odwith earmold and/or aid.
The value of the 4-to-5 dB improvement in SNR shouldnot be underestimated . Under noisy conditions, an im-pmvemeniofthiamuohcaninonaaomopemnhinteUig\bi!i1y
scores by as much as 20 percent . It is difficult to achievesuch an increase by any other means (e .g ., optimizationof frequency response or complex signal processing).
It is well known that localization in the vertical plane ismade possible by the spectral shaping caused by reflec-tions from, and coupling with, the convoluted folds of thepinna. Use of a directional microphone located above thepinna will destroy those cues, and the new set providedwill almost certainly not be as strong . However, the cuesfor vertical localization are high-frequency cues, certainlyabove 3kHz and probably above 6 kHz, and it is unfortunatethat, it is in the high-frequency region that present aidshave their poorest performance—and hearing lossestend to be greatest . The reduced fcequency-resolution ofhearing impaired people will probably make the use ofhigh-frequency spectral cues more difficult as well . Onbalance, it seems that, at present, most hearing impairedpeople are probably better served by adiroo1iona! micro-phone than by a non-directional microphone in the ear.
While some aids on the market allow the user to closethe rear port to remove the directionality, this does notseem to be a necessary option from the point of view ofachieving different types of directional responses, as thevast majority of clients simply leave the switch in theUineotionml position* . However, it does seem a worth-while fitter-operated option, as the aid's response canthen be measured in a standard test box or single-portmicrophone coupler while in its non-directional mode,and a standard correction applied to determine itsdirectional performance . Additionally, those people the
*Private communication : H . G . Mueller, Ph . D ., Supervisor Audi-ology soouion .wa!to,noodxrmymedica!Conmr, Washington,D .C.
~~~~ . . .~ .~~~~~ .~
250
u
500
~~ . .~~ .~~ . . .~~~
a
1K
2K
4K
Frequency (Hz)
FIGURE 8Directivity index for the unaided ear,nonmmonona!microphone, anuoa,m'o!ddimcoona! microphone . The lattertwo were mounted in a behind-the-earaid case on KEMAR.
-
12
DILLON and MACRAE: Performance Requirements
fitter believes would be better served by an extended flatlow-frequency response than by a directional aid couldbe fitted with the rear port closed off.
The optimum directional pattern (Le ., the shape of thepolar diagram) depends on where the signal is and wherethe noise is—and obviously these vary from moment tomoment . However, if one assumes that the noise is ran-domly inciden1fromaUdirochono,undthewontodoigna!only from directly in front, then the hypercardioid responseis optimal . Alternatively, if the wanted signal can comefrom anywhere in the front hemisphere, then the super-cardioid response is optimal . These terms, which aredefined with reference to their identifying equations inTable I, represent particular responses from an infinitefamily of responses all having the general sensitivityequation S=Sv(1+R cos 0) where the value R variesamong the members of the family.
However, these response shapes are obtained onlywhen the aid is measured in isolation in a free field ; whentheoid)oworn they anamodified substantially bythoaid-wearer's head . Madaffari in 1983 published polar responsesmeasured on KEMAR for aids with various values of R(14) .
TxBLe1Different polar response characteristics, their equations, andtheir identifying names.
Response Equation R
Omnidirectional So 0
Cardioid Go(1+0000) 1
Super-cardioid So (1 +noo0) 1 .73
Hype pnan1io!d So(1 +000a0) 3
Various directivity figures of merit have been abstractedfrom those responses (Fig . 9) . Some of the indices (suchas maximum sensitivity relative to average sensitivity)are little affected by the value of Ruhoaan . Ovond!, how-ever, there is a tendency for the indices to be maximizedfor responses near that of thooavdioid . It is thus recom-mended that a value of R of about 1 .0 be adopted, but asthe maximum is rather broad and occurs at differentvalues of R for different figures of merit, any value in the
r
6).,-.
.-4-
4
10
8
O0.33
2
v
,_ .IN^
,....
Front Semicircle re Average (Front fro "f ' .-a--b,random)
o"
ae
ax.t.]~
u
x
x
Index)""° ~- ~
~"~~~oV«"~-
~
Maximum re 4'
%
0 .44
0 .58
0 .76
1 .0
1 .32
1.73
Microphone Response Ratio , R
FIGURE 9Various directivity figures of merit plotted as a function of the variable oin the sensitivity equation . The
values of R are those applicable for the aid in isolation, while the actual directivity measurements are forthe aid mounted on KEMAR . A double set of values is present fo,n= 0 .76 because the four smallest values
of R were obtained with a 6 .3-mm port spacing and the four largest with a 14 .4-mm port spacing . All points
are averages of the results at 1, 1 .6, and 2 .5 kHz.
13
Journal of Rehabilitation Research and Development Vol . 23 No . 1 1986
0 ®
10
-10
FIGURE 10Just-acceptable noise relative to the sig-nal level for each of eight subjects . Re-sults are shown for five one-octave noisebands and one speech-shaped noiseband.
-20
-40 100
250
500
1k
2' k
Frequency (Hz)
4k "WideBand
range 0 .7 to 1 .4 must be considered satisfactory . In prac-tice various values of R are readily obtained because R isequal simply to the ratio of the external delay betweenmicrophone ports to the internal microphone delay in therear port path (14).
Maximum Internal Aid Noise
The work reported in this section was performed in anattempt to determine the level at which the internallygenerated noise in hearing aids becomes a problem . Noisein hearing aids is most usefully expressed as "equivalent"noise needed to be applied at the input in order to producethe same output noise. The equivalent input noise is mostfully specified if it is measured on a band-by-band basis.That is, the output noise is measured in separate (usually1/3-octave) frequency bands, and from the measurementsis subtracted the gain at the center frequency of eachband. The result is a complete Y3-octave spectrum of theaid noise referred to the input . This method obviously givesmore complete information about the aid's noise, but howcan it be used to arrive at a subjectively determinedcriterion?
If we were to know the spectrum (and level) of the signalinput to the aid, the equivalent input noise spectrum couldbe used to determine the SNR (at each frequency) receivedby the aid user . Fortunately, we are in a position to esti-mate the average input signal, since it will have the sameshape as the long-term spectrum of speech : e .g., seeByrne, 1977 (3) . (This estimate assumes that the primaryuse of hearing aids is for speech communication .)
Having now characterized the typical input signal, weneed to determine a criterion SNR at which the aid will bedeemed "just acceptable ." To find this, an experimentwas conducted in which eight normal-hearing subjectswere asked to adjust the level of a wide-band and variousnarrow-band noise signals . The wide-band noise wasshaped to have the same longterm rms spectrum as the
speech, and the narrow-band noise was filtered from thisby using octave-wide filters at 250, 500, 1000, 2000, and4000 Hz. A tracking procedure was used in which thebackground noise level was controlled by the subject viaa recording attenuator . The subjects were instructed thatthey were to press the button whenever the noise becameobjectionable for longterm listening and to release itwhen the noise was no longer objectionable. All subjectswere experienced at the tracking procedure, and all wereaware that the purpose of the experiment was to establisha criterion for the maximum allowable internal aid noise.The subjects, with one exception, were tested twice foreach of the six noise stimuli.
Noise criterion test results—From a comparison ofthe test and retest data (for the seven subjects testedtwice), it was quite obvious that all except one were ableto establish a noise criterion and stick to it . Test and retestfor each stimulus were separated by about 15 minutesbut the standard deviation of the test-minus-retest differ-ences, averaged across six subjects and all six noisestimuli, was only 2 .4 dB. This is even lower than that ex-pected for threshold . The seventh subject had an averagetest-retest standard deviation of 11 .5 dB, but as his over-all results were very close to the average results of therest of the group, his data were retained . The results foreach subject, expressed as the just-acceptable noiserelative to the signal, are shown in Figure 10 . Despite thethe internal consistency of each subject, there are reason-able differences among the SNRs deemed acceptable forthe various subjects . The results for one subject are ob-viously quite different from the rest ; these were excludedfrom further consideration . The remaining subjects wereall in agreement that a poorer SNR can be tolerated at thelowest and highest frequencies . A SNR of about 30 dB isneeded to satisfy most of the subjects at the 3 mid-fre-quencies, with the value falling to 18 dB at 250 Hz and to25 dB at 4 kHz . This is represented in Figure 10 by the
14
DILLON and MACRAE: Performance Requirements
heavy line with its dotted extrapolations to lower andhigher frequencies.
The final criterion input noise can thus be calculatedby subtracting the SNR given in Figure 10 from the long-term average speech spectrum with an overall level of 65 dBSPL The criterion equivalent input noise (in ½-octaveband levels) is shown in Figure 11 . It is equivalent to anA-weighted noise !oval of 33.7 dB, but not all noise spectrawith anA+woighiad noise level of337dBwill meet this cri-terion! (TheA+woighied !evel of a sound is the overall levelwith the low and high frequency components attenuatedto compensate for the reduced sensitivity of the normalhuman ear to low and high frequency sounds .) For interest,the ½*o1aveaquiva!ent noise levels of the no/mal humanear are shown in Figure 11 as the dotted line . This curvewas calculated by Killion (1976) by considering the normalthreshold in quiet to be a masked threshold, with the masking
provided by the equivalent input noise of the ear (15).
The criterion outlined above is strictly valid only if thenoise present is completely contained within a singleoctave band, because that is the experimental conditionunder which it was obtained . We should thus ask whetherthe criterion SNR in a given frequency region depends onthe amount of noise present in other frequency regions.Examination of the data for each subject showed that thejust-acceptable !nvel ofthm500 Hz and 1000 Hz bandswas not greatly affected by whether they were presentedin isolation or as part of the wideband noise . Averagedacross subjects, a noise level only 1 dB higher could betolerated when either of these bands was presentedalone than when they were presented as part of a wide-band noise . From this it follows that the criterion noiselevel (or SNR) previously suggested is not strongly influ-enced by the epoutnul shape of the noise under
Frequency (Hz)
FIGURE 11Maximum equivalent noise (measured in*oota,* levels) deemed to be accept-able when listening to a speech signalwith a longterm rms level of 65 dB soundpressure level (curve), and equivalent in-put noise of the normal ear (circles).
FIGURE 12Hypothetical examples of an objection-ably noisy aid (A) with noise rating of+1n go . and aqu!o\ aid (B) with a no serating of —ode.
1k
Frequency (Hz)
0125 250 500 4k2k 8 h
10
0125 250 500 1k 2k 4 h 8k
~
~
^
+
=
°
15
Journal of Rehabilitation Research and Development Vol . 23 No. //S80
oonnid*nakm.The result suggests that the overall objectionability of
an aid's inkernal noise can be assessed by noting thegreatest amount by which the noise in any octave (or one-third octave) exceeds the criterion noise in the respectiveoctave (or third of an octave). Similarly, a negative figure willindicate how far short of objectionable a particular equiv-alent inpui nniae in . Figura 12givea1womxump!an : oid ^A^
would be given a noiaooNootionebi!iiy rating of +1U dBand so would probably be considered unsuitable, whileaid "B" would be given a rating of -3 dB and be consid-ered sufficiently quiet.
The criterion can be relaxed somewhat in the low-frequency region whenever an effectively vented ear-mold is to be used. If the sound at the eardrum is domi-nated by vent-transmitted sound, then noise originatingin the aid is clearly of little consequence . In fact, for everyd*oibel by which the combined transmission gain (aidplus vent) lies above the aid transmission gain alone (asmodified by the vent, but not including sound transmittedin through the vent), one decibel can be added to the cri-terion input signal level . As the mold characteristics areknown only when individual clients are fitted, a practicalapproximation for the purposes of aid type testing wouldbe to relax the criterion by 1 dB for every dB by which theaid gain falls below 0 dB transmission gain when meas-ured in an artificial ear (or below -4 dB when measuredin a2cocoup!er) . This relaxed criterion will be valid onlyprovided the aid is fitted with a suitably vented earmold.
Existing electret hearing-aid microphones (except forthose with sharply rising responses achieved by acousti-cal means, and directional microphones with very smallport spacings) are just capable of meeting the noisecriterion proposed here.
Conclusions
Selected aspects of hearing aid electroacoustic per-formanoohave been examined . An analysis of the rangeof slopes required in aid frequency response (based onthe Byrne and Tonisson selection procedure) indicatedthat most hearing aid candidates require as much gain at250 Hz as they do at 500 Hz . Very few existing aids meetthis requirement . Inspection of the range of LDL curveshapes measured on hearing impaired people revealsthat, if hearing aid SSPL curves are to be matched to LDLon an individual basis, then adjustable-shape SSPL curvesare required in hearing aids. This can be achieved in asingle-band aid employing compression limiting . Theneed for directional characteristics in hearing aid micro-phones seems well documented, and available data indi-oaietha\the response should be approximately cardi-oid VVheD\hS aid is measured in isolation if it is to haveop1imal performance when mounted on the head . Leat,acriterion for the maximum tolerable internal noise ofhearing aids has been proposed based on experimentaldata obtained with normal hearing people . n
REFERENCES
1. Byrne, D, Tonisson W : Selecting thegain of hearing aids for persons withsensorineural hearing impairments.Srandxuuio!5o1-59 .1870.
2. Punch JL: Subjective approaches tohearing aid evaluation . Hearing In-struments 32(11)1 2-14, 65, 1981.
3. Byrne D : The speech spectrum—some aspects of its significance forhearing aid selection and evaluation.Brit J Audio! 11 :40-46, 1977.
4. Plant GL: A diagnostic speech testfor severely and profoundly hearing-impaired children . Aust J Aumo!6 :1 -9 .1S84.
5. Speaks C : Intelligibility of filteredsynthetic sentences . J Speech Hearneo10:oOe-nV8 1967.
6. Hawkins DB : Overamplification—awell-documented case report. JSpeech Hear Disord 47 :382-384,1982.
7. Dillon H, Macrae JH : PerformanceSpecifications for Hearing Aids . Nowpublished as : Dillon H, Macrae JH:Performance Specifications vo Hear-ing
Labora-tories <wAVu/Australia, 1984.
8. Walker G, Dillon H : Compression inHearing Aids—an Analysis, a Review,and some Recommendations . NALReport 90 . Sydney, Australia : NationalAcoustic Laboratories (NAL) of Aus-tralia . 1982.
o Risberg A, uuumorJL Prosody andspeech reading . Speech Transmis-sion Laboratory Quarterly Progressand Status Report (Q .P .S .R .) 4 :1-15,1978.
10. Dillon H, Walker G : Compression-input or -output control? Hearing In-struments a4(9):2nu2, 42, 1983.
11. Mueller HG : Directional microphonehearings aids—a 10-year report . Hear-ing Instruments %2(11} .1881.
12 . Hillman NS: Directional hearing aidcapabilities . Hearing Instruments
1 .198n.13. Shaw EAG: transformation of sound
pressure level from the free field tothe eardrum in the horizontal plane.J Acoust aocxmmo184o-180t1874.
14. Madaffari PL : Directional MatrixTechnical Bulletin . Industrial Re-search Products, Chicago, Illinois,Report No . 10554-1, 1983.
15. Killion MC: Noise of ears and micro-phones. J Acoust Soc Am 59 :424-433, 1976 .
