Secondary Logo

Journal Logo

Proprietary fitting algorithms compared with one another and with generic formulas

Keidser, Gitte; Brew, Christopher; Peck, Andrea

doi: 10.1097/01.HJ.0000293014.56004.ee
Article
Free

Compared with generic hearing aid fitting prescriptions, much less information is available about proprietary algorithms developed by manufacturers for their hearing aids. After studying several widely used proprietary prescriptions, the authors report on how they differ from or are similar to each other and to the leading generic procedures.

Gitte Keidser, PhD, is Senior Research Scientist, National Acoustic Laboratories (NAL), Sydney, Australia. Christopher Brew, MClinAud, is a research audiologist at NAL. Andrea Peck, BS, is an AuD candidate at Central Michigan University. Correspondence to Dr. Keid-ser at National Acoustic Laboratories, 126 Greville Street, Chatswood, NSW 2067, Australia; e-mail: gitte.keidser@nal.gov.au.

Technology sometimes advances faster than our understanding of how to best apply it. This is certainly the case with hearing aids. In recent years, hearing aids have been introduced with parameters that no generic procedures prescribe. Therefore, hearing aid manufacturers are forced to consider how their new products are best fitted to the hearing aid user before releasing them. While some manufacturers (e.g., Bernafon, Siemens, and Unitron), recommend using established prescription procedures such as NAL-NL11 and DSL[i/o]2 as a baseline for fitting their devices, others have introduced their own proprietary fitting algorithms (e.g., Oticon, Phonak, GN ReSound, and Widex).

Typically, generic fitting procedures are based on thorough research conducted by people not associated with hearing aid manufacturers, and the results and evaluation data are published for scientific scrutiny. In contrast, the proprietary fitting algorithms are developed based on research conducted by, or overseen by, the manufacturers. Although manufacturers commonly explain the basic rationale behind the proprietary algorithms to clinicians through brochures or publications,3,4 these presentations neither present the fitting targets, discuss how they vary with hearing loss, nor compare the proprietary prescription with other fitting algorithms. In fact, very little information is available on how similar or different the amplification characteristics recommended by each manufacturer are from those recommended by other companies, or how their suggestions compare with the better known prescriptive targets.

At the National Acoustic Laboratories (NAL), we attempted to investigate this by comparing the insertion gain targets prescribed by four proprietary fitting algorithms (DigiFocus II [Oticon], Claro [Phonak], Danalogic [GN ReSound], and Senso Diva [Widex], respectively) and by two generic prescription algorithms (NAL-NL1 and DSL[i/o]) for five diverse audiogram configurations. In particular, we studied how the targets vary in prescribed overall gain and in prescribed shape across the low and high frequencies. To better understand the characteristic of each algorithm we also determined whether the algorithm was dependent or independent of the hearing loss configuration.

Back to Top | Article Outline

STUDY METHODOLOGY

Table 1 summarizes the fitting algorithms selected for this study and their aims. In broad terms, the algorithms seem either to aim at normalizing loudness or compensating for loudness recruitment (DSL[i/o],2 Phonak,5 GN ReSound,6 and Widex7), or at maximizing speech intelligibility (NAL-NL11 and Oticon4).

Table 1

Table 1

Information about the DSL[i/o] and the NAL-NL1 algorithms has been published in peer-reviewed journals1,2 and some validation data are available for both methods.8-10 These two algorithms represent the two main categories (loudness normalization and speech-intelligibility maximization) and are known to prescribe very different targets.1 If the proprietary methods achieve their stated aims, we would expect them to prescribe targets that are close in shape to the generic method within their category.

Back to Top | Article Outline

Extracting insertion gain targets

We extracted the insertion gain targets for the six fitting algorithms in Table 1 for each of the five audiograms listed in Table 2. The audiograms consist of a flat loss, a reverse sloping loss, a gently sloping high-frequency loss, a steeply sloping high-frequency loss with normal threshold across the low frequencies, and a steeply sloping high-frequency loss with a mild low-frequency hearing loss.

Table 2

Table 2

We used stand-alone software to extract the insertion gain curves for the two generic algorithms (see Table 1). Specifically, for the NAL-NL1 software, pure-tone targets were derived for a three-channel device, considering a unilateral fitting and multi-channel limiting.* We converted the real-ear aided gain (REAG) targets derived from the DSL software to real-ear insertion gain (REIG) by subtracting the real-ear unaided gain (REUG) values from Dillon.11

For the proprietary algorithms, we extracted the insertion gain curves from the manufacturers' fitting software after entering the audiograms into NOAH (see Table 1 for an overview). To get to the simulated insertion gain curves in the manufacturers' software, we often were required to select the device, acoustic parameters, and adaptation level.

Insertion gain should not be dependent on device or acoustic parameters. However, in OtiSet, PFG, and Compass (device only), these factors did affect the simulated insertion gain curves. Therefore, we ensured that (1) the combination of device and acoustic parameters was appropriate for the audiogram in question and largely in agreement with the manufacturers' recommendation and (2) the selected parameters were comparable across software. For consistency, we always selected a behind-the-ear (BTE) device. Where possible, we used the same device for all hearing loss configurations. However, if a device was not suitable for a particular loss, we selected an appropriate device according to the manufacturer's recommendation (see Table I for an overview).

We selected regular tubing (2 mm) for all audiograms with the DigiFocus II and Claro devices. Table 3 shows an overview of the vent sizes selected for the two devices to assure comparability. Note that the choice of earhook did not affect the insertion gain targets. Finally, if adaptation level was a parameter, we always selected the highest level, assuming this reflected the most desirable target.

Table 3

Table 3

Two manufacturers (Oticon and Phonak) have introduced algorithms specifically designed for fitting steeply sloping (ski slope) losses. We followed the companies' recommendations in using their algorithms for the two steeply sloping losses. Finally, we converted the REAG values displayed in Phonak's fitting software to REIG by zeroing the open-ear response.

For all six algorithms, we obtained the insertion gain targets at frequencies from 250 to 6000 Hz for input levels of 50, 65, and 80 dB SPL. In cases where the insertion gain could not be derived for input levels of 50, 65, or 80 dB SPL, the values for these levels were interpolated or extrapolated from the values that could be extracted from the software. We did this on the assumption that the prescribed pure-tone compression threshold is no more than 50 dB SPL across frequencies.

From Widex's fitting software (Compass), insertion gain values could be extracted only for 500, 1000, 2000, and 4000 Hz as a function of input level in dB HL. To convert the input levels from dB HL to dB SPL, we shifted the input levels down by 7 dB for 500 Hz, 4 dB for 1000 Hz, 8 dB for 2000 Hz, and 1.5 dB for 4000 Hz.

Back to Top | Article Outline

FINDINGS

Differences in overall gain

To compare differences in overall gain, we averaged the prescribed insertion gains at 500, 1000, 2000, and 4000 Hz. We did this separately at each of three input levels (50, 65, and 80 dB) for each hearing loss configuration and prescription. The results of this analysis are presented in Figure 1.

Figure 1

Figure 1

Figure 1 shows that DSL[i/o] provides more gain than any other procedure at all three input levels for all five hearing losses. Apart from this observation, there is no clear pattern. Based on the average gain prescribed across input levels and hearing loss, the procedures rank in order from most to least prescribed gain as follows: DSL[i/o], 21.4 dB; NAL-NL1, 17.8 dB; Danalogic,16.5 dB; DigiFocus II, 16.0 dB; Claro,15.2 dB; and Senso Diva 14.1 dB.

Generally, the amount of overall gain prescribed by the six procedures at any one input level varies by about 10 dB, a substantial amount. The variation is, on average, slightly smaller for the gently sloping loss and slightly larger for the steeply sloping loss with a mild low-frequency loss. Interestingly, for the steeply sloping loss with a mild low-frequency loss, both the ASA2 and the SKI algorithm were accepted by the Otiset fitting software as appropriate. Whereas the SKI algorithm (see Figure 1) on average across input level prescribed 9.7 dB of overall gain for this hearing loss, ASA2 prescribed 18.2 dB of overall gain.

Back to Top | Article Outline

Differences in shape

The general difference in prescribed overall gain of 10 dB is large, but this difference may be compensated for by the use of a volume control or by adjusting the overall gain to the client's preferred listening level for an average input level (about 65 dB SPL). On the other hand, the prescribed slope typically cannot be adjusted by the aid user and is preferably not compromised by the audiologist.

Figures 2 to 6 show the differences in insertion gain data obtained for each of the five audiograms for a 65-dB input when the curves are normalized to the same gain level at 1000 Hz. We did the normalization so that we could compare differences in the shapes of the prescriptions independently from the overall gain that was prescribed. For simplicity, we are reporting this comparison for only the 65-dB input level, at which level the prescribed gain-frequency response shapes are those recommended for listening to average speech. Note that for this input level the Senso Diva aims at equalizing loudness of speech bands (cf. Table I).

Figure 2

Figure 2

Figure 3

Figure 3

Figure 4

Figure 4

Figure 5

Figure 5

Figure 6

Figure 6

Figures 2 and 3 show the difference in response shapes prescribed for the flat and reverse sloping loss, respectively. For these audiogram configurations, the insertion gain curves display differences in targets of 15–20 dB at 250 Hz and of 10–15 dB at 500 and 4000 Hz after normalization of gain at 1000 Hz. That is, the prescribed response shape varies substantially between procedures.

Below 1000 Hz where the hearing loss is severe in both cases, there is a distinct difference in the shapes of the targets prescribed by the two generic methods. While the DSL[i/o] target curve, aiming at loudness normalization, is relatively flat, the NAL-NL1 target curve, aiming at speech intelligibility maximization, rises steeply.

In this frequency region, the targets prescribed by Senso Diva and DigiFocus II are most similar in shape to the DSL[i/o] targets, even though none aims at pure loudness normalization. Danalogic, which does use a pure loudness-normalization rationale, prescribes targets most similar to the NAL-NL1 targets, whereas the Claro targets, also aiming at normalizing loudness, seem to produce a shape somewhere between NAL-NL1 and DSL[i/o].

Above 1000 Hz, the shapes of the two generic target curves are very similar. For the flat loss, Senso Diva and DigiFocus II follow the generic targets, whereas the Danalogic target curve is flat across the high frequencies and the Claro target curve rolls off steeply at 2000 Hz. For the reverse sloping loss, in comparison to the two generic methods, the proprietary methods all prescribe a shallower downward-sloping target curve across the frequencies above 1000 Hz. Across the entire frequency range, Senso Diva prescribes for both audiograms the flattest response of the six methods.

The shapes of the target curves for the gently sloping high-frequency loss are reported in Figure 4. For this audiogram configuration, a moderate 10-dB variation in gain is observed at both low and high frequencies after normalization of gain at 1000 Hz. In this case, both the generic methods prescribe a similar upward-sloping target curve from 500 to 2000 Hz. Outside this range, DSL[i/o] prescribes relatively more gain than NAL-NL1. Apart from Senso Diva, the proprietary methods prescribe, like NAL-NL1, an upward-sloping target curve from 250 to 2000 Hz. Above 2000 Hz, Danalogic and DigiFocus II more closely follow the DSL[i/o] targets, whereas Claro presents a roll-off like NAL-NL1, which is not typical of a loudness-normalization procedure. As for the flat and reverse sloping loss, Senso Diva prescribes a much flatter target curve than the other five methods.

After normalization of gain at 1000 Hz, spreads of more than 20 dB are seen across targets at the high frequencies for the two steeply sloping high-frequency losses (see Figures 5 and 6). For both of these, the generic targets differ above 1000 Hz. NAL-NL1 prescribes a slightly shallower slope than DSL[i/o] and its targets roll off at 3000 Hz while the DSL[i/o] targets continue to rise steeply.

Except that the Senso Diva target curve is the most shallow, there is no consistent pattern in the proprietary target shapes for the two steeply sloping losses above 1000 Hz. For example, Danalogic's targets are closest to those of DSL[i/o] for the steeply sloping loss with normal threshold up to 1000 Hz (Figure 5), but they are closest to NAL-NL1 for the steeply sloping loss with a mild low-frequency hearing loss (Figure 6).

For the Claro target, the pattern is reversed. Both of these proprietary methods aim at normalizing loudness. For the steeply sloping high-frequency loss with normal threshold up to 1000 Hz, all the generic and proprietary methods prescribe a similar flat target curve across the frequencies up to and including 1000 Hz. Figure 6, on the other hand, reveals variations of 10 dB at 250 Hz and of 15 dB at 500 Hz between targets at the frequencies below 1000 Hz for the steeply sloping high-frequency loss with a mild low-frequency hearing loss. Whereas the targets of NAL-NL1, Senso Diva, and Danalogic gently rise across the low frequencies, the DSL[i/o], Claro, and DigiFocus targets are flatter. Note that the two manufacturers (Claro and DigiFocus II) that have introduced a target curve specifically for ski-slope losses recommend very different targets across the high frequencies. In particular, Claro presents a characteristic roll-off at 3000 Hz.

Overall, for an input level of 65 dB SPL, the two generic methods, which represent different and well-defined rationales, prescribe responses of very different shapes. None of the proprietary methods consistently prescribes a response shape similar to that prescribed by either DSL[i/o] or NAL-NL1. However, for frequencies below 1000 Hz, the Danalogic targets, which are based on the rationale of loudness normalization, are often much more similar to the NAL-NL1 targets. On the other hand, DigiFocus II, which has a rationale similar to that of NAL-NL1, has targets that are generally more similar to the DSL[i/o] targets below 1000 Hz.

When normalizing the targets at 1000 Hz, the various methods generally produce variations in gain at low and high frequencies between 10 and 20 dB. With a few exceptions, the differences in shapes of target curves for input levels of 50 dB and 80 dB are the same as those reported for a 65-dB input level.

Back to Top | Article Outline

Hearing loss configuration dependency/independency

Our final analysis investigated hearing loss configuration dependency. Hearing loss configuration dependency refers to the way that gain changes as the slope of the audiogram changes. A prescription is independent of hearing loss configuration if the gain at any one frequency is independent of the threshold at any other frequency. Conversely, hearing loss configuration dependency means that there is a relationship between the gain prescribed at one particular frequency and the hearing threshold at more than one frequency. In such a case, the slope of the loss together with the degree of loss influences the prescribed insertion gain at any frequency.

To test the prescriptions under investigation here for hearing loss configuration dependency, we entered into each software module four new audiograms of differing slopes that all passed through 50 dB at 1000 Hz. We examined the gain prescribed by each procedure for each hearing loss at 1000 Hz. If the prescription is independent of hearing loss configuration, then the slope of the loss would be irrelevant and the same gain would be prescribed at 1000 Hz for all audiograms. On the other hand, we would expect the gain to vary with the slope if the prescription is hearing loss configuration dependent.

Table 4 shows the results of the analysis. The three loudness-normalization procedures and the loudness-mapping procedure prescribed exactly the same gain at 1000 Hz regardless of the slope of the audiogram, indicating that they are hearing loss configuration independent. On the other hand, NAL-NL1 and DigiFocus II produced different gains for different slopes, indicating that they are hearing loss configuration dependent. It should be added that the acoustic characteristics affect the prescribed gain for the DigiFocus II. Therefore, we cannot be entirely sure if the observed differences between gains for each loss were due to hearing loss configuration dependency or to change in vent size.

Table 4

Table 4

Back to Top | Article Outline

DISCUSSION AND SUMMARY

The differences in the targets prescribed by the two generic prescription procedures (DSL[i/o] and NAL-NL1) have been reported before1 and can be explained from the assumptions and principles underlying the two procedures. For example, DSL[i/o] consistently prescribes more gain than NAL-NL1 at frequencies where the hearing loss is severe. It does so because DSL[i/o] is aiming at restoring loudness and ensuring audibility. However, NAL-NL1 includes a desensitization factor because it is believed that hearing-impaired listeners have reduced ability to extract useful information from speech at frequencies where the hearing loss is severe.12

On the other hand, little is known about the proprietary fitting methods and we were interested to find how much the targets from such procedures differed from those prescribed by the two generic procedures and from each other. The variation of 10 dB in prescribed overall gain for various audiogram configurations and input levels is not so critical, because this difference is easily overcome by the use of a volume control or by adjustment of the overall gain.

However, the variation in the targets for a 65-dB input is more significant. When all the targets are normalized to the same gain level at 1000 Hz, we see large differences in prescribed response shapes that produce variations in gain at lower and higher frequencies of between 10 and 30 dB (Figures 2–6).

Another study that presented insertion gain curves derived from 2-cc coupler measurements of selected generic and proprietary fitting methods implemented in commercial devices reported similar gain variations of 10 to 20 dB.13 The same study concluded that the differences have a large effect on estimated loudness, but only a small effect on estimated speech intelligibility. However, it is unclear from that study how much of the discrepancy is due to differences in prescribed overall gain (which can be compensated for) and how much to differences in response shape.

In the comparison of target curves, those for Senso Diva differed from the others in having a flatter shape and generally less overall gain. While the targets for all the other procedures are derived for a pure-tone test signal, this choice was not specifically available in Compass. We think—but have not confirmed—that the targets for Senso Diva are more likely related to a broadband complex signal, which would explain the lower overall gain. Further, whereas we obtained gain values for a range of audiometric frequencies for most of the other procedures, gain values could be extracted at only four octave frequencies for Senso Diva. That is, we obtained less detail on the shape of the targets for this device.

It should also be noted that the insertion gain targets for DigiFocus II, Claro, and Senso Diva were affected by the acoustic parameters and device selected. As far as possible, we equalized these factors across procedures in our investigation. We are unsure if this dependency was intended by the manufacturers, but we cannot see why it should occur.

A study of the target curves suggests that the three procedures (one generic and two proprietary) that share the rationale of normalizing frequency-specific loudness have very little in common. This may be partly because the loudness data used for each procedure are measured with different loudness tests and/or because the procedures are based on different operational principles, which in the case of the proprietary methods are virtually unknown.

With respect to hearing loss configuration dependency, the loudness-normalization and loudness-mapping methods (DSL[i/o], Claro, Danalogic, and Senso Diva) all seem to prescribe gain independent of hearing loss configuration.

In contrast, the gain prescribed by NAL-NL1 at each frequency depends on the degree of loss at several frequencies. Our investigation suggests that the prescription for DigiFocus II is also hearing loss configuration dependent. However, this observation may be a result of selecting different acoustic parameters for the different audiograms. We note that data collected at NAL in the 1980s showed that the gain-frequency response preferred by hearing-impaired listeners varied in a hearing loss configuration-dependent way.14,15

Overall, we find the outcome of this investigation thought provoking. Our data suggest that if the targets prescribed by the various fitting algorithms examined here are reached, clients with similar hearing losses can easily walk away with extremely different amplification characteristics, depending on which device and/or fitting method is chosen. In practice, however, it is likely that, due to vented earmolds, feedback, and limitations in the electroacoustic characteristic, the achieved differences in the real ear are smaller than the target differences reported here. Even so, the real-ear differences could be significant.

Personally, we think that the current paucity of information available on the various proprietary fitting methods makes it impossible for the audiologist to make an informed choice of which device and procedure will best benefit the client.

Back to Top | Article Outline

REFERENCES

1. Byrne D, Dillon H, Katsch R, et al.: The NAL-NL1 procedure for fitting non-linear hearing aids: Characteristics and comparisons with other procedures. JAAA 2001;12(1):37–51.
2. Cornelisse L, Seewald, R, Jamieson D: The input/output formula: A theoretical approach to the fitting of personal amplification devices. J Acoust Soc Am 1995;97(3):1854–1864.
3. Ludvigsen C: Basic amplification rationale of a DSP hearing instrument. Hear Rev 1997;4(3):58–64,67,70.
4. Schum DJ: Adaptive speech alignment: A new fitting rationale made possible by DSP. Hear J 1996;49(5):25–30.
5. Phonak: Claro Loudness Perception Profile. Stafa, Switzerland.
6. GN ReSound: Fitting Danalogic: The Reasoning Behind the Technology.Tastrup, Denmark.
7. Ludvigsen C: Audiological background and design rationale of Senso Diva. Widexpress 2001;15.
8. Jenstad LM, Seewald RC, Cornelisse LE, Shantz J: Comparison of linear gain and wide dynamic range compression hearing aid circuits: Aided speech perception measures. Ear Hear 1999;20(2):117–126.
9. Jenstad LM, Pumford J, Seewald RC, Cornelisse LE: Comparison of linear gain and wide dynamic range compression hearing aid circuits II: Aided loudness measures. Ear Hear 2000;21(1):32–44.
10. Keidser G, Grant F: Comparing loudness normalization (IHAFF) with speech intelligibility maximization (NAL-NL1) when implemented in a two-channel device. Ear Hear 2001;22(6):501–515.
11. Dillon H: Hearing Aids. Sydney, Australia: Boomerang Press 2001: 99.
12. Ching TYC, Dillon H, Katsch R, Byrne D: Maximizing effective audibility in hearing aid fitting. Ear Hear 2001;22:212–224.
13. Smeds K, Leijon A: Threshold-based fitting methods for non-linear (WDRC) hearing instruments: Comparison of acoustic characteristics. Scand Audiol 2001;30:213–222.
14. Byrne D, Murray N: Predictability of the required frequency response characteristic of a hearing aid from the pure-tone audiogram. Ear Hear 1986;7(2):63–70.
15. Byrne D, Parkinson A, Newall P: Hearing aid gain and frequency response requirements for the severely/profoundly hearing impaired. Ear Hear 1990;11(1):40–49.
© 2003 Lippincott Williams & Wilkins, Inc.