British Journal of Audiology, 1999,33, 157-170

Use of a loudness model for hearing aid fitting: 11. Hearing aids with multi-channel compression

B.C.J. Moore, J.I. AlcBntara, M.A. Stone and B.R. Glasberg

University of Cambridge, UK

(Received 23 Ju1.y 1998; accepted 21 October 1998)

Abstract A model for predicting loudness for people with cochlear hearing loss was applied to the problem of the initial fitting of a multi-channel compression hearing aid. The fitting was based on two constraints: (1) The specific loudness pattern evoked by speech of a moderate level (65 dB SPL) should be reasonably flat (equal loudness per critical band), and the overall loudness should be similar to that evoked in a normal listener by 65-dB speech (about 23 sones for binaural listening); (2) Speech with an overall level of 45 dB SPL should just be audible in all frequency bands from 500 Hz up to about 4 kHz, provided that this does not require compression ratios exceeding about 3. These two constraints were used to deter- mine initial values for the gain, compression ratio and compression threshold in each channel of a multi- channel compression system. This initial fitting was based entirely on audiometric thresholds; it does not require suprathreshold loudness measures. The fitting method was evaluated using an experimen- tal fast-acting four-channel compression system. The initial fitting was followed by an adaptive proce dure to ‘fine tune’ the fitting, and the aids were then used in everyday life. Performance was evaluated by use of questionnaires and by measures of speech intelligibility. Although the fine tuning resulted in modest changes in the fitting parameters for some subjects, on average the frequency response shapes and compression ratios were similar before and after the fine tuning. The fittings led to satisfactory loudness impressions in everyday life and to high speech intelligibility over a wide range of levels. It was concluded that the initial fitting method gives reasonable starting values for the fine tuning.

Key words: hearing aids, hearing loss, compression, prescriptive methods, loudness

Introduction In a previous paper (Moore and Glasberg, 1998) we described an application of a loudness model (Moore and Glasberg, 1997) to the derivation of a prescription formula for linear hearing aids, i.e. aids not incorporating any form of compression (except for high-level limiting), and for aids incor- porating slow-acting automatic gain control, such as the dual front-end AGC system developed at Cambridge (Moore and Glasberg, 1988; Moore et al., 1991; Moore, 1993). Here, we describe the application of the same model to the fitting of

Address for correspondence: B.C.J. Moore, Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge CB2 3EB.

hearing aids with fast-acting multi-channel com- pression. The previous paper gives a description of the loudness model and of the concepts under- lying it; the reader is referred there for details.

It should be emphasized that the procedure described in this paper is intended only for setting initial parameter values for multi-channel com- pression. It is envisaged that some form of fine tun- ing to suit the individual patient would always be used after the initial fitting. One way of perform- ing such fine tuning is described in Moore et al. (1998a). A similar approach is described as part of this paper, where the extent of change performed during fine tuning is used as a measure of the ade- quacy of the initial fit. A virtue of the initial fitting procedure is that it is based entirely on the pure

0300-5364/99/330157+ 13 $03.50/0 0 1999 British Society of Audiology

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

158 B.C.J. Moore et al.

tone audiogram; no measures such as loudness scaling are required. This means that the initial fit can be determined more or less instantly from the audiogram (by use of a computer program), leav- ing the maximum possible time for fine tuning.

It should be noted that the initial fitting proce- dure described here is intended for multi-channel compression systems with low compression thresh- olds, and with compression ratios that are constant over a wide range of levels. The procedure would need to be modified for systems with high compres- sion thresholds or ‘curvilinear’ compression. How- ever, the same basic principles could be used to derive fitting procedures for such systems.

Rationale behind the initialfittingprocedure Loudness models can be used in a variety of ways for fitting hearing aids. One approach is to attempt to restore loudness perception to ‘nor- mal’, by means of multi-channel fast-acting Compression. With this approach, the frequency- dependent gain for any given sound should be such that the loudness perceived by the hearing- impaired listener is the same as would be per- ceived by a normally hearing listener without amplification. The loudness model is used to cal- culate the required gain in each channel on a moment-by-moment basis, over time scales of 1&50 ms (Kollmeier and Hohmann, 1995). How- ever, one might question whether restoration of loudness to ’normal’ is an appropriate goal for multi-channel compression systems (Moore, 1990; Stone et al.. 1999). We have argued that a reasonable alternative is to give good audibility for speech over a wide range of input sound lev- els, whilst maintaining listening comfort (Moore et al., 1998). Of course, if this can be done by restoring loudness to ‘normal’ then there is no conflict between these two goals. However, it is not clear that restoration of ‘normal’ loudness should be the primary goal, especially given the variability associated with loudness judgements.

A different approach is to apply frequency selective amplification so as to make the loudness of speech equal (as far as possible) in different

frequency bands. This approach has been used especially for the fitting of linear hearing aids. The basic idea is to place as much of the speech spectrum as possible above absolute threshold for a given overall loudness. This idea lies behind the NAL(R) formula for prescribing the gain of a linear hearing aid as a function of frequency (Byrne and Dillon, 1986), and it is also the basis of the ‘Cambridge’ formula (Moore and Glas- berg, 1998). In principle, the frequency-selective insertion gain (IG) prescribed by the Cambridge formula should result in a near-normal overall loudness (about 23 sones for binaural listening) for speech with an overall level of 65 dBSPL. It also should result in a specific loudness pattern that is flat (constant specific loudness or loudness per critical band) over the frequency range that is most important for speech, namely 5004000 Hz. The Cambridge formula is:

IG = HL x 0.48 + INT (1)

where HL stands for the absolute threshold in dB HL and INT is a frequency-dependent inter- cept. The value of INT is given in Table 1 for each audiometric frequency. Above 5 kHz, gains are limited to the value at 5 kHz.

The goal of achieving a flat specific loudness pattern is restricted to the frequency range 500-4000 Hz, since that is the frequency range that is most important for speech intelligibility. Below 500 Hz, the gains are reduced below those required to achieve a flat specific loudness pat- tern, in order to reduce masking of the speech by low-frequency environmental sounds, such as car noise and noise from ventilation and air-condi- tioning systems. These sounds have considerable energy at low frequencies which can have a mask- ing effect on the medium and high frequencies in speech. Also, when several people are talking in a typical room, the higher frequencies in the speech are absorbed by room reflections much more than the low frequencies. The reverberant sound field therefore contains predominantly low frequencies, which again can have a masking

Table 1. Values of the intercept (INT) in the Cambridge formula for each audiometric frequency ( k m

~_______

Frequency 0.125 0.25 0.5 0.75 1.0 1.5 2.0 3.0 4.0 5.0

INT -1 1 -10 -8 4 0 -1 1 -1 0 1

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

Fitting compression hearing aids 159

effect on higher frequencies. Such masking may be especially important for people with cochlear hearing loss, since they suffer more than nor- mally hearing people from the ‘upward spread of masking’ (Glasberg and Moore, 1986). The inter- cept values at 125 Hz and 250 Hz were chosen so as to reduce upward spread of masking while avoiding making the speech quality too ‘tinny’. In theory, for mild hearing losses, the intercept val- ues at low frequencies could lead to negative gains. In practice, where gains would be negative, they are set to zero.

Our proposed procedure for the initial fitting of multi-channel compression hearing aids is based on two constraints. The first is that for a speech- shaped noise with a level of 65 dBSPL (corre- sponding to typical conversational speech), the response should be as prescribed by the Cam- bridge formula. This ensures that most of the (65-dBSPL) speech spectrum is audible (giving a high value of the articulation index) whilst giving a comfortable loudness and an acceptable tonal quality (Moore and Glasberg, 1998).

The second constraint is that, as far as possible, speech should be audible (but less loud) over its entire dynamic range in each frequency band when the overall speech level is 45 dBSPL. The value of 45 dB SPL was chosen for two reasons. Firstly, it corresponds roughly to the lowest level of speech that a person normally needs to under- stand in everyday life (Pearsons et al., 1976; Killion, 1997). Secondly, empirical evidence sug- gests that if the low-level gains are set so as to make speech intelligible for lower levels than this, users complain that noises in the environ- ment arc too intrusive (Laurence et al., 1983; Moore et al., 1992a).

Taken together, these two constraints can be used to define the required gain, compression ratio and compression threshold in each channel of a multi-channel compression system. In prac- tice, however, the second constraint is modified to take into account the empirical finding that high compression ratios (greater than 2-3) have deleterious effects on speech intelligibility in quiet and in noise (Moore et al., 1992b; Plomp, 1994; Goedegebure et al., 1996; Verschuure and Dreschler, 1996). To avoid such effects, the maxi- mum compression ratio in any one channel is lim- ited to 2.92 (the reason for this particular value will become apparent later). This has the effect that, when the hearing loss at high frequencies is greater than 50-60 dB, the audibility of speech at

45 dBSPL is not fully restored. Notice that the main aim of this approach is not to restore loud- ness perception to ‘normal’. Rather, the goal is to give good audibility for low level speech and to maintain intelligibility and listening comfort over a wide range of sound levels. However, the approach should lead to a reasonably natural variation in loudness with sound level.

Application ofthe procedure to an eight-channel compression system As an example, we describe how the procedure can be applied to the initial fitting of an eight- channel compression system; the fitting proce- dure is similar to that used in a laboratory study of compression that did not employ any fine tun- ing procedure (Moore et al., 1999). Then, we show how the procedure can be adapted to the fitting of a four-channel system, and we describe a fine-tuning procedure using the four-channel system. The eight-channel system had channels centred at nominal frequencies of 250,500,1000, 1500,2000,3000,4000 and 6000 Hz (the actual centre frequencies were 183,488,849,1342,1952, 2684,3660 and 5917 Hz). The absolute thresholds at each of the nominal centre frequencies were denoted by HL(i), where i = 1-8. The absolute threshold is specified in dBHL. The crossover frequencies between the channels were 366,610, 1098,1586,2318,3050 and 4270 Hz.

Recall that the two constraints used for initial fitting are:

1. The gains for a speech-shaped noise with an overall level of 65 dB SPL should be as pre- scribed by the Cambridge formula, using Equa- tion (I). The gains at the nominal centre frequencies of the channels, i.e. 250,500,1000, 1500,2000,3000,4000 and 6000 Hz, are called Glin(l), Glin(2) . . . Glin(8). The values of INT(i) were-10,43,0,-1,1,-1,Oandl dB (seeTable 1).

2. As far as possible, speech should be audible over its entire dynamic range in each fre- quency band when the overall speech level is 45 dBSPL. In each frequency band, the dynamic range of the speech was assumed to extend from 18 dB below the RMS level to 12 dB above it; such an assumption is usually made in calculations of the articulation index (French and Steinberg, 1947; Pavlovic, 1987).

The steps to fulfil these two constraints required the following preliminary calculations:

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

160 B.C.J. Moore et al.

1. The absolute thresholds in dBHL at the

2.

centre frequencies of the bands, HL(i), were converted to equivalent free-field dBSPL. To achieve this, the following conversion factors in dB were added to the values of HL(i): 13,5, 4,2,0, -4, -5 and 4 dB, for the frequencies 0.25, 0.5,1,1.5,2,3,4 and 6 kHz, respectively. These conversion factors, Conv(i), were derived from the free-field absolute threshold values specified in I S 0 389-7 (1996), except that 2 dB was added to each threshold value to allow for the average difference between monaural and binaural thresholds (Moore et al., 1997). For a speech-shaped noise input with an over- all (free field) level of 65 dBSPL, the input level for each channel in the eight-channel compression system was calculated using the average speech-spectrum shape published in Byrne et al. (1994). The resultingvalues were:

I8(i) =61.3,60.1,57.6,49.5,47.0,44.1,45.1, 39.5 dB SPL, for i = 1-8

When speech with an overall input level of 45 dB SPL is applied, the level in the ith channel is 20 dB lower than for 65-dB speech, i.e. I8(i) - 20 dB. The speech minima in the ith channel have a level 18 dB below this. Thus the minima have a level I8(i) - 38 dB. The gain in the ith channel for that input level should be such that the input of (I8(i) - 38) is amplified to the absolute threshold. The absolute threshold in dBSPL in the ith chan- nel is HL(i) + Conv(i). Therefore, the gain required for the speech minima when the speech has an overall input level of 45 dB SPL is:

(2)

G(i)=HL(i)+Conv(i)-I8(i)+38 (3)

For that same input level (I8(i) - 38 dB), the output is:

I8(i) - 38 + HL(i) + Conv(i) -I8(i) + 38 = HL( i) + Conv( i)

For an input of I8(i), the output is 18(i) + Glin(i). Therefore, over an input range of 38 dB (from I8(i) - 38 to I8(i)), the output changes by 18(i) + Glin(i) - HL(i) - Conv(i). Therefore the compression ratio for the ith channel is:

(4)

CR8(i) = 38/{ 18(i) + Glin(i) -HL(i) - Conv(i)} ( 5 )

For example, if the hearing loss at 3 kHz is

HL(6) = 50 dB then Glin(6) = 23 dB. Using the values I8(6) = 44.1 dB and Conv(6) = -4 dB, the compression ratio calculated from Equation 5 is CR8(6) = 38/(44.1+ 23 - 50 + 4) = 1.80. In prac- tice, as described earlier, a limitation is placed on the amount of low-level gain. This was done by applying the following constraint: if the denomi- nator of Equation ( 5 ) was less than 13 then this quantity was set to 13. This gave a maximum com- pression ratio of 2.92. For a small hearing loss, Equation ( 5 ) can call for a compression ratio less than 1. In this case, the ratio was set to 1.

It was desired to apply the appropriate amplifi- cation for levels in the ith channel down to I8(i) - 38 dBSPL, in order to restore audibility of the minima in speech with an overall level of 45 dBSPL. The compression threshold in the ith channel, CT8(i), was therefore set to I8(i) - 38 dB SPL. This led to very low compression thresh- olds, ranging from 23.3 dBSPL to 1.5 dBSPL. In practice, such compression thresholds might be difficult to achieve in a hearing aid, due to micro- phone noise and the limited dynamic range of the analogue-to-digital converter in a digital hearing aid. However, the compression thresholds were higher (26-9.8 dBSPL) for the four-channel sys- tem described in the next section.

In summary, it can be seen how the procedure can be used to derive initial values for gains, compression ratios and compression thresholds from the audiometric thresholds, HL(i). It is straightforward to write a computer program to perform the calculation automatically. A fitting procedure exactly of this type was used in the study of Moore et al. (1999). Although that study was a laboratory study of speech intelligi- bility, and was not specifically concerned with the subjective quality of signals processed through the compression hearing aid, all sub- jects commented that the subjective quality of speech was acceptable.

Adaptation of the initial fittingprocedure to a four-channel system In this section, we describe how the procedure can be adapted to the initial fitting of a four-chan- nel compression hearing aid. We then go on to describe work using an experimental wearable digital four-channel system; the technical details of this system are very similar to those described in Moore et al. (1999) and Stone et al. (1999). After initial fitting, we conducted a fine tuning procedure similar to that described by Moore

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

Fitting compression hearing aids 161

et al. (1998). Changcs made on the basis of this fine tuning were used to assess the adequacy of the initial fit. We then briefly describe the out- come of trials of the aid.

In what follows, bands refers to frequency ranges used for frequency response shaping, whereas channels refers to frequency ranges used for compression. The bands had nominal centre frequencies (CFs) of 250,500,1000,1500,2000, 3000,4000 and 6000 Hz, and actual centre fre- quencies of 183,488,849,1342,1952,2684,3660 and 5917 Hz, the same as the channels of the eight-channel system described above. For the four-channel system, bands were combined as follows:

Channel 1: bands 1 and 2. Channel 2: bands 3 and 4. Channel 3: bands 5 and 6. Channel 4: bands 7 and 8.

When fitting the four-channel system, it was desired to retain the same degree of flexibility and accuracy in shaping the frequency response as was used with the eight-channel system. This was achieved in the following way. For a 65-dB speech-shapednoise input (i.e. for aninput to the first channel of 64.0 dBSPL), the gain in channel 1, G(ch1) was initially set to the mean of Glin(1) and Glin(2). To preserve the appropriate fre- quency-response shaping, the relative gains in bands 1 and 2 should be the same as for the eight- channel system. Therefore for band 1, the gain was increased by (Glin(1) - Glin(2)}/2 and for band 2, the gain was decreased by (Glin(1) - Glin(2)}/2. A similar procedure was applied for calculating the gains in the other channels and bands for a 65-dB speech-shaped noise input. For a speech-shaped noise with an overall input level of 65 dBSPL, this gave the same output as for the eight-channel system.

The compression ratio for channel n was set to be the mean of the compression ratios 2n -1 and 2n in the eight-channel system. The compression thresholds, CT4(i), were calculated as for the eight-channel system: CT4(i) = I4(i) -38 dBSPL. Note that the values I4(i) here are those appro- priate for the four-channel system, i.e. 64.0,57.3, 48.8 and 47.8 dB, for i = 14.

Fine tuning using an adaptiveprocedure The adaptive fine-tuning procedure was very sim- ilar to that described by Moore et al. (1998); the

reader is referred to that paper for details. The goal of the procedure is that speech at 85 dBSPL should be judged as ‘loud’, speech at 60 dBSPL should be judged as ‘quiet’, and speech at both lev- els should have an acceptable tonal quality. The procedure involves adjusting the low-level (50 dB) gain and high-level (80 dB) gain at low fre- quencies and at high frequencies. Call these gains G50(lf), GSO(hf), G8D(lf) and G80(hf). The low- frequency region was nominally centred at 750 Hz (i.e. roughly the boundary of the two lower chan- nels) and the high-frequency region was nomi- nally centred at 3000 Hz (i.e. roughly the boundary of the two upper channels).

On each trial, a single sentence was presented, with a level of either 60 dB SPL or 85 dB SPL. The sentences were taken from the ASL lists (MacLeod and Summerfield, 1990). The speech at 60 dB SPL was digitally filtered so as to have the long-term average speech spectrum pub- lished by Byrne et al. (1994). The speech at 85 dB SPL was additionally digitally filtered to give a slight boost to the medium and high fre- quencies, using the filter characteristic illus- trated in Fig. 1. This made the spectrum more like that of speech spoken with ‘raised effort’ (Pearsons et al., 1976). The overall level of the speech was 85 dBSPL after taking into account this filtering. In everyday life, speech rareiy occurs in the absence of some background noise, especially when the speech has a high level. To make the stimuli more representative of every- day life, a speech-shaped background noise was presented with each sentence. The noise level was 15 dB lower than that of the speech. The level of the noise was chosen so that it was clearly audible, but low enough to produce minimal masking of the speech. The sentences were stored on computer disk, and replayed via a 16-bit digital-to-analogue converter on a card inside the PC (SoundblasterB compatible). The output of the card was fed via an amplifier to a Bowers and Wilkins DM600 loudspeaker. The listener was seated about 70 cm from the loudspeaker in a small room (3 m x 1.5 m x 2.3 m) with moderate reverberation (reverberation time = 380 ms). There was a moderate level of background noise (about 45 dB SPL) generated by the computer fan. We chose these conditions deliberately (rather than testing in asound-attenuating cham- ber), to ensure that the procedure would work in a setting typical of that found clinically. The fol- lowing stages were used:

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

162 B. C. J. Moore et al.

10 I I I I , , > I I I I I l l I I I 1 1 1 1

9

E

7

6

5

m a 4

3

2

I

0

- m U

C

- .Fl

0 . 1 I

Frequency (kHz1

10

Fig. 1. Characteristic of the digital filter used to shape the spectrum of the85 dB SPL speech. The speech level was specified after this filtering had been applied.

1. A sentence was presented at an overalllevel of 85 dBSPL. The target was that this should be judged as ‘loud’. If its loudness was judged as ‘comfortable’ or lower, all post-AGC gains (i.e. the linear gains following the compres- sion processing) were increased by the same amount; the compression ratios were not altered. If the loudness was judged as ‘very loud’ or higher, all post-AGC gains were decreased by the same amount. The step size in gain was as described by Moore et al. (1998). This process was repeated until the target judgement of ‘loud’ was obtained.

2. A sentence was presented at an overall level of 60 dBSPL. The target was that this should be judged as ‘quiet’. If its loudness was judged as ‘comfortable’ or higher, the low-level gains, G50(lf) and G5O(hf), were both decreased by the same amount, leaving the high-level gains unaltered (details of exactly how this was achieved are given below; note that this altered the compression ratios). If its loudness was judged as ‘very quiet’ or ‘inaudible’, the low-level gains, G50(lf) and GSO(hf), were both increased by the same amount, leaving

the high-level gains unaltered. The step size in gain was as described by Moore et al. (1998). This process was repeated until the target judgement of ‘quiet’ was obtained.

3. A sentence with a level of 85 dBSPL was pre- sented. The listener was required to judge the quality of the speech on a seven-point scale going from ‘uncomfortably tinny’ to ‘uncom- fortably boomy’. The response ‘neither tinny nor boomy’ was deemed optimal. The post- AGC gains at 750 and 3000 Hz were adjusted (in opposite directions) until the target response was achieved. For example, if the sound was rated as ‘tinny’, the post-AGC gain at 3000 Hz was decreased by 1 dB and the post- AGC gain at 750 Hz was increased by 1 dB; steps of 2 and 3 dB were used for responses that deviated more from the target response. Examples of such adjustments are shown in the upper panel of Fig. 2. The solid curves illustrate typical starting gain functions, mea- sured in a 2-cc coupler using speech-weighted noise with levels of 60 or 85 dBSPL as the input (Rastronics Portarem 2000 system). The gains are lower for the higher noise level.

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

Fitting compression hearing aids 163

Following a response of ‘very tinny’, the post- AGC gain at 3000 Hz would be decreased by 2 dB, whereas the gain at 750 Hz would be increased by 2 dB, with smooth changes in between, giving the dotted curves. The changes are similar at the two levels. Follow- ing a response of ‘very boomy’, the post-AGC gain at 3000 Hz would be increased by 2 dB, whereas the gain at 750 Hz would be decreased by 2 dB, with smooth changes in between, giving the dashed curves.

4. A sentence with a level of 60 dBSPL was pre- sented. The listener was required to judge the quality of the speech on a seven-point scale going from ‘uncomfortably shrill’ to ‘very muffled’. The response ‘neither shrill nor muf- fled’ was deemed optimal. The low-level gains at 750 and 3000Hz, GSO(1f) and GSO(hf), were adjusted (in opposite directions) until the tar- get response was achieved, without changes in the high-level gains, G80(lf) and G80(hf). For example, if the sound was rated as ‘muffled’, GSO(hf) was increased by 1 dB and GSO(1f) was decreascd by 1 dB; steps of 2 and 3 dB were used for responses that deviated more from the target response. Examples of such adjustments are shown in the lower panel of Fig. 2. Note that the adjustments (shown as the dotted curve for a response of ‘very shrill’ and the dashed curve for a response of ‘very muf- fled’) mainly affect the gain for the 60-dB input level.

This whole sequence was repeated until stable values were achieved and thc target responses were given consistently. Generally, only one rep- etition was requircd, although two repetitions were sometimes performed. The whole proce- dure typically took about five minutes per ear.

Note that only four parameters were adjusted. It is important to minimize the number of parame- ters that must be adjusted in the clinic, otherwise the fitting procedure becomes unwieldy and unre- liable, and takes an excessive time. However, we wished to apply a fine tuning procedure with four adjustable parameters to a four-channel compres- sion system that has. in principle, eight adjustable parameters (the values of G50 and G80 in each of four bands; this assumes that the crossover fre- quencies between bands and the compression thresholds remain fixed). The following para- graphs describe how the fine tuning procedure was adapted to the four-channel system.

I I 30.

250 500 1000 2000 4000 8000 Frequency (Hz)

30 ~

m

85-dB SW noise

250 500 1000 2000 4000 8000 Frequency (Hz)

Fig. 2. Illustration of the changes infrequency- gain characteristic made in response to judge- ments of the tone quality of speech. The gain measured in a 2-cc coupler is shown f o r a speech- weighted noise input with a level o f 60 or 85 d B SPL. The upper panel shows changes made in response to judgements of 85 d B SPL speech on a t inny-boomy scale. The solid curves show the starting frequency-gain characteristics while the dotted and dashed curves show the characteristics following responses o f ‘very tinny’ and ‘very boomy’, respectively. The lower panel shows changes made in response to judgements of 6OdB SPL speech on a shrill-muffled scale. The solid curves show the starting frequency-gain charac- teristics, whereas the dotted and dashed curves shows the characteristics following responses o f ‘very shrill’and ‘very muffled’, respectively.

Implementation of the adaptiveprocedure on the four-channel system In some parts of the proccdurc, the gain in a given frequency region is adjusted by the same amount for all levels, i.c. the compression ratio is not changed. In other parts of the procedure, the gain

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

164 B.C.J. Moore et al.

for a 50-dB input may be altered, whereas the gain for an 80-dB input is left unchanged. This requires a change of the compression ratio (CR). To see how this was achieved, consider the fol- lowing. The gains at the two input levels, G50 and G80 are related to the CR by:

GSO = G80 + 30 - 30/CR

For example, if the CR is 2, then the gain G50 is 15 dB more than the gain G80. To increase G50 by Y dB relative to G80, the CR needs to be increased as follows:

(6)

CR,,, = 30/[(30/CRold) - Y] (7)

For example, if CR,,, is 2, and it is desired to increase G50 by 2 dB relative to G80 then CR,,, is 2.31.

The adaptive procedure involves adjustment of gains and compression ratios only at the two reference frequencies, 750 Hz and 3000 Hz. This gives two new compression ratios, CR(If),,, and CR(hf),,,, (one eachfor the lower and upper fre- quency ranges) and two values of the post-AGC gain adjustment, which will be denoted Gout(1f) and Gout(hf). Consider now how to extend these for use in the four-channel system. The compres- sion ratios determined by the initial fitting proce- dure often change progressively with frequency (typically being greater at high frequencies). To preserve the form of the changes, the following is done. The amount of compression is represented by the amount by which the CR exceeds unity; if CR = 1, there is no compression. Therefore, one can define ‘scaling factors’ for the compression as:

and

W h f ) = (CR(hf),,, - l)/(CR(hf),,,j - 1) (9)

For example, if CR,,, and CR,,, are 2 and 2.31 then the scaling factor is 1.31. In practice, the values of the compression ratios are not allowed to be less than 1.001, to prevent the denominator in equations 8 and 9 from becoming zero. The appropriate scaling factor is applied to the old compression ratios, CROld(i), to give new compression ratios, CR,,,,(i), for each band. Specifically:

CR(2),,, = SF(lf){ CR(2),,, - 1 } + 1 (1 1)

CR(4),,,=SF(hf){CR(4),,,- 11 + 1 (13)

Say, for example, that CR(3),,, and CR(4),,, are 1.7 and 2.3,respectively, so that CR(hf),,, = 2. If the adaptive procedure calls for CR(hf),,, = 2.31 then:

SF(hf) = 1.31, CR(3),, = 1.92 andCR(4),,, = 2.7 (14)

In practice, this procedure may sometimes call for a compression ratio greater than 2.92. If this happens, the value is limited to 2.92. However, it is desirable to preserve the relative amounts of compression in adjacent channels (channels 1 and 2 or 3 and 4). This can be done in the follow- ing way. Before the adaptive procedure starts, two compression factors are calculated:

CF(1Q = {CR(2)-I}/{CR(l) -1) and CF(hQ = {CR(4)-1}/{CR(3)-1}

If, during the adaptive procedure, CR(2) is set to 2.92 then CR(1) is set to (1.92/CF(lf)) + 1.

If, during the adaptive procedure, CR(1) is set to 2.92 then CR(2) is set to (1.92*CF(lf)] + 1.

If, during the adaptive procedure, CR(4) is set to 2.92 then CR(3) is set to (1.92/CF(hf)] + 1.

If, during the adaptive procedure, CR(3) is set to 2.92 thenCR(4) is set to (1.92*CF(hf)] + 1.

If any compression ratio would be given avalue less than 1.001, its value is set to 1.001. The result- ing compression ratios are denoted CR(i),,,.

In summary, the adaptive fitting procedure based on the adjustment of four parameters, can be used to adjust the compression ratios and post- AGC gains in all four channels of a four-channel compression system. This is done in such a way as to preserve the patterning of the compression ratios across channels, as determined by the ini- tial fitting procedure. Although the procedure appears complex, it is easy to implement in auto- matic form, as a computer program.

Outcome of the fitting procedure Eight subjects with bilateral hearing loss of cochlear origin were fitted with four-channel fast-acting compression hearing aids, using the

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

Fitting compression hearing aids 165

initial procedure followed by the adaptive proce- dure. The audiograms of the subjects are given in Table 2. The hearing aids were experimental dig- ital hearing aids - called Audallions - manufac- tured by Audiologic. Full details of how the aids were programmed are given in Moore et al. (1999) and Stone et al. (1999). The main interest here is in the outcomes of the fitting procedurcs. If there are consistent differences between the parameter settings of the initial fitting and the fit- ting following fine tuning with the adaptive pro- cedure, this would indicate that the initial fitting procedure needs adjustment. On the other hand, if the average parameter values determined by the initial fitting procedure are reasonably close to those obtained after fine tuning, this would indicate that the initial fitting procedure does not lead to systematic errors, and that it provides a good starting point for the fine tuning.

After the experiment was completed, it was discovercd that all programmed insertion gains were actually 3 dB higher than intended. If the intended gains prescribed by the initial fitting procedure were appropriate, then one would expect the gains to be slightly reduced as a result of the adaptive procedure. This did, in fact hap- pen. For a 65 dBSPL speech-shaped noise input,

the decrease in gain, averaged across the fre- quencies 0.25, 0.5, 1, 1.5, 2, 3 4 and 6 kHz and across the 16 ears, was 3.7 dB. This means that the average final gain was within 1 dB of the value prescribed by the Cambridge formula. The largest decrease in gain averaged across frequen- cies for any individual ear was 5.5 dB, whereas the largest increase was 3 dB. In any case, it should be noted that individual preferences in terms of overall gain could easily be accommo- dated by adjustment of a volume control designed to affect post-AGC gains.

More important issues are: did the overall fre- quency response shape and the compression ratios change markedly as a result of the adaptive fitting procedure? The first of these questions was addressed in the following way. For each ear of each subject, the mean change in gain as a result of the fine tuning, averaged across the fre- quencies 0.25,0.5,1.1.5,2,3,4 and 6 kHz, was cal- culated. The gains here are those for a 65 dBSPL speech-shaped noise input. This mean change was subtracted from the change at each channel centre frequency. Any deviations of the resulting numbers fromzero indicate changes in frequency response shape as a result of the fine tuning. Figure 3 shows histograms of the deviations for

Table 2. Audiometric thresholds for each ear of the eight subjects used in the trial

Subjedear Frequency (kHz)

0.25 0.5 1.0 1.5 2.0 3.0 4.0 6.0 -

DPIR 10 15 40 50 40 50 55 45 DPIL 5 5 25 40 40 55 50 45 ETIR 50 55 55 55 55 60 70 60 ETIL 55 60 70 65 65 70 75 70 GWIR 40 50 65 60 60 60 65 55 GWIL 35 40 60 60 55 55 60 55 RWIR 30 30 45 50 50 55 60 55 R W L 60 60 65 65 65 60 65 60 VWIR 40 55 70 75 65 75 85 75 VWIL 45 60 70 65 70 55 50 60 PYIR 30 40 45 50 55 55 55 65 PYlL 35 40 45 50 60 60 65 70 SZ/R 40 40 35 35 40 40 45 60 SZIL 35 35 35 45 45 45 55 70 JWIR 15 25 40 40 40 55 65 65 JWIL 15 30 40 40 45 55 65 65

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

166 B.C.J. Moore et al.

Channel i Channel 2 Channel 3 Channel 4 100

m

0 C

L

0

90

00

70 - 60 - 50 -

- -

-4-

a, 40 -

a, 30 - a, 20 n

c, C

0 L -

10 - r UI

0 0 0 0 0 0 0 0 0 0 0 0 c,+Jc, w u c , w w c , W + c ,

C U - d N - d N - d C U - n I I I I I I I I

G a i n d e v i a t i o n (dB)

Figure 3. Histograms showing deviations between the gains prescribed b y the Cambridge formula and the gains achieved after application of the adaptive fine tuningprocedure, for a 65 d B SPL speech- shaped noise input. Gain deviations are shown separately for each channel of the four-channel system. For each ear of each subject, the mean change in gain as a result of the fine tuning averaged across the frequencies 0.25, 0.5,1,1.5,2, 3, 4 and 6 kHz, was calculated. This mean changes was subtractedfrom the change at each channel centre frequency. A n y deviations of the resultingnumbers from zero indicate changes infrequency response shape as a result of the fine tuning.

the frequencies 0.5, I, 2 and 4 kHz. It may be seen that, in the majority of cases, the gain deviated by less than 1 dB. In other words, changes in fre- quency response shape were generally very small. In the worst case (GW, right ear), the fre- quency response shape after fine tuning differed from that determined by the initial fitting proce- dure by about 5 dB across the range 0.25-6 kHz; the final gain was about 0.5 dB higher than the initial gain at 0.25 kHz and 4.4 dB lower than the initial gain at 6 kHz. Averaged across all 16 ears, the mean deviation was -0.1 dB at 0.5 kHz, -0.1 dB at 1 kHz, +0.2 dB at 2 kHz and -0.3 dB at 4 kHz. These mean deviations are all very small, indicating that the initial fitting procedure led to frequency response shapes that were not system- atically changed by the adaptive fine tuning pro- cedure.

To assess the extent to which compression ratios changed as a result of the adaptive fine tun- ing, we calculated the ratio of the Compression

ratio for a givenchannel at the end of the fine tun- ing, called CR(final), and the compression ratio determined by the initial fitting procedure, called CR(start). This was done separately for each ear of each subject. Figure 4 shows histograms of the ratios for each of the four channels. Ratios close to 1 indicate that the compression ratio following fine tuning, CR(final), was close to the compres- sion ratio determined by the initial fitting proce- dure, CR(start). Many of the ratios were in the range 0.9-1.1, indicating little change in compres- sion ratio as a result of the fine tuning. The largest ratio was 1.34 (2.92/2.17), whereas the smallest ratio was 0.56 (1.4N2.63). The geometric mean ratios were 1.06 for channel 1,1.03 for channel 2, 1.02 for channel 3, and 1.00 for channel 4. The mean ratios are all close to 1, indicating that the initial fitting procedure is not systematically in error.

Given that the frequency-response shapes and compression ratios were not changed markedly

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

90 (0

0 C

L 3

0 0

0

ol * a, 0

t3 80

70

0 60

- -

-

u- 50 -

0) 40 - m

30 -

20 - a

- Channel 1 Channel 2 Channel 3 Channel 4 -

* . . . . . . .

167

O d d 0 4 - O d d 0-4.- I l l I l l I I l l

L D U l d L o o l d mch.!I h-?m‘: . . . . . . . . . 00-4 00.- 00.- 0 0 4

CR (final) / CR (start)

Fig. 4. Histograms of ratios of the compression ratios obtained after the adaplive fine tuningprocedure (CR(fina1) ) and the compression ratiosprescribed by the initialfittingprocedure (CR(start) ). Results arc shown separately for each channel of the four-channel system. Ratios close to I indicate that the compression ratio did not change markedly as a result of the adaptive fine tuningprocedure.

by the adaptive fine-tuningprocedure, one might question whether the procedure is necessary or justified. Certainly, for the majority of the sub- jects tested, the initial fit, based only on the audiogram, would have given satisfactory results. However, in two subjects out of the eight tested, the compression ratios did change markedly as a result of the fine tuning. This sug- gests that it is not safe to rely on the initial fitting alone. In any case, a hearing aid user is likely to view a hearing aid more favourably if it is clear to them that the aid has been adjusted to suit them as an individual.

Further checks of thefittings usingfield trials Once the aids had been fitted as described above, subjects wore them for a period of at least two weeks in their everyday lives. The main purpose of this part of the study was to compare the four- channel fast-acting compression system with other forms of automatic gain control, including the slow-acting single-channel dual front-end AGC system (Moore and Glasberg, 1988; Moore

et al., 1991). Full results are reported elsewhere (Stone et al., 1999). However, it is of interest to consider here thc informal reports of the sub- jects, their responses to a modified version of the APHAB questionnaire (Cox and Alexander, 1995), and the results of speech intelligibility tests.

All subjects reported that the aids gave accept- able loudness in most everyday listening situa- tions. They were not unduly bothered by low-level background sounds, and intense sounds were generally not uncomfortably loud. This is confirmed by the results of the APHAB test (Cox and Alexander, 1995), the mean results of which are shown in Fig. 5. The test requires subjects to rate how often they have problems in specific situations, such as ‘Unexpected sounds, like a smoke detector or alarm bell are uncomfort- able’ or ‘When Z am having a quiet conversation with a friend, I have difficulty understanding’. Response alternatives range from ‘Always (99%)’ to ‘Never (1%)’. Low percentages indi- cate good performance. The results are grouped

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

R. C. .I. Moore et al.

loo 90 i- In E 80 a, rl

0 L

60 4-

U L

a a, 20

10

EC R V 1 BN

APHAB s u b s c a l e

Fig. 5. Mean results of the A P H A B test. The scores indicate the percentage of times thatproblems are encountered in specific situations. Low percentages indicate good performance. The results are grouped into four sub-scales: ease of communication (EC); understanding in reverberent environments (RV); understanding in background noise (BN); and aversiveness of sounds (AV). A totalscore is also given.

into four sub-scales: ease of communication (EC), understanding in reverberant environ- ments (RV), understanding in background noise (BN) and aversiveness of sounds (AV). A total score is also given.

Two scores are of particular interest here. The score for EC is based partly on questions about the ability to hear low-level speech in quiet situa- tions. Cox and Alexander (1995) reported that the mean frequency of problems for the EC scale for subjects wearing hearing aids was 24%. Our mean score was 21%, slightly, but not materially better. The score for AV primarily reflects aver- sive response to intense environmental sounds. Cox and Alexander (1995) reported that the mean frequency of problems for the AV scale for subjects wearing ‘conventional analogue’ hear- ing aids was 55%. Our subjects gave a mean score of 31%, markedly lower than reported by Cox and Alexander (1995). This indicates that the compression systems in our aids effectively pro- tected the users from unpleasantly loud sounds. In fact, the AV scores for our subjects are only slightly higher than the mean value of 26% reported by Cox and Alexander (1995) for

unaided listening. The APHAB results support the idea that our method of fitting the compres- sion systems led to satislactory results in every- day life, as least with regard to the loudness of sounds.

We stated in the introduction that a reasonable goal for a multi-channel compression system is to give good audibility for speech over a wide range of input sound levels while maintaining listening comfort. To assess whether our fitting method did indeed provide good audibility for speech over a wide range of levels, we measured the intelligibility of speech in quiet for AB word lists (Boothroyd, 1968) at levels of 50 and 80 dBSPL (binaurally aided). The scores were very high for all subjects at both levels. The mean scores were 93% at 50 dBSPL and 98% at 80 dBSPL.

Summary and conclusions The fitting of multi-channel compression systems involves two stages: initial fitting based on audio- metric and/or psychoacoustic data, and fine tun- ing to suit the individual. We have described an initial fitting procedure which uses only informa- tion from the pure tone audiogram; it does not

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

Fitting Compression hearing aids 169

require information about loudness growth at different frequencies. The initial fitting proce- dure is based partly on the use of a loudness model. This model was used to derive the ‘Cam- bridge’ formula, which gives target insertion gains when the input is a speech-shaped noise with a level of 65 dBSPL (Moore and Glasberg, 1998). The initial fitting procedure also calculates the low-level gains that would be required to give full audibility for speech with an overall level of 45 dBSPL, subject to the constraint that the com- pression ratio in any channel of the compression system should not exceed 2.92. Together, these two aspects of the fitting procedure definc thc gains, compression ratios and compression thresholds for each channel of a multi-channel compression system. Examples were given for an eight-channel system and a four-channel system.

Previously, we described an adaptive proce- dure for the fine tuning of a two-channel com- pression system (Moore et al., 19Y8). Here, we described how this procedure was modified to apply to a four-channel system, and we described the results of applying it to each ear of eight sub- jects with cochlear hearing loss. The results showed that, following application of the adap- tive fine tuning, the mean gain for a speech- shaped noise with a level of 65 dBSPLwas within 1 dB of that prescribed by the Cambridge for- mula (after allowing for the initial 3-dB error in calibration of gains). Also, the adaptive proce- dure generally resulted in only small changes in frequency response shape and in compression ratios. This indicates that thc initial fitting proce- dure was not systematically in error; it gave appropriate parameter starting values from which to begin the fine tuning procedure.

Trials of hearing aids fitted using the proce- dure indicated very satisfactory results. Scores from the APHAB test indicated low frequencies of problems in understanding quiet speech and in aversion from intense sounds. Speech intelligibil- ity measures indicated that speech in quiet could be understood very well for sound levels of 50 dBSPL and SO dBSPL.

Acknowledgements This work was supported by the European Union (SPACE project) and the Medical Research Council. We thank Audiologic, Resound and Danavox for providing the Audallion hearing aids, together with the hardware and interface software necessary to program the aids. We also

thank an anonymous reviewer and a ubiquitous, anonymous antipodean for helpful comments on an earlier version of this paper.

References Boothroyd A. Developments in speech audiome-

try. Sound 1968; 2: 3-10, Byrne D, Dillon H. The National Acoustic Labo-

ratories’ (NAL) new procedure for selecting the gain and frequency response of a hearing aid. Ear Hear 1986; 7: 257-65.

Byrne D,Dillon H, Tran Ket al. An international comparison of long-term average speech spec- tra. J Acoust Soc Am 1994; 96: 2108-20.

Cox RM, Alexander GC. The abbreviated pro- file of hearing aid benefit. Ear Hear 1995; 16: 176-86.

French NR, Steinberg JC. Factors governing the intelligibility of speech sounds. J Acoust SOC Am 1947; 19: 9C119.

Glasberg BR, Moore BCJ. Auditory filter shapes in subjects with unilateral and bilateral cochlear impairments. J Acoust SOC Am 1986;

Goedcgebure A, Hulshof M, Maas AJJ, Ver- schuure J. The effects of syllabic compression on speech intelligibility in hearing impaired. In: Kollmeier B, ed. Psychoacoustics, Speech and Hearing Aids, Singapore: World Scien- tific, 1996; 165-70.

I S 0 389-7. Acoustics - Reference zero for the calibration of audiometric equipment. Part 7: Reference threshold of hearing under free- field and diffuse-field listening conditions. Geneva: International Organization for Stan- dardization, 1996.

Killion MC. Hearing aids: Past, present and future: Moving toward normal conversations innoise. Br JAudioll997; 31: 141-8.

Kollmeier B, Hohmann V. Loudness estimation and compensation employing a categorical scale. In: Manley GA, Klump GM, Koppl C, Fast1 H, Oeckinghaus H, eds. Advances in Hearing Research, Singapore: World Scientific Publishers, 1995; 441-51.

Laurence RF, Moore BCJ, Glasberg BR. A com- parison of bchind-the-ear high-fidelity linear aids and two-channel compression hearing aids in the laboratory and in everyday life. Br J Audio1 1983; 17: 31-48.

MacLeod A , Summerfield 0. A procedure for measuring auditory and audio-visual speech- receDtion thresholds for sentences in noise:

79: 1020-33.

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.

170 B.C.J. Moore et al.

rationale, evaluation, and recommendations for use. Br J Audiol 1990; 24: 29-43.

Moore BCJ. How much do we gain by gain con- trol in hearing aids? Acta Otolaryngol 1990; Suppl. 469: 250-6.

Moore BCJ. Signal processing to compensate for reduced dynamic range. In: Beilin J, Jensen GR, eds. Recent Developments in Hearing Instrument Technology, Copenhagen: Stougaard Jensen, 1993; 14745.

Moore BCJ, Glasberg B R. A comparison of four methods of implementing automatic gain con- trol (AGC) in hearing aids. Br J Audio1 1988;

Moore BCJ, Glasberg BR. A model of loudness perception applied to cochlear hearing loss. Auditory Neurosci 1997; 3: 289-311.

Moore BCJ, Glasberg BR. Use of a loudness model for hearing aid fitting. I. Linear hearing aids. Br J Audiol 1998; 32: 301-19.

Moore BCJ, Alcintara JI, Glasberg BR. Devel- opment and evaluation of a procedure for fit- ting multi-channel compression hearing aids. Br J Audiol 1998; 32: 177-95.

Moore BCJ, Glasberg BR, Stone MA. Optimiza- tion of a slow-acting automatic gain control system for use in hearing aids. Br J Audiol 1991; 25: 171-82.

Moore BCJ, Johnson JS, Clark TM, Pluvinage V. Evaluation of a dual-channel full dynamic range compression system for people with sen- sorineural hearing loss. Ear Hear 1992a; 13: 349-70.

22: 93-104.

Moore BCJ, Lynch C, Stone MA. Effects of the fitting parameters of a two-channel compres- sion system on the intelligibility of speech in quiet and in noise. Br J Audiol 1992b; 26: 369-79.

Moore BCJ, Glasberg BR, Baer T. A model for the prediction of thresholds, loudness and par- tial loudness. J Audio Eng SOC 1997; 45: 224-40.

Moore BCJ, Peters RW, Stone MA. Benefits of linear amplification and multi-channel com- pression for speech comprehension in back- grounds with spectral and temporal dips. J Acoust SOC Am 1999; 105: 400-11.

Pavlovic C. Derivation of primary parameters and procedures for use in speech intelligibility predictions. J Acoust SOC Am 1987; 82: 413-22.

Pearsons KS, Bennett RL, Fidell S. Speech Lev- els in Various Environments. Report No. 3281. Cambridge, MA: Bolt, Beranek and Newman, 1976.

Plomp R. Noise, amplification, and compression: Considerations of three main issues in hearing aid design. Ear Hear 1994; 15: 2-12.

Stone MA, Moore BCJ, Alcintara JI, Glasberg BR. Comparison of different forms of com- pression using wearable digital hearing aids. J Acoust SOC Am 1999; (submitted).

Verschuure J, Dreschler WA. Dynamic compres- sion in hearing aids. In: Kollmeier B, ed. Psychoacoustics, Speech and Hearing Aids, Singapore: World Scientific, 1996; 15344.

Bri

tish

Jour

nal o

f A

udio

logy

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

Seri

als

Uni

t - L

ibra

ry o

n 10

/04/

13Fo

r pe

rson

al u

se o

nly.