hearing aid selection for the 1990s: where to? · hearing aid selection for the 1990s: where to?...

J Am Acad Audiol 7 : 377-395 (1996)

Hearing Aid Selection for the 1990s: Where to? Denis Byrne*

Abstract

This paper discusses trends and issues in selecting amplification for new, mainly nonlinear, types of hearing aids . Some concerns are that nonvalidated prescription formulae may pro-liferate ; that, through concentration on loudness normalization, alternative rationales are not being evaluated; and that most new fitting procedures for nonlinear hearing aids ignore the findings of research with linear amplification . A possible philosophical concern is that the hearing aid fitter may lose control of the fitting process. The achieving of any particular loud-ness relationship, among different frequency components of amplified speech, may be unimportant in itself but may be critical for achieving other objectives such as maximizing audibility over a wide frequency range when speech is at a comfortable level. Amplification selection should seek to achieve objectives by the simplest means (linear frequency shap-ing, automatic volume control) and should proceed to more complex, but potentially detrimental, processing (syllabic compression) only to the extent necessary. New prescriptive procedures should build on established procedures because, for typical speech levels, amplification requirements should be similar, regardless of whether linear or nonlinear amplification is used .

Key Words: Compression amplification, hearing aid fitting, loudness normalization, nonlinear amplification

D

etermining amplification requirements for people with hearing impairments is vital for the design and fitting of

hearing aids . Although substantial advances have been made, the challenge is as great as ever because new hearing aid technology has greatly increased amplification options. Considerable effort is being directed towards the develop-ment of new hearing aid fitting procedures and systems, as well as new types of hearing aids . I shall argue, however, that some current direc-tions give cause for concern and that there is a need for some reorientation of thinking and research effort.

This paper discusses the broad philosophi-cal and theoretical issues relating to the fitting of new types of hearing aids . The discussion will be preceded by outlining the history and cur-rent trends in hearing aid selection, the major characteristics of new types of hearing aids and their implications for fitting practices, and some

`National Acoustic Laboratories, New South Wales, Australia

Reprint requests : Denis Byrne, National Acoustic Lab-oratories, 126 Greville St ., Chatswood 2067, NSW, Australia

current design and fitting rationales for non-linear amplification. I shall conclude by sum-marizing some concerns about current directions and suggesting some other possible directions and approaches . This paper challenges certain dominant ideas that tend to be taken to be self-evident, particularly the uncritical acceptance of loudness normalization as the only prescrip-tive rationale worthy of consideration. I shall argue that new fitting procedures should build on current knowledge and procedures and that a variety of fitting rationales should be evalu-ated with a view to developing validated pre-scriptive procedures .

TRENDS IN HEARING AID PRESCRIPTION

Present Situation

Over the past 20 or so years, the subject of hearing aid selection has received considerable attention, following a long period of comparative neglect (Studebaker, 1982). This interest has included the development of several new for-mulae for prescribing the gain and frequency response of a hearing aid (Crouch and Pendry,

377

Journal of the American Academy of Audiology/Volume 7, Number 6, December 1996

1975a, b; Byrne and Tonisson, 1976; Berger et al, 1977 ; Cox, 1983, 1988 ; McCandless and Lyregaard, 1983; Seewald et al, 1985 ; Byrne and Dillon, 1986 ; Schwartz et al, 1988 ; Skinner, 1988 ; Byrne et al, 1991). Such formulae have been supported by varying degrees of valida-tion research (reviewed in Byrne, 1983, 1993 ; Skinner, 1988). The selection of maximum power output (MPO) has received less attention. Nonetheless, useful recommendations, together with a limited amount of validation evidence, are available (Dillon et al, 1984, 1994 ; Hawkins et al, 1987).

Recently, there has been a renewed interest in procedures for evaluating hearing aid fit-tings. These include paired-comparison judg-ments, or magnitude estimations, of speech quality or intelligibility (e.g., Punch and Parker, 1981 ; Studebaker et al, 1982 ; Levitt et al, 1987; Purdy and Pavlovic, 1992) and other techniques such as cued listening (Jerger et al, 1994). In some research (Neuman et al, 1987; Byrne and Cotton, 1988), hearing aid selection has been achieved by comparing frequency response char-acteristics representing systematic variations from a response derived from a prescriptive pro-cedure . Clinical procedures along these lines have also been suggested (Byrne and Cotton, 1988 ; Kuk, 1994).

At present, no single procedure is universally accepted as the best way to select a hearing aid. Nonetheless, most audiologists would agree that, as a result of substantial research from many laboratories, scientific bases exist for selecting gain, frequency response, and MPO by using one of the better-validated prescriptive procedures or a combination of such a proce-dure with an evaluation method. It is, therefore, disconcerting to find that methods proposed for fitting nonlinear hearing aids (discussed later) mostly ignore the principles established for selecting linear amplification . Indeed, there seems to be an increasingly common view (e.g., Brunved, 1994) that current hearing aid selec-tion methods have no application to new types of hearing aids .

hearing aid parameters . This is not necessarily the way that all manufacturers would wish to go but it may be unavoidable if scientists and clinicians fail to develop properly researched selection methods that are applicable to new types of hearing aids. The scenario described raises concerns, both scientific and philosophi-cal. Scientifically, the concern is that amplifi-cation may become prescribed by a wide variety of proprietary formulae of which few, if any, are validated by published research . A possible philo-sophical problem is that control of the fitting process is taken away from the fitter who is responsible for care of the client . This concern has already been expressed in relation to the common practice of "selecting" custom in-the-ear aids by simply sending an audiogram to the manufacturer (Bratt and Sammeth, 1991). It may be argued that it does not matter who con-trols the fitting process, provided that it is done effectively. However, at the least, the fitter should understand the fitting procedure and be assured that it is appropriate for his or her client. Another concern is that the choice of a particular type of hearing aid would oblige the fitter to use that manufacturer's fitting formula.

The trend towards manufacturer-specific fitting formulae is not universal. Some general prescriptive formulae, applicable to nonlinear as well as linear amplification, have been sug-gested, of which the IHAFF protocol (Cox, 1995) is a good example. Such formulae, however, have not been validated and most are based on the same rationale, loudness normalization, which is acknowledged to be an "intuitive, though as yet unproven, notion" (Mueller, 1995).

In view of the above concerns, it seems desir-able to examine the hearing aid selection and fit-ting process in terms of how it applies to current and future hearing aids . What are the major issues that need to be resolved in order to develop validated fitting procedures? To what extent are existing methods applicable (i.e., do we really need to start all over again)? What are appro-priate roles for researchers, manufacturers, and clinicians in the process?

The Future?

What, then, is the future for hearing aid selection? One possible scenario is that virtually all new types of hearing aids will be fitted by using a manufacturer-specific fitting system and that such a system will include proprietary formulae for calculating amplification require-ments, as well as methods for adjusting the

NEED FOR NEW HEARING AID FITTING PROCEDURES

N ew hearing aia fitting procedures are needed because new types of amplification have recently become available. We need to decide what type of amplification system is best for an individual and, for any particular type, we may need to prescribe values for the various

378

Hearing Aid Selection / Byrne

amplification parameters . The major complica-tion is that most recently developed hearing aids are nonlinear amplification systems, of which there are several types. A smaller com-plication is that many current aids provide mul-tiple memories that enable the user to select different types of amplification. Both of these developments stem from the belief that there is no single set of amplification characteristics (e.g ., combination of overall gain, frequency response, compression characteristics, and MPO) that will be optimal for an individual at all times. Nonlinear systems vary amplification depending on changes in acoustic input. Multi-ple memory aids can provide variations, in lin-ear or nonlinear amplification, that may be chosen either to accommodate changes in acoustic input or changing preferences in sound quality. Multiple memories complicate hearing aid selection in that two or more sets of ampli-fication characteristics need to be selected . Such selection has been discussed by Keidser et al (1995, 1996). Dillon (1996) has reviewed the various types of nonlinear amplification sys-tems and the evidence evaluating each type . Such a review will not be attempted here but it is necessary to outline the major characteristics of compression systems in order to indicate their implications for hearing aid selection issues .

Characteristics of Nonlinear Amplification Systems

Figure 1 illustrates three variations in non-linear amplification expressed as input/output (I/O) functions. In Panel a, compression is used simply for limiting. In addition to gain and fre-quency response, the required descriptive para-meters are the kneepoint at which limiting

CR LI~

5 0

(a)

CR um CR Li.

(6) (c)

INPUT INPUT INPUT

Figure 1 Input/output functions (schematic represen-tations) of three types o£ nonlinear amplification sys-tems : (a) compression limiting, (b) whole-range compression, (c) low-level compression. K = knee point, CR = compression ratio, lim = limiting, WR = wide range, Lo = low level, Hi = high level.

begins (K lim) and the compression ratio (CR) of the function above K lim, CR lim.

Panel b illustrates whole range compres-sion . Starting from a low input level, gain decreases as input increases, with the same CR until the limiting level is reached. If the I/O function is independent of frequency, the required descriptive parameters are gain and frequency response for a typical input level (e .g ., 70 dB SPL), the low kneepoint (K Lo), CR-wide range (CR WR), K lim, and CR lim. If the I/O function varies with frequency, the hearing aid's perfor-mance can be described either by obtaining VO functions at different frequencies or by obtain-ing multiple frequency responses with differ-ent input levels . (Here, and later, it is assumed that frequency response is measured with a sta-tionary broadband signal . However, such mea-surements may not accurately indicate the effect of compression on the spectra of speech signals that have a spectral shape that varies with time .)

Panel c shows a system that may be desig-nated as low level compression (Dillon, 1996). Compression occurs over most of the input range but amplification becomes linear (and may pro-vide zero gain) for high-level inputs . It now becomes necessary to specify three CRs (of which CR Hi may be unity) and three kneepoints . A fur-ther complication occurs with compression sys-tems that produce a curvilinear I/O function . Although these may require complex equations to be described precisely, they may be approxi-mated by straight-line VO functions with one or more kneepoints . The kneepoints of compres-sion systems can be specified as either input or output levels . Logically, the choice should depend on fitting rationale; output level would be appro-priate for a kneepoint prescription derived from a hearing loss characteristic, whereas input level would be appropriate if the kneepoint is not pre-scribed. For example, the kneepoint for limiting should be specified as an output level if it is intended to correspond to the individual's loud-ness discomfort level (LDL). On the other hand, a kneepoint could be specified as an input level if compression is intended for noise reduction and is not varied for individuals .

Compression systems may also be classi-fied on the basis of whether their time con-stants, attack and release times, are relatively fast or slow. Fast-acting systems (release time less than 150 msec) are usually called "syllabic" compression, whereas slow-acting systems are called automatic volume control (AVC) (Walker and Dillon, 1982). In most systems, the time

379


constants cannot be varied by the hearing aid fit-ter: they are either fixed or adaptive, that is, the release time varies automatically depending on durational characteristics of the input. How-ever, the facility for the hearing aid fitter to vary time constants can be provided and, there-fore, these characteristics are potentially pre-scription parameters .

Multiband compression systems will almost certainly require different characteristics for each band as the usual rationale for multiband systems is to compensate for frequency-depen-dent recruitment.

Dillon (1996) has proposed a three-way clas-sification for describing compression systems. They may be classified according to whether (a) they provide most compression for low- or high-level inputs, (b) the compression has most effect on low-frequency or high-frequency sounds, and (c) whether the time constants are short or long . Examples exist of nearly all possible combina-tions and some systems include two or more types of compression.

Implications of Nonlinear Amplification for Hearing Aid Selection

The hearing aid selection and fitting process has been conceptualized as having four stages (Byrne, 1979). The first stage is prescribing amplification characteristics to suit the client . The second stage is matching the prescription by finding a suitable combination of hearing aid, settings, and attachments. The third stage is verifying that the required amplification has been provided in the ear of the individual . The fourth stage is evaluating how the client hears with the fitted hearing aid and, where indi-cated, readjusting ("finetuning") the hearing aid fitting.

The use of nonlinear amplification (and/or multiple memories) complicates all stages of the process. The verification and evaluation stages require a greater number of test signals and various problems may be encountered in select-ing appropriate signals for the system concerned. The matching stage has been affected by the advent of programmable aids in that it may be a matter of programming the appropriate settings into a single, flexible aid rather than choosing among different instruments. (The audiologist should not, however, overlook the possible value of acoustic modifications rather than relying exclusively on electronic variations .)

Although the use of nonlinear hearing aids affects all stages of the selection process, its

implications are greatest for prescription . This may involve three steps. The first is determin-ing what type of compression system is best for the individual . (If multiple memories are avail-able, it may be appropriate to select two or more types.) This step presupposes that no single compression system is best for all people . Whether that is true remains to be established, although there is some indication that different systems may be best under different conditions (Keidser, 1995) and that some systems are most suitable for certain degrees or types of hearing loss (Killion, 1993).

The second step involves determining which parameters of the chosen system need to be indi-vidually selected. This issue, which cannot be resolved until the validity of different prescrip-tion rationales has been tested, will be discussed later. The third step is prescribing values for the prescriptive parameters . Possible prescrip-tion parameters are outlined in the following discussion and are summarized in Table 1.

Gain and frequency response are individual prescription parameters. They can be varied extensively, are currently prescribed, and have strong theoretical and validated rationales . The use of compression should not change gain and frequency response requirements for a typical input level (e .g ., speech with an overall rms level of 65-70 dB SPL) .

K lim is an individual prescription para-meter. It is equivalent to MPO and the same pre-scription principles apply. Nearly all aids provide

Table 1 Possible Prescriptive Parameters for Nonlinear Hearing Aids

Need for Measurement Amplification Parameter Prescription Requirements

Gain Yes HTL or MCL Frequency response Yes HTL or MCL Maximum output or

limiting kneepoint (K lim) Yes LDL &/or HTL Compression ratio,

wide range (CR WR) Yes Dynamic range or loudness scaling

Low-level kneepoint (K Lo) Yes HTL? High-level kneepoint (K Hi) Yes HTL or

loudness scaling Attack time (Ta) Maybe ? ? Release time (Tr) Maybe ? ? Compression ratio,

limiting (CR lim) No Compression ratio,

high level (CR Hi) No

380


scope to vary this parameter, although usually not on a frequency by frequency basis. Good rationales exist for prescribing MPO or K lim.

CR WR is usually treated as a prescription parameter although an exception may be when compression is regarded purely as a noise reduc-tion system (see below) . Prescription rationales (discussed later) are based broadly on the idea of compensating for a hearing-impaired per-son's reduced dynamic range.

In some hearing aids, "K Lo" (compression threshold) is variable . The undesirability of applying compression to sounds that are inaudi-ble could lead to prescriptive formulae for appro-priate combinations of gain and compression threshold. Furthermore, Killion (1993) has sug-gested that compression threshold may some-times need to be set at a higher level than is usual for the K-Amp, to avoid complaints of cir-cuit noise.

K Hi is intended to represent the level at which recruitment is complete or at which the loudness growth function asymptotes and assume a normal shape. As this level does not vary greatly for people with mild and moderate hearing losses (Lyregaard, 1988), it may be rea-sonable to use the same value for all such peo-ple. However, K Hi will certainly need to be varied if this type of system is used for people with severe hearing losses .

The choice of "Ta" and "Tr" define whether the compression system is of the syllabic orAVC type . They are not currently treated as pre-scription parameters . However, there may be possibilities for developing prescription ratio-nales based on individual differences in tempo-ral masking or on information concerning the listening environments most commonly encoun-tered by the hearing aid wearer.

For various reasons, the other parameters shown on Figure 1 are probably not prescription parameters . CR lim will always need to be high to serve its limiting function and there is no reason to think that it should be varied . CR Hi will always be unity (or close to unity) as the rationale of low-level compression systems is to provide linear amplification above the level at which the loudness growth function becomes normal.

DESIGN AND FITTING RATIONALES FOR NONLINEAR HEARING AIDS

T here are several rationales for using non-linear amplification. The simplest, which will not be discussed further, is using compression

to provide low-distortion limiting (Walker and Dillon, 1982). Four additional rationales are to reduce noise, to improve audibility for soft sounds, to normalize loudness perceptions, and to minimize the need for volume control adjust-ments. These four key concepts are illustrated and discussed by presenting an overview of some system designs and fitting procedures. How-ever, it is not intended to describe fully or to clas-sify particular devices or fitting procedures . In fact, any such classification would be arbitrary as some devices and procedures incorporate two or more of the concepts .

Noise Reduction

The earliest examples of the new wave of nonlinear hearing aids were designed as noise reduction systems . For low-level inputs, these aids provide their broadest and flattest fre-quency response while, for higher inputs, the bass response is progressively cut (i .e ., the response has more high-frequency emphasis). The rationale is that high inputs are presumed to include high levels of noise and many noises are predominantly low frequencies. Reducing the low-frequency amplification will reduce the noise but also the low-frequency components of speech. Nonetheless, there could theoretically be some advantage in speech intelligibility through a reduction in upward spread of masking, and/or some advantages in listening comfort and decreased distortion because the reduced gain for high-intensity low-frequency inputs would reduce the likelihood of saturation .

A recent variation of this type of system (i .e ., one in which compression affects mainly the low frequencies) is the MultiFocus hearing aid (Brunved, 1994). This is a two-band hearing aid with wide range (low-level) compression in the low-frequency band only. The compression ratio is varied in a manner designed to compensate for the average loudness growth functions of persons with various degrees of sensorineural hearing loss . Thus, with respect to the low-frequency band, amplification is prescribed to compensate for the individual's abnormal loud-ness growth function, as predicted from hearing level. The high-frequency band has compres-sion limiting only (although capable of being set at a relatively low level) because it is reported that the use of dynamic range compression in both bands resulted in reduced speech intelli-gibility in noise, presumed to arise from a reduc-tion in spectral contrasts .

381


Improving Audibility

Most nonlinear hearing aid fitting proce-dures or design rationales include the concept of normalizing loudness . Some rationales also emphasize using compression to improve audi-bility for low-intensity speech sounds . Such ratio-nales may be distinguished from others that are unconcerned with audibility or that simply assume (wrongly) that normalizing loudness will ensure optimal audibility. (Normalization does not, in itself, optimize audibility because some components of speech may be inaudible when the speech is soft or when masking occurs.) A ratio-nale that emphasizes audibility, as well as nor-malization, is that of the K-Amp amplifier in its level-dependent frequency response configuration (Killion, 1993). For sloping high-frequency hear-ing losses, this provides a treble boost, as well as more gain, for low-level inputs. The rationale is that, when voice levels are low, additional high-frequency amplification is needed to provide audibility of the low-intensity high-frequency components of speech . Very pertinently, Killion (1993, p. 60) points out that recruitment should be viewed as "an abnormal loss of sensitivity for quiet sounds ." In keeping with this view, the K-Amp provides the most gain for low-level inputs in order to restore audibility. (The fitting procedure, Fig. 6, proposed by Killion [1994] pre-scribes gain for low-level inputs to be equal to HTL-20 dB for hearing losses 20 to 60 dB HTL.) The gain decreases for higher level inputs up to about 90 dB SPL, where the system becomes linear and gain becomes zero (or relatively little) because people with mild and moderate sen-sorineural hearing losses are considered to have no loss (or little loss) of loudness for high-level sounds . The gain for mid-level inputs is based on the principle of gradually reducing the gain from what is required for low levels to that required for high levels (possibly zero) where the system becomes linear.

The complex compression system of Moore and colleagues (Laurence et al, 1983 ; Moore et al' 1985 ; Moore and Glasberg, 1986 ; 1988; Moore, 1987) may be considered under the audibility rationale, although it also has strong elements of the AVC, as well as the loudness normaliza-tion, rationales . This approach combines slow-acting (AVC) and fast-acting (syllabic) types of compression. An AGC amplifier at the front of the system is usually triggered by a moderate input (about 63 dB SPL) and operates with long time constants (several hundred msec) to reduce the variations in long-term speech levels . It can,

however, also be triggered by higher level sounds to operate as a fast-acting limiter. The signal is then split into two bands and is processed by a syllabic compressor to reduce the short-term variations in speech sounds . In the latest ver-sion of the system, the syllabic compression is applied to the high-frequency band only. It is reported that "the use of dual front-end AGC eliminates the need for compression in the low-frequency band" and that "this reduces spurious spectral and temporal distortions introduced by the syllabic compression" (Moore and Glasberg, 1988, p. 103) . The need for a two-band system is argued on the grounds that recruitment often varies across frequency and that "relatively weak high-frequency components in speech . . . are often accompanied by, or follow rapidly after, relatively intense low-frequency components . The use of fast-acting AGC in two or more sep-arate bands can ensure that these weak high-frequency components are always audible" (Moore, 1993, p. 152) . This conclusion contrasts with the MultiFocus rationale that speech intel-ligibility will be optimized with a system that uses wide-range compression in the low-fre-quency band only. The use of two or more types of compression complicates fitting considera-tions. With the Moore and Glasberg system, the AVC characteristics are the same for everyone but the syllabic compression is adjusted to suit individual hearing losses, partly on an empiri-cal basis rather than strictly according to a pre-scriptive procedure.

Loudness Normalization (Recruitment Compensation)

Loudness normalization is a prominent fea-ture of the K-amp rationale and, to a lesser extent, the Moore rationale. However, those rationales also emphasize audibility and it is a major consideration in their system designs. By contrast, there are some suggested fitting procedures in which loudness normalization is certainly the primary aim and it is presumed that normalizing loudness will ensure an opti-mal fitting.

An example is the loudness growth in half-octave bands (LGOB) method recommended for fitting the ReSound hearing aids . Loudness growth functions are measured with 1/2-octave bands of noise (Pluvinage, 1989) and the com-pression characteristics are selected to restore normal loudness . In practice, this involves select-ing the crossover frequency for a two-band

382

Hearing Aid Selection / Byme

compression system and prescribing the gain for 50 dB SPL and 80 dB SPL inputs for each band (Johnson et al, 1989). For most clients with slop-ing high-frequency losses, this aid would provide most compression in the high frequencies because a higher compression ratio would be required in the high band than in the low band to restore normal loudness .

For fitting a research system, Kollmeier et al (1993) used a recruitment compensation pro-cedure based on loudness scaling measurements with narrow bands of noise. The dynamic range is defined by measuring the levels required at each frequency to elicit loudness impressions of "very soft," "comfortable," and "very loud." Gain is provided to compensate for the differ-ences between the measured levels of a hearing-impaired subject and the corresponding levels for a normal-hearing person .

A recent example of loudness normaliza-tion is the IHAFF protocol . According to Mueller (1995), this procedure is based on the "notion that when hearing aid processing is matched to the patient's loudness growth function across fre-quencies, greatest user benefit will result" (p . 10). Cox (1995) states that "amplification should normalize the relationships between envi-ronmental sounds and loudness perception . This means that a sound that appears soft to a normal-hearing listener should be audible, but soft, after amplification, to the hearing-impaired person . Similarly, sounds that are com-fortable or loud to a normal-hearing listener should be comfortable or loud, respectively, after amplification.. ." (p . 10). The procedure is based on measuring loudness growth functions by determining the levels of pulsed warble tones that correspond to seven loudness categories . In principle, measurements could be made at sev-eral frequencies but, for clinical use, Cox suggests that two frequencies should be sufficient . The prescription formula calculates the gain required at each frequency to amplify soft, average, and loud speech to the levels required for them to be perceived normally. The formula was derived using data showing the relationship between loudness perceptions of warble tones and speech. The computerized selection procedure, VIOLA, uses the loudness scaling data to calculate tar-get levels, for two or more frequency bands, for soft, average, and loud speech inputs . These define an I/O function that can be compared with the 1/O functions of one or more hearing aids to determine their suitability. Cox stresses that the formula is provisional and may change after evaluation of its underlying assumptions.

Automatic Volume Control

A major element of some fitting rationales is that compression minimizes the need for vol-ume adjustments by the user and thereby makes aid usage easier and more comfortable. A notable example is the MultiFocus type of hearing aid, which does not have a volume control because it is considered unnecessary. The slow-acting AVC of the Moore system is also claimed to min-imize the need for volume adjustments. Similar statements have been made by some Australian Hearing Services' clinicians with regard to some single-band, whole-range compression hearing aids . Furthermore, after reviewing the evidence for different types of compression systems, Dillon (1996) has concluded that the main advantage demonstrated for compression, whatever type, is that it reduces the long-term level variations of speech and thereby tends to maintain audi-bility for soft speech, together with comfort for loud sounds, without much need for volume con-trol adjustments.

Evidence for Fitting Rationales

The evidence supporting the various types of compression amplification systems has been reviewed by Dillon (1996) . In that such evi-dence is, indirectly and in varying degrees, evi-dence for different fitting rationales, it will be briefly reviewed here . Several evaluation stud-ies of the older noise reduction systems have generally found little, if any, advantage over an optimally fitted linear system (Dillon and Lovegrove, 1993). The MultiFocus rationale is supported by a series of in-house studies in which field trials showed this type of amplifi-cation to be preferred to linear amplification or to a system with wide-range compression in both the low and high bands (Lundh, 1981, 1983 ; Lundh and Nielsen, 1992). Biering-Sorensen et al (1995) evaluated the MultiFocus hearing aids and concluded that, compared with linear hearing aids, they did not provide any bet-ter speech intelligibility in noise but were pre-ferred by the majority of clients. The nonlinear aid was rated as superior for listening in quiet, for brightness of sound, and for reducing the annoyance of traffic noise .

The various studies by Moore and colleagues (Laurence et al, 1983 ; Moore et al, 1985; Moore, 1987) have shown that their system can provide improved intelligibility for soft speech (when volume control adjustment is not permitted) and improved understanding of speech in noise.

383


Moore et al (1992) found similar, but much smaller, advantages for the ReSound hearing aid, which differs from the Moore system in various respects, notably that it does not have slow-acting compression preceding the syllabic com-pression . There seems to be no commercial hearing aid with slow-acting compression only but, in experiments with a master hearing aid, Neuman et al (1994) found slow-acting com-pression, with a CR of 1 .5 to 2, to be beneficial for listening in noise for people with small dynamic ranges .

Apart from the above-mentioned studies, the support for currently available compression systems seems to be mainly anecdotal or based on comparisons with the clients' own hearing aids, which are often of unspecified condition, quality, or suitability. Taken together, the vari-ous case study reports (e .g ., Killion and Villchur, 1993) provide general support for the value of compression but no conclusive evidence for the superiority of any specific compression system or fitting rationale.

CRITICAL FITTING ISSUES

0 n the available evidence, it is not possible to decide which of the various design and

fitting rationales is most valid. Yet, different rationales would have very different implica-tions for selecting amplification and for deter-mining what types of measurements and prescription formulae are necessary. Therefore, the most basic research need is a thorough eval-uation of fitting rationales before proceeding to develop any particular rationale into a fitting procedure. Such research should consider the fol-lowing issues .

Loudness Considerations

Virtually all hearing aid selection rationales are concerned with loudness in that it is gener-ally agreed that speech should be amplified to audible and comfortable levels and that no sounds should reach uncomfortable levels . Fur-thermore, many procedures, either explicitly or implicitly, aim to deliver the various frequency bands of speech with a particular loudness rela-tionship (e.g., Byrne and Tonisson, 1976; Cox, 1983, 1988 ; Seewald et al, 1985 ; Byrne and Dil-lon, 1986 ; Tyler, 1986 ; Skinner, 1988). Consid-eration of loudness relationships has received even greater prominence in procedures for fit-ting nonlinear hearing aids (e.g., Cox, 1995). It is, therefore, pertinent to ask: How important is

it to amplify speech to provide some specified loudness relationship between different fre-quency bands? I shall argue that, in one sense, it is critically important but that, in another sense, it may not be important at all.

Unimportance of Loudness Relationships

Van Tasell (1993) has observed that "there is no conclusive evidence that loudness rela-tionships among the various frequency regions of speech are important for their intelligibility, as long as the regions are sufficiently audible and not uncomfortably loud" (p . 237) . In the present context, the relationship of most interest is loud-ness normalization, as this concept is a major ele-ment of most fitting rationales for nonlinear hearing aids . Not only is there no evidence that loudness normalization will optimize speech intelligibility or quality, but there are argu-ments why it may be unimportant or even detrimental under some circumstances. Nor-mal-hearing people experience a wide range of loudness levels depending on the intensity of sound at the source, its distance from the listener, and various acoustic conditions . The long-term spectrum of speech, and the short-term spectra of particular sounds, will also vary consider-ably for different talkers and depending on reflection and baffle effects . Variations in loud-ness and spectra do not cause any problems provided that they remain within the range that is both audible and comfortable . Why then should it matter whether a hearing-impaired person experiences exactly the same loudness sensation that a normal-hearing person would in the same situation? Research has confirmed that a wide range of speech spectra can provide equivalent speech intelligibility (Horwitz et al, 1991 ; Van Buuren et al, 1995).

Inherent in the loudness normalization rationale is the assumption that normal is best . Yet, under some difficult listening conditions, such as in many moderate to high noise levels, speech intelligibility would be improved for normal-hearing people by providing amplifica-tion with high-frequency emphasis. If we accept the unprovable, philosophical position that what is normal must be best, we might surmise that the unimpaired human auditory system pro-vides loudness sensations that constitute the best compromise among somewhat conflicting requirements of auditory functioning that include understanding speech, enjoying the qual-ity of sounds, and detecting and locating sounds . However, such a compromise might not be

384

optimal for a hearing-impaired listener who has difficulty understanding speech in a greater range of situations than a normal-hearing lis-tener. It may well be that something different from normal loudness sensations will provide the best speech recognition or a better compromise for intelligibility, detection, quality, and local-ization. Indeed, restoring normality to some aspect of audition (e.g ., recruitment compensa-tion) could be counterproductive if other aspects (e.g ., frequency resolution) remain abnormal .

Importance of Loudness Relationships

Although there is no evidence that any par-ticular loudness relationship is intrinsically important, the achievement of a certain rela-tionship may be essential to meet other objec-tives, notably to maximize audibility consistent with comfort. This is explicit in the rationale of the original (Byrne and Tonisson, 1976) and revised (Byrne and Dillon, 1986) National Acoustic Laboratories' (NAL) procedures. Specif-ically, it is hypothesized that, with volume adjusted for comfortable listening, the audibil-ity of speech averaged across frequencies will be maximized if all frequency bands are equally loud . Conversely, if some frequencies, say the low frequencies, were amplified to higher loudness levels than other frequencies, then the overall level required for comfortable listening (pre-sumably reflected in the preferred hearing aid volume control setting) would be determined mainly by the level of the louder frequencies while the other frequencies could be at an inad-equate, or even inaudible, level.

Figure 2 illustrates how audibility is affected by the choice of frequency response . The data, in addition to the threshold and discomfort levels of a particular client, are the peak (rms + 12 dB) levels of speech amplified with three frequency responses and with gain adjusted for comfortable listening. Response 1 provides the most signal at frequencies below about 800 Hz but the least sig-nal at frequencies above 1600 Hz . Similarly, the other two responses each provide the most sig-nal at some frequencies but the least signal at other frequencies. Thus, the choice of frequency response affects the audibility of speech in dif-ferent frequency regions and the crux of the selection problem is to find the best trade-off between greater audibility in some regions at the expense of less audibility in others .

The selection problem has been quantified by Studebaker (1992), who has combined the Articulation Index (Weighted Audibility Index)


with a loudness model to calculate how audibility of speech is affected by changes in frequency response when a constant loudness level is main-tained . An example calculated by this program (Studebaker and Sherbecoe, 1992) is shown in Figure 3.

The calculations are based on the frequency responses prescribed by four popular procedures for a person with a steeply sloping high-fre-quency hearing loss . The data points on the "0" vertical line show the Weighted Audibility Index for each procedure with the overall gain adjusted to provide the same loudness as the NAL pre-scription. The data points to the left and right of the line indicate how the index would change with variations in the overall slope of the fre-quency response . The significance of these data, to this discussion, is that they show that the choice of frequency response (i .e ., the provision of a particular loudness relationship) can make a considerable difference to the Weighted Audi-bility Index, that is, to potential for under-standing speech. Studebaker (1992) commented that ".. .when loudness was held constant, max-imizing weighted audibility produced frequency-gain prescriptions that agreed with the NAL-R

DISCOMFORT LEVEL

SPEECH PERK LEVELS [C1/3 OCTAVE BAND)

i

0

- 11 THRESHOLD

.25 .5 1 2 4 FREQUENCY (kHz)

I I I I l I 1 -L I I I I I I

Figure 2 Speech peak (1/3-octave band) levels for three frequency response characteristics (1, 2, and 3) with the overall gain adjusted to provide the preferred listening level.

385


w

0 . 4

0 . 35

0 . 3

q w 0 .2S

0 . 2 -10 -5 0 5 10

SLOPE (dB/OCTAVE) RE PRESCRIBED HEARING RID RESPONSE

Figure 3 The Weighted Audibility Index (a form of the Articulation Index) for the prescriptions of four proce-dures, and variations in slope from those procedures, for a steeply sloping high-frequency hearing loss . For all prescriptions and variations, overall gain has been adjusted to provide the same loudness as the NAL pre-scription.

prescription to a surprising extent" (p . 118) . This appears to confirm that the loudness rela-tionship provided by the NAL procedure, namely, equal loudness across frequencies, does tend to achieve its aim of maximizing audibility at a com-fortable level.

The above discussion has referred to fre-quency responses prescribed for linear amplifi-cation. However, nonlinear systems could also be considered in terms of what loudness rela-tionship is best, for a particular listening condition or over a range of conditions, for maximizing some objective such as audibility and/or comfort and/or sound quality. This ori-entation may be more productive than trying to restore normal loudness or to compensate specif-ically for some other hearing loss characteristic .

What Bandwidths Should Be Normalized or Equalized?

A methodological problem when trying to normalize, or equalize, loudness across frequency is deciding what bandwidths to consider. Equal-izing loudness for narrow bands, say 1/3- or 1/2-octave bands, may require different amplification from equalizing loudness for broader bands (Byrne, 1986). There is no obvious simple way of decid-ing what bandwidth is most appropriate, espe-cially as it probably varies (as critical bandwidths do) across individuals. This issue needs clarifi-cation before any loudness rationales can be evaluated with confidence . It is also possible that there may be less difference between equal loudness and normal loudness if it is appropri-ate to consider broad rather than narrow bands.

Are Loudness Measurements Always Valid?

Research has demonstrated acceptable reli-ability for loudness scaling measurements and that such measurements produce similar loud-ness growth functions for nearly all normal-hearing listeners (Elberling and Nielsen, 1993 ; Hohmann and Kollmeier, 1995 ; Kiessling, 1995). However, this is not sufficient to show that loud-ness scaling always gives a valid indication of amplification requirements for hearing-impaired listeners. The justification for using loudness scaling, rather than simply measuring thresh-old or threshold and loudness discomfort level, is that some individuals will show loudness growth functions that are not predictable from other measurements . However, there is an unan-swered question of whether, for such individu-als, the loudness scaling data are the best indication of amplification requirements . It is possible that such "aberrant" scaling results (i .e ., ones that are greatly different from those of other people with similar hearing losses) reflect differences in using the loudness labels, rather than or in addition to differences in loud-ness perception . Procedures based on loudness scaling assume that the difference between two labels (e .g ., "very soft" and "very loud") corre-sponds to a constant loudness difference across people, and we cannot be sure that this is entirely true . This poses a methodological problem for val-idation research and for the application of loud-ness scaling based fitting procedures . A research need is to investigate whether fittings based on unusual loudness data are better or poorer than ones that assume average loudness growth functions.

Is One Fitting Rationale Sufficient?

Even for linear amplification, a single fitting rationale is not sufficient for all degrees of hear-ing loss . Studies by Schwartz et al (1988) and Byrne et al (1990a) have indicated that fitting procedures need to be modified substantially when hearing losses exceed about 60 dB HTL. Theoretical reasons for this have been discussed by Van Tasell (1993), among others . Byrne et al (1990b) have shown that the revised NAL pro-cedure achieved its aim of amplifying all speech regions to most comfortable loudness level (MCL) but that this rationale did not provide optimal amplification for some people with severe or profound hearing losses . Particular types of compression systems are usually designed for a

386


certain range of hearing losses and sometimes it is specifically acknowledged that other types of systems may be required for other degrees of hearing loss (e .g ., Killion, 1993).

A further issue is whether a single fitting rationale, such as amplifying all speech bands to MCL, is appropriate for all listening condi-tions? Data from studies of multiple memory hearing aids (Keidser et al, 1996) suggest not. Keidser (1995) found that, under one condition, one type of compression was best, whereas a different type was best under another condi-tion . Furthermore, both types of compression were inferior to linear amplification for some con-ditions . Specifically, high-frequency compres-sion was best for understanding a dialogue between two voices that varied substantially (by 10 dB) in overall level; low-frequency com-pression was best for minimizing the annoy-ance of traffic noise; linear amplification with the NAL frequency response was best for listening to speech in quiet or in speech spectrum shaped noise; linear amplification with a low cut, rela-tive to the NAL response, was best for under-standing speech in traffic noise. These findings suggest that an adequate rationale for fitting compression amplification may need to consider differences in listening conditions as well as hearing loss characteristics. Such a rationale could lead to prescribing two or more types of compression, possibly including linear amplifi-cation, to be provided by a multiple memory hearing aid. Indeed, the same hearing aid may be fitted by different strategies (choices of which parameters are prescribed and on what basis), depending on what is considered to be the pur-pose of using compression (Dreschler, 1992).

Consistency with Current Fitting Rationales

With just a few exceptions (e .g., Kiessling and Steffens, 1991; Fabry, 1993; Cornelisse et al, 1995), the procedures suggested for fitting non-linear hearing aids start from "scratch" rather than attempting to build on existing procedures . Loudness normalization is, in fact, contrary to the rationale of most currently popular procedures for prescribing the frequency response of linear hearing aids . The NAL (Byrne and Dillon, 1986), Desirable Sensation Level (Seewald et al, 1985, 1991), and Memphis State University (Cox,1983, 1988) are based on the rationale of amplifying all speech bands to MCL (or something equivalent), and the general idea is that this will maximize the potential for understanding speech when

listening at a comfortable level. This means that the procedure is intended to compensate for interfrequency differences in the long-term speech spectrum as well as interfrequency differences in preferred listening levels . Some other proce-dures (e.g ., Berger et al, 1977; McCandless and Lyregaard, 1983) have similar adjustments that result in prescribing less gain, for a given hear-ing level, at the low frequencies than at the mid and high frequencies. A considerable body of research (reviewed in Skinner, 1988; Byrne, 1993) supports the rationale of amplifying speech bands to MCL or similar rationales . That research supports "equal" rather than "normal" loudness as being the optimal relationship among the var-ious frequency bands of speech . Furthermore, for listening in low-frequency weighted noise, such as traffic noise, the best frequency response may have less low frequencies than the NAL response (Keidser et al, 1996). Such a response would be further removed than the NAL response from what would be required to normalize loudness .

Comparisons of Fitting Rationales and Procedures

What differences will result from using dif-ferent fitting rationales, as implemented in the various fitting procedures? An example com-paring the loudness normalization and loud-ness equalization rationales is presented in Figure 4. This shows the gain and frequency response prescribed by the NAL and Fig. 6 pro-cedures, for an average speech level, for 50 dB HTL across all frequencies.

The main difference between the procedures is that Fig. 6 prescribes a flat frequency response, whereas NAL prescribes a low-cut response, specifically less gain at the frequencies below 1000 Hz than at the mid and high frequencies. The main reason for the difference between pre-scriptions is that, in order to equalize loudness across frequencies, it is necessary to "flatten" the long-term average speech spectrum (LTASS) by varying the gain across frequency to compensate for the intensity differences of the different frequency bands of speech . The figure also shows the LTASS for speech, spoken with normal vocal effort (Byrne et al, 1994). Note that the LTASS and NAL response curves are approximately mirror images of each other in that the 8 dB higher speech level at 500 Hz compared with 1000 Hz is closely matched by the reduction in gain at 500 Hz . Therefore, when speech is ampli-fied with the NAL response, the 500-Hz band will be about equal in intensity to the 1000-Hz band,

387


(0 CE . F-0 J

LL I

M Z v U

I n 00 J w

w J

Z H M Q Z 0 M Z o: 0o;

Q:f ~ w U

z ° H M

A LLJ CO Ld H >

U I-vl

30

20

10

w J 0.25 0.5 1 2 4 8

0- W FREQUENCY (kHz)

Figure 4 Comparison of prescriptions of NAL and Fig . 6 procedures for a flat 50 dB HTL audiogram . Fig. 6 pre-scriptions shown for low, MCL, and high inputs . Relative 1/3-octave band levels of LTASS are also shown .

which will correspond (approximately) to equal loudness . By contrast, when speech is amplified with the Fig. 6 response, the normal intensity and loudness differences among different speech bands will be retained, that is, the low fre-quencies will be louder than the mid and high frequencies.

For flat audiograms (of 95 dB HTL or less), the frequency responses (but not absolute gain values) for NAL and Fig. 6 will always be the same as shown in Figure 4. Of greater interest is how the difference between loudness nor-malization and loudness equalization will vary for different input levels . Figure 4 shows the pre-scriptions of Fig. 6 for low and high, as well as for average, speech inputs . Although gain varies with input, the frequency response is flat for all levels . Although the NAL procedure does not provide prescriptions for low or high inputs, we might deduce that the NAL frequency response shown probably would not equalize loudness for all input levels . The procedure is, in effect, attempting to overlay the amplified LTASS on an equal loudness contour and the shape of such contours may vary with level. However, the cal-culation of equalization, or normalization, responses for various speech levels raises the complex and unresolved issue of what band-widths to consider (see previous discussion) and the further complication, which is considered in

the IHAFF procedure, that the LTASS varies with vocal effort .

A further illustration of the main difference between normalization and equalization, for an average input, is shown in Figure 5. For this client, whose audiogram slopes gently from 30 dB HTL at 500 Hz to 55 dB HTL at 6000 Hz, pre-scriptions were calculated according to two equalization procedures (NAL and DSL) and three normalization procedures (Fig. 6, LGOB, and IHAFF) . Loudness scaling data were obtained for the LGOB procedure and these were also used, with some interpolation, to approximate the IHAFF procedure. The most obvious difference between the equalization and normalization response is that the former have a low cut below 1000 Hz, again demonstrating that equalization requires flattening of the LTASS, whereas normalization does not.

The five procedures were compared for seven clients, selected to present a wide vari-ety of audiograms . In the example shown (see Fig. 5), the two equalization procedures pre-scribe responses that are similar (but certainly not identical) to each other, as do the three nor-malization procedures . However, this is not so for all clients. For steeply sloping high-fre-quency audiograms, DSL will prescribe far more high-frequency emphasis than NAL. This is because variations in audiogram slope are matched by about one-third as much variation in the frequency response slope for NAL but by about two-thirds as much variation for DSL. Thus, although the two procedures have essen-tially the same rationale, which is to amplify all speech bands to MCL, a characteristic of the cal-culation procedure (the slope rule) will result in considerably different prescriptions for some

F EQUAL LOUDNESS - NAL

DSL

NORMAL LOUDNESS o FIG. 6 o LGOB x IHAFF

P -

2S 5 1 2 4 6 2S 5 1 2 4 6

FREQUENCY (kHz)

Figure 5 Comparison of prescriptions for two loudness equalization (NAL, DSL) and three loudness normaliza-tion (Fig . 6, LGOB, IHAFF) fitting procedures, for a per-son with a moderate, gently sloping audiogram (30 dB HTL at 500 Hz to 55 dB HTL at 6000 Hz).

388

audiograms . A similar situation exists for the normalization procedures . Although the LGOB and IHAFF procedures have similar rationales, and the calculations used the same loudness scaling data, the prescribed responses (and overall gain) were considerably different in some instances. For examplb, for one client, for an average input level, LGOB prescribed a 1 dB/ octave rising response between 500 Hz and 2000 Hz, whereas IHAFF prescribed a 5 dB/octave falling response . For this client, Fig. 6 prescribed a 6 dB/octave falling response and was, thus, very similar to IHAFF but, in other instances, Fig. 6 and IHAFF differed substan-tially. Such differences may result from differ-ences in the calculation methods of the two procedures and/or because one procedure was based on loudness data and the other on HTLs .

It is evident from these comparisons that procedures based on the same rationale can result in substantially different prescriptions because of differences in how the rationale is implemented, that is, differences in the calcu-lation methods. It is also true that the effects of differences in calculation methods can, coinci-dentally, tend to offset the effects of differences in rationale. This can mean that, for some peo-ple, basically different procedures can happen to provide similar prescriptions. For example, for moderately sloping high-frequency losses, NAL and Fig. 6 may give similar prescriptions despite the differences in rationales, which will lead to substantially different prescriptions for other audiogram configurations . A challenge for research is to evaluate different rationales with-out being misled by the effects of procedural differences that are incidental to those rationales .

Is Prescription Necessary for Compression Parameters?

Earlier (see Table 1), a list of possible pre-scriptive parameters was presented. However, do all of these parameters really need to be pre-scribed, that is, calculated in a relatively precise fashion from individually measured hearing loss characteristics? The answer will depend on what fitting rationale is adopted. Advocates of loud-ness normalization (e.g ., Pluvinage, 1989 ; Cox, 1995) often suggest deriving a loudness growth function from measurements at several levels and at two or more frequencies. This informa-tion is used to calculate the required CRs in dif-ferent bands and, possibly, the point at which compression reverts to linear amplification, if such a system is used . Such a procedure is only


justified if one believes that it is important to achieve a close match between compression characteristics and loudness growth functions. If less precision is acceptable, then compres-sion characteristics could be calculated from a more simple set of measurements . Loudness growth functions can be estimated reasonably well from dynamic range (LDL-HTL) (Elberling and Nielsen, 1993) or even from thresholds only (Hellman and Meiselman, 1990), although it appears that the accuracy of such prediction decreases as HTL becomes greater (Kiessling, 1995).

If the rationale for using compression is to reduce long-term variations in speech levels, then a slow-acting system could be used and there would seem to be little justification for precise selection of compression parameters . It may be reasonable to use the same compres-sion for everyone or to use just two or three variations in parameter values, selected for clients grouped broadly according to dynamic ranges or hearing levels .

At present, it is not clear whether, or how much, the use of compression complicates the hearing aid selection process because it is uncer-tain to what extent compression parameters need to be prescribed . Resolving this question is important for developing effective and efficient selection procedures .

STAGED APPROACH FOR FITTING NONLINEAR HEARING AIDS

T he demonstrated or hypothesized advan-tages of compression amplification have

been indicated above. However, many authors have suggested that some types of compression could also be detrimental. The main problem is associated with fast-acting (syllabic) compression that reduces intensity differences between suc-cessive speech sounds and, if multiband com-pression is used, between simultaneously occurring components such as the vowel for-mants. This reduction in intensity differences, together with possible distortions of temporal information, may reduce important cues for speech perception (Boothroyd et al, 1988 ; Plomp, 1988 ; Moore, 1993 ; Van Tasell, 1993). Indeed, in a study with a master hearing aid system in which long-term variations in speech level were minimized and the output was delivered at the highest comfortable level, Boothroyd et al (1988) found that the addition of syllabic compression resulted in a decrement in speech perception for the majority of their profoundly hearing-

389


impaired subjects . Compression may also increase noise during the gaps between speech and may increase the likelihood of acoustic feed-back, especially when it involves applying high gain to low-intensity, high-frequency sounds (Dillon, 1996).

The problem of using compression optimally can be viewed as finding ways to maximize the potentially beneficial effects while minimizing the potentially detrimental effects . The Moore system reduces the need for syllabic compression by preceding it with slow-acting compression to reduce the long-term variations in speech levels . It now seems widely believed that, to avoid excessive reduction of spectral contrasts in complex stimuli, multiband systems should use just two or three bands rather than a large number. Moore (1993) recommends that the bands of a compression system should be broad compared with auditory filters to avoid adding to speech perception problems arising from reduced frequency selectivity. Despite this rather general view, recent research by Crain and Yund (1995) suggests that it may not necessarily be detrimental to use as many as 31 bands in com-pression systems when accompanied by indi-vidual fitting.

In line with the suggestions of others, notably Boothroyd et al (1988) and Van Tasell (1993), a logical approach to using nonlinear amplification (or any other form of signal pro-cessing) would be to define an objective (e .g., maximizing audibility at a comfortable level), seek to achieve the objective to the greatest extent with the most simple means, and then use more complex processing only if it is advanta-geous and only to the extent necessary.

In the present context, application of the above approach would be to optimize linear amplification, then (if necessary) use slow-act-ing compression (which is mostly free of the above-mentioned detrimental effects), and finally to use syllabic compression only to the extent required .

A staged approach to providing speech audi-bility, starting with the least drastic "medicine," is illustrated in Figures 6 and 7 . The left panel of Figure 6 shows the LTASS for speech with an overall level of 70 dB SPL, relative to a flat hearing loss of 40 dB HTL. The greater part of the speech signal is audible, even without ampli-fication, but the less intense components are not. Obviously, if the voice level drops below 70 dB SPL, a greater amount of speech will become inaudible. The right panel shows the LTASS amplified according to the NAL procedure

for prescribing gain and frequency response . Except for very low frequencies, all of the speech signal is audible and would remain so even if the voice level dropped by 5 dB . A person with such a hearing loss should hear relatively well with linear amplification with suitable frequency shaping. However, the addition of compression might be helpful to cope with particularly low voice levels and to minimize the need for volume adjustments. These requirements could be met by a slow-acting system .

Figure 7 shows the situation for a person with a moderate low-frequency hearing loss sloping to a severe high-frequency hearing loss . In the unaided condition (left panel), only the low frequencies of speech are audible. If aided with a flat frequency response (right panel), audibil-ity is improved for the mid frequencies but a large proportion of the signal above 1000 Hz is still inaudible. When aided with the NAL response, there is a further improvement but much of the high-frequency signal is still inaudi-ble. What are the options for improving audibility of the high frequencies? One recurring sugges-tion is to provide whatever high-frequency emphasis is needed to maximize the AI (e.g ., Dugal et al, 1980 ; Humes, 1986 ; Rancovic, 1991). However, providing more high-frequency gain will result in increased loudness, which will lead to a reduction in overall gain to maintain the preferred listening level. When the preferred loudness level is maintained, high-frequency emphasis in excess of that prescribed by the NAL procedure usually will not increase the AI when an appropriate hearing loss desensitiza-tion correction is made (Studebaker, 1992). This agrees with experimental findings that the extreme high-frequency emphasis indicated by

LBL LBL

100

` I -- ~) <,/ =TAVE emp~, -

80 0-

80

T-_ 40dB HTL 40 w ___,`

J 20

NO AMPLIFIG TION 70dB SPL SPEECH 0 FLAT HERRING LOSS . PMPLIFIEB BY NAL GRIN

.25 .5 1 2 4 .25 .5

FREQUENCY (kHz)

Figure 6 The long-term-average (Y3-octave band) speech spectrum (broken lines show peak levels and minima) for an overall level of 70 dB SPL, shown in relation to a mod-erate flat hearing loss, without amplification (left panel) and amplified according to the NAL prescription.

390


100

J 80

m 60

J 40 w W 20 J

higher CR would be required to provide a given degree of audibility.

The above analysis of when and what type of compression is needed is consistent with the findings of Moore and Glasberg (1993) who examined the effects of amplification on simu-lated hearing losses with recruitment. They con-cluded that linear amplification with the NAL frequency response was effective for listening to speech in quiet at an average level, but that an AVC system, or manual volume adjustments, would be required to maintain audibility for low-level inputs and comfort for high-level inputs. For listening in noise, consisting of a single talker, they concluded that syllabic com-pression at the high frequencies would be desir-able for a sloping hearing loss similar to that illustrated in Figure 7. They suggest that, for such a case, the most likely explanation for the poor performance of linear amplification was that "the amount of high frequency amplification was not sufficient to restore audibility to all the important high-frequency components in speech" (p . 2060).

Although the approach outlined in this sec-tion is logical, it must be admitted that there is little solid evidence that syllabic compression sys-tems can be detrimental (Dillon, 1996). It may be that a variety of fast-acting and slow-acting systems will have a similar beneficial effect in improving audibility and comfort for a wide range of input levels and with minimal use of vol-ume adjustments.

.25 .5 1 2 4 .25 .5

FREQUENCY (kHz)

Figure 7 The LTASS (broken lines show peak levels and minima) for an overall level of 70 dB SPL, shown in rela-tion to a hearing loss that slopes from mild at low fre-quencies to severe at high frequencies. Left panel shows unamplified speech ; right panel shows speech with flat amplification (equal gain at all frequencies) and with gain according to NAL prescription.

an AI-type prescription does not provide best speech intelligibility for people with steeply sloping hearing losses (Byrne, 1986, 1992 ; Skinner, 1988 ; Rancovic, 1991). Realistically, the situation shown in Figure 7 probably illus-trates the best that can be achieved by frequency shaping alone. There is, therefore, a good theo-retical case for attempting to improve audibil-ity by using fast-acting wide-range compression at the high frequencies, as well as frequency shaping and probably AVC to compensate for long-term fluctuations in voice levels . (Of course, it is still necessary to test whether the overall effect of compression actually is beneficial in a case such as this .)

The figure also illustrates the advantage of doing as much as possible with frequency shap-ing before implementing compression. (This com-ment refers to the conceptual strategy to be employed, not to the pros and cons of placing the frequency shaping before or after compression in the hearing aid circuit.) If the frequency response were flat, rather than shaped, then higher CRs, and presumably more risk of detrimental effects, would be required to restore the same degree of audibility. This illustration provides a further argument against loudness normalization. If the LTASS were amplified to provide normal loud-ness, it would fall about mid way between the flat and the NAL amplified LTASS. (For a sloping audiogram, the achievement of normal loudness would require a frequency response with some high-frequency emphasis, but not as much as that required for equal loudness [i .e ., the NAL response]) . Thus, if normal rather than equal interfrequency loudness were the aim, then a

CONCLUSIONS AND RECOMMENDATIONS

have discussed some basic issues concerning the fitting of advanced hearing aids . Pri-

marily, these relate to the selection of com-pression characteristics, together with the further complication that some aids also have multiple memories . In part, this article has been prompted by three concerns about current trends in fitting advanced hearing aids . First is a concern that, with the increasing development of pro-prietary fitting systems, there may be a prolif-eration of manufacturer-specific fitting formulae that may not be validated and, indeed, may not even be open to scrutiny. A related concern is that the clinician may finish up with little control over, or understanding of, the fitting process. A second concern is that nearly all current attempts to devise fitting procedures are based, to a greater or lesser extent, on a single ratio-nale, namely, loudness normalization . Associated

391


with this is a lack of interest in critically eval-uating that rationale or in considering other possible fitting rationales . Although loudness normalization is intuitively appealing, there are reasons to suspect that it may not be an optimal fitting rationale. A third concern is that most pro-posed procedures ignore the results of years of research into the amplification requirements of hearing-impaired people . With these concerns in mind, the following principles are suggested for approaching the development of fitting proce-dures for advanced hearing aids .

First, methods for prescribing advanced hearing aids should build on current prescrip-tive principles and methods. Although the new hearing aids add complexities, there are common points of reference. For all hearing aids, we need to prescribe limiting and, for any given input, the amount of amplification (overall gain) and the frequency response shaping. For non-linear aids, we need a series of different gain and frequency response prescriptions for different input levels and spectra. Most research under-lying prescriptive procedures has been conducted with sensorineural listeners with, presumably, typical auditory characteristics such as loud-ness growth functions and dynamic range. To the extent that prescriptive procedures have been validated, the prescriptions may be considered appropriate for clients with the specified degree of hearing loss when receiving a typical hearing aid input. Typical overall speech input levels appear to be about 70 dB SPL for clients with moderate hearing losses (Martin et al, 1976 ; Walden et al, 1977 ; Farrell et al, 1979) and a few dB higher for clients with severe hearing losses (Byrne and Cotton, 1987 ; Parkinson et al, 1989). For a typical input level, the amplification requirements defined by research, using linear hearing aids, should be approximately correct for nonlinear hearing aids as well . Therefore, the better-validated current prescriptive procedures can provide a prescription for one input level for nonlinear hearing aids and this may provide a useful reference from which to calculate the variations in prescriptions required for other input levels . This seems preferable to develop-ing completely new procedures that ignore the results of years of research . It should be stressed that the potential advantages of compression apply only to low inputs (improved audibility, no need to turn up volume) and to high inputs (greater comfort and/or reduction of masking, no need to turn down volume). There is little rea-son to suppose that the use of compression will change amplification requirements for moderate

input levels . As moderate input levels are more common than either very low or very high lev-els, it would seem especially important to get this aspect of the prescription right.

In a similar vein, current prescriptive pro-cedures provide the logical starting point for deriving a set of amplification prescriptions for multiple memory hearing aids . The prescrip-tion will be the best estimate of what is required for the most typical listening conditions and research will indicate how this needs to be var-ied to suit other acoustic conditions or listening needs. The research of Keidser et al (1996) sup-ports this position in that the NAL prescription was usually effective for listening to speech in quiet or in noises with speech-like long-term spectra, whereas certain other frequency responses or compression conditions were bet-ter for some other listening conditions .

Second, research is essential to develop val-idated procedures for prescribing nonlinear amplification. Such procedures need to be sup-ported by experimental evidence rather than logic alone or anecdotes. This is a substantial task that will probably require the efforts of many sci-entists from various academic and commercial settings . With a few exceptions (e.g ., Lundh, 1981, 1983), there seems regrettably little inter-est in tackling such basic questions as whether there is any value in attempting to normalize loudness growth functions. On the contrary, there is often a distressing tendency to assume that what seems right must be right even though some "right-sounding" theories are mutually contradictory. Preferably, research should eluci-date the principles relevant to prescribing ampli-fication rather than simply comparing two or more available systems. While it may be of some value to know that system A works better than system B, it is much more valuable to understand the relevant principles and thereby permit infer-ences about the likely effect of any of a wide range of systems. Indeed, this provides the basis for developing better systems. Research should include independent scientists and clinicians and should produce prescriptive procedures that can be applied generally rather than being spe-cific to a particular brand of hearing aid (THe IHAFF, Fig. 6, and DSL [i/o] are examples of such procedures). Research findings must be published in scientific journals so that they can undergo rigorous scrutiny and, hopefully, lead to independent confirmation . This process may be slow but the alternative would be that each type of advanced hearing aid will be fitted according to a different prescriptive formula supported by

392


little (if any) published research . Among other problems, that situation would make it difficult to evaluate the effectiveness of fitting formulae and the systems with which they are used with-out confounding one with the other.

Finally, this article is based on the common assumption that compressidn systems can be beneficial . Although compression reduces the need for manual volume adjustments, other benefits largely remain to be demonstrated (Dil-lon, 1996). The extent to which research can demonstrate such benefits, and their specific nature, should be a major consideration in devel-oping fitting procedures .

Acknowledgment . I am very grateful to two anony-mous reviewers and to the following people for their

comments on an earlier draft of this manuscript : Arthur Boothroyd, Robyn Cox, Harvey Dillon, Donald Dirks, Claus Elberling, Gitte Keidser, and Richard Tyler.

REFERENCES

Berger KW, Hagberg EN, Rane RL . (1977) . Prescription of HearingAids: Rationale, Procedures, and Results. Kent, OH : Herald .

Biering-Sorensen M, Riess H, Boisen G, ParvingA. (1995) . A clinical comparative investigation of a non-linear versus linear hearing aid. Scand Audiol 24:125-132 .

Boothroyd A, Springer N, Smith L, Schulman J. (1988) . Amplitude compression and profound hearing loss. J Speech Hear Res 31:362-376 .

Bratt GW Sammeth CA. (1991). Clinical implications of prescriptive formulas for hearing aid selection. In : Studebaker GA, Bess FH, Beck LB, eds. The Vanderbilt Hearing Aid Report II. Parkton, MD : York Press, 23-33.

Brunved PB . (1994) . How studying loudness growth led to the development of MultiFocus . Hear Instr Suppl 1 45:8-10.

Byrne D. (1979) . Selective amplification : some psychoa-coustic considerations. In : Bess FH, Freeman BA, Sinclair JS, eds. Amplification in Education. Washington, DC : AG Bell, 260-285.

Byrne D. (1983) . Theoretical prescriptive approaches to selecting the gain and frequency response of a hearing aid. Monogr Contemp Audiol 4(1):1-40.

Byrne D, Cotton S. (1987) . Preferred listening levels of persons with sensorineural hearing losses . Aust JAudiol 9:7-14.

Byrne D, Cotton S. (1988) . Evaluation of the National Acoustic Laboratories' new hearing aid selection proce-dure . J Speech Hear Res 31:178-186 .

Byrne D, Dillon H. (1986) . The National Acoustic Laboratories' (NAL) new procedure for selecting the gain and frequency response of a hearing aid. Ear Hear 7:257-265 .

Byrne D, Dillon H, Tran K, Arlinger S, Wilbraham K, Cox R, Hagerman B, Hetu R, Kei J, Lui C, Kiessling J, Kotby NM, Nasser NHA, El Kholy WAH, Nakanishi Y, Oyer H, Powell R, Stephens D, Meredith R, Sirimanna T, Tavartkiladze G, Frolenkov GI, Westerman S, Ludvigsen C. (1994) . An international comparison of long-term average speech spectra. J Acoust Soc Am 96 : 2108-2120.

Byrne D, Parkinson A, Newall P. (1990a). Hearing aid gain and frequency response requirements for the severely/profoundly hearing impaired . Ear Hear 11:40-49 .

Byrne D, Parkinson A, Newall P. (1990b). Comparisons of NAL and MCL based hearing aid frequency response prescriptions for severely/profoundly hearing-impaired clients. Aust JAudiol 12:1-9 .

Byrne D, Parkinson A, Newall P (1991) . Modified hear-ing aid selection procedures for severe/profound hearing losses . In : Studebaker GA, Bess FH, Beck L, eds. The Vanderbilt Hearing Aid Report 77. Parkton, MD: York Press, 295-300.

Byrne D, Tonisson W. (1976) . Selecting the gain of hear-ing aids for persons with sensorineural hearing losses . Scand Audiol 5:51-59 .

Cornelisse L, Seewald RC, Jamieson DG . (1995) . The input/output formula: a theoretical approach to the fit-ting of personal amplification devices. J Acoust Soc Am 97:1854-1864 .

Cox R. (1988) . The MSU hearing instrument prescrip-tion procedure. Hear Instr 39(1):6-10.

Cox RM. (1983) . Using ULCL measures to find fre-quency/gain and SSPL90 . Hear Instr 34(7):17-21, 39 .

Cox RM. (1995) . Using loudness data for hearing aid selection : the IHAFF approach . Hear J 48(2):10-44 .

Crain TR, Yund EW (1995) . The effect of multichannel compression on vowel and stop-consonant discrimination in normal-hearing and hearing-impaired subjects . Ear Hear 16:529-543 .

Byrne D. (1986). Effects of frequency response charac-teristics on speech discrimination and perceived intelligibility and pleasantness of speech for hearing-impaired listeners. JAcoust Soc Am 80:494-504.

Byrne D. (1992) . Key issues in hearing aid selection and evaluation . J Am Acad Audiol 3:67-80 .

Byrne D. (1993) . Implications of National Acoustic Laboratories' (NAL) research for hearing aid gain and frequency response selection strategies . In : Studebaker GA, Hochberg I, eds. Acoustical Factors Affecting Hearing Aid Performance. Boston : Allyn and Bacon, 119-131.

Crouch JD, Pendry BL . (1975a). Otometry in clinical hearing aid dispensing . Hear Aid J 28(10) :12, 42-48.

Crouch JD, Pendry BL . (1975b). Otometry in clinical hearing aid dispensing . Hear Aid J 28(11) :18, 31-33.

Dillon H. (1996) . Compression? Yes, but for low or high frequencies, for low or high intensities, and for what response times? Ear Hear 17:287-307 .

Dillon H, Chew R, Deans M. (1984) . Loudness discom-fort level measurements and their implications for the design and fitting of hearing aids . Aust JAudiol 6:73-79 .

393


Dillon H, Lovegrove R. (1993) . Single microphone noise reduction systems for hearing aids : a review and an eval-uation. In : Studebaker GA, Hochberg 1, eds. Acoustical Factors Affecting HearingAidPerformance. Boston : Allyn and Bacon, 353-372.

Dillon H, Yeend I, Storey L, Wigney D. (1994, September) . Selecting the Maximum Output Level of Hearing Aids. Paper presented at the Second International Symposium on Speech and Hearing Sciences, Osaka, Japan.

Dreschler WA. (1992) . Fitting multi-channel compres-sion hearing aids . Audiology 31:121-131 .

Dugal RL, Braida LD, Durlach NI . (1980) . Implications of previous research for the selection of frequency-gain characteristics . In : Studebaker GA, Hochberg I, eds. Acoustical Factors Affecting Hearing Aid Performance. Baltimore: University Park Press, 379-403.

Elberling C, Nielsen C. (1993). The dynamics of speech and the auditory dynamic range in sensorineural hear-ing impairment . In : Beilin J, Jensen GR, eds. Recent Developments in Hearing Instrument Technology. 15th Danavox Symposium. Copenhagen : Stougaard Jensen, 99-133 .

Keidser G, Dillon H, Byrne D. (1996) . Guidelines for fit-ting multiple-memory hearing aids . J Am Acad Audiol 7:406-418 .

Kiessling J. (1995) . Loudness growth in sensorineural hearing loss-consequences for hearing aid design and fitting . Audiol Akust 34(2):82-89 .

Kiessling J, Steffens T. (1991). Clinical evaluation of a programmable three-channel automatic gain control amplification system. Audiology 30:70-81 .

Killion M. (1993) . The K-Amp hearing aid: an attempt to present high fidelity for persons with impaired hear-ing. Am J Audiol 2:52-74 .

Killion M. (1994) . Fig 6. WQ1: Loudness-based fitting Formulae . Elk Grove Village, IL : Etymotic Research.

Killion MC, Villchur E. (1993). Kessler was right-partly : but SIN test shows some aids improve hearing in noise. Hear J 46(9):31-35 .

Kollmeier B, Peissig J, Hohmann V. (1993) . Real-time multiband dynamic compression and noise reduction for binaural hearing aids . J Rehab Res Dev 30(1):82-94 .

Fabry DA . (1993). A clinical procedure described for eval-uating compression hearing aids . Hear J 46(11) :25-30 .

Farrell DE, Jansen JW, Sweetman RH. (1979, November). Clinical versus Real-life Gain Setting for Experienced Hearing Aid Wearers. Paper presented at the Annual Convention of American Speech and Hearing Association, Atlanta.

Hawkins DB, Walden BE, Montgomery A, Prosek RA. (1987) . Description and validation of an LDL procedure designed to select SSPL90 . Ear Hear 8:162-169 .

Hellman RP, Meiselman CH. (1990) . Loudness relations for individuals and groups in normal and impaired hear-ing. JAcoust SocAm 88:2596-2606.

Hohmann V, Kollmeier B. (1995) . Further development and use of hearing field scaling in clinics . Audiol Akust 34(2):48-59.

Horwitz AR, Turner CW Fabry DA . (1991) . Effects of dif-ferent frequency response strategies upon recognition and preference for audible speech stimuli. J Speech Hear Res 34:1185-1196 .

Humes L. (1986) . An evaluation of several rationales for selecting hearing aid gain . J Speech Hear Disord 51:272-281 .

Jerger J, Darling R, Florin E. (1994). Efficacy of the cued-listefiing task in the evaluation of binaural hearing aids . J Am Acad Audiol 5:279-285 .

Johnson JS, Pluvinage V, Benson D . (1989) . Digitally pro-grammable full dynamic range compression technology. Hear Instr 40(10) :26-30 .

Keidser G. (1995) . The relationship between listening conditions and alternative amplification schemes for mul-tiple-memory hearing aids . Ear Hear 16:575-586 .

Keidser G, Dillon H, Byrne D. (1995) . Candidates for multiple frequency response characteristics . Ear Hear 16:562-574 .

Kuk F. (1994) . A screening procedure for modified sim-plex in frequency-gain response selection . Ear Hear 15 :62-70 .

Laurence RF, Moore BJ, Glasberg BR. (1983) . A com-parison of behind-the-ear linear hearing aids and two-channel compression aids, in the laboratory and in everyday life. Br JAudiol 17 :31-48 .

Levitt H, Sullivan J, Neumann A, Rubin-Spitz JA. (1987) . Experiments with a programmable master hearing aid. J Rehab Res Dev 24(4):29-54 .

Lundh P (1981) . Field Test for Two Channel Compression Hearing Aid. Report No . 14-8-6 . Denmark: Oticon A/S.

Lundh P. (1983) . Field Test for Two Channel Hearing Aid with Compressor in Low-frequency Channel. Report No . 14-8-7 . Denmark: Oticon A/S.

Lundh P, Nielsen C. (1992) . Multi-Focus Field Test . Report No . 14-8-7. Denmark: Oticon A/S.

Lyregaard PE . (1988) . POGO and the theory behind . In : Jensen HJ, ed . Hearing Aid Fitting: Theoretical and Practical Viewpoints . 13th Danavox Symposium. Copenhagen : Stougaard Jensen,81-96 .

Martin MC, Grover BC, Worrall JJ, Williams V (1976) . The effectiveness of hearing aids in a school population . Br J Audiol 10:33-44.

McCandless GA, Lyregaard PE . (1983) . Prescription of gain/output (POGO) for hearing aids . Hear Instr 34(1): 16-17,19-21.

Moore BJ . (1987) . Design and evaluation of a two-chan-nel compression hearing aid. J Rehab Res Dev 24(4): 181-192.

Moore BJ . (1993) . Signal processing to compensate for reduced dynamic range. In : Beilin J, Jensen GR, eds. Recent Developments in Hearing Instrument Technology. 15th Danavox Symposium. Copenhagen : Stougaard Jensen,147-165 .

394


Moore BJ, Glasberg BR. (1986) . A comparison of two-channel and single-channel compression aids . Audiology 25:210-226 .

Rankovic CM. (1991) . An application of the Articulation Index to hearing aid fitting. J Speech Hear Res 34:391-402.

Moore BJ, Glasberg BR . (1988) . A comparison of four methods for implementing automatic gain control (AGC) in hearing aids . Br JAudiol 22:93-104 .

Moore BJ, Glasberg BR . (1993) . Simulation of the effects of loudness recruitment and threshold elevation on the intelligibility of speech in quiet and in a background of speech . JAcoust Soc Am 94:2050-2062 .

Moore BJ, Laurence RF, Wright D. (1985). Improvements in speech intelligibility in quiet and in noise produced by a two-channel compression hearing aid. Br J Audiol 19:175-187 .

Moore BC, Johnson JS, Clark TM, Pluvinage V (1992) . Evaluation of a dual-channel full dynamic range com-pression system for people with sensorineural hearing loss . Ear Hear 13:349-370 .

Mueller G. (1995) . Editor's introduction to Cox RM (1995). Using loudness data for hearing aid selection : the IHAFF approach . Hear J 48(2):10-44 .

Neuman A, Levitt H, Mills R, Schwander T. (1987) . An evaluation of three adaptive hearing aid selection strate-gies . JAcoust Soc Am 82:1967-1976 .

Neuman AC, Bakke MH, Hellman S, Levitt H. (1994) . The effect of compression ratio in a slow-acting com-presssion hearing aid: paired comparison judgments of quality. J Acoust Soc Am 96:1471-1478 .

Parkinson A, Byrne D, Newall P. (1989) . Aided speech sensation levels received by severely hearing-impaired listeners. Aust JAudiol 11:27-36 .

Plomp R. (1988). The negative effect of amplitude com-presssion in multichannel hearing aids in the light of modulation-transfer function . J Acoust Soc Am 83:2322-2327 .

Pluvinage V (1989) . Clinical measurement of loudness growth. Hear Instr 39(3):28-32 .

Punch J, Parker C. (1981) . Pairwise listener preferences in hearing aid selection. J Speech Hear Res 24:366-374.

Purdy SC, Pavlovic CV (1992) . Reliability, sensitivity and validity of magnitude estimation, category scaling and paired-comparison judgments of speech intelligibil-ity by older listeners. Audiology 31:254-271 .

Schwartz D, Lyregaard PE, Lundh P (1988) . Hearing aid selection for severe/profound hearing losses . Hear J 41(2):13-17 .

Seewald RC, Ross M, Spiro MK. (1985) . Selecting ampli-fication characteristics for young hearing-impaired children . Ear Hear 6:48-53 .

Seewald RC, Zelisko DL, Ramji KV, Jamieson DG . (1991) . A Computer-assisted Implementation of the Desired Sensation Level Method for Electroacoustic Selection and Fitting in Children: Version 3.0 . DSL 3.0 user's manual . London, ON : University of Western Ontario.

Skinner MW (1988) . Hearing Aid Evaluation. Englewood Cliffs, NJ : Prentice Hall .

Studebaker GA. (1982) . Hearing aid selection ; an overview. In : Studebaker GA, Bess FH, eds. The Vanderbilt Hearing Aid Report . Upper Darby, PA: Monographs in Contemporary Audiology, 147-155.

Studebaker GA . (1992) . The effect of equating loudness on audibility-based hearing aid selection procedures . J Am Acad Audiol 3:113-118 .

Studebaker GA, Bisset JD, Van Ort VM, Hoffnung S. (1982) . Paired comparison judgments of relative intelli-gibility in noise. J Acoust Soc Am 72:80-92 .

Studebaker GA, Sherbecoe RL . (1992) . LASR3 (SSB): A Model for the Prediction of Average Speech Recognition Performance of Normal Hearing and Hearing-impaired Persons. Laboratory Report 92-02 Hearing Sciences Laboratory. Memphis: Memphis State University.

Tyler R . (1986) . Adjusting a hearing aid to amplify speech to MCL. Hear J 39(8):24-27 .

Van Buuren RA, Festen JM, Plomp R . (1995) . Evaluation of a wide range of amplitude-frequency responses for the hearing-impaired. J Speech Hear Res 38:211-221 .

Van Tasell DJ . (1993) . Hearing loss, speech, and hearing aids. J Speech Hear Res 36:228-244 .

Walden BE, Schuchman GI, Sedge RK. (1977) . The reli-ability and validity of the comfort level method of setting hearing aid gain . J Speech Hear Disord 48:264-273 .

Walker G, Dillon H. (1982) . Compression in Hearing Aids : An Analysis, a Review and Some Recommendations. NAL Report No . 90 . Sydney : National Acoustic Laboratories .

hearing aid selection for the 1990s: where to? · hearing aid selection for the 1990s: where to?...

Documents