Journal Logo

Article

Digital hearing aid design: Fact vs. fantasy

Trine, Timothy D.; Van Tasell, Dianne

Author Information
doi: 10.1097/01.HJ.0000292490.67222.a2
  • Free

It is often said and written these days that digital technology will make possible what was not possible in analog hearing aids. This is true. It is also true that we are—as we always will be—at a point where what we wish digital hearing aids could do far exceeds what they actually can do. This has created some confusion among hearing professionals and consumers alike, and has given rise to some very legitimate questions: What can digital aids really do? What can't they do? How do we know?

In this article, we take a hard look at what digital hearing aids can do, and what is, at the moment, only wishful thinking. First, some definitions:

What digital hearing aids can do:

This includes all hearing aid functionality that has been proven efficacious via research published in the scientific literature. A good example of proven efficacy is the wealth of evidence from many scientific articles that aided hearing for speech is better than unaided hearing for speech, provided that the hearing aid provides sufficient audibility over a wide frequency bandwidth.1,2

Wishful thinking:

In this category falls hearing aid functionality that may be advertised, but for which the weight of evidence for efficacy is at best insufficient and at worst non-existent. A good example of this is the frequent contention that a non-directional hearing aid can reduce the level of background noise without reducing the audibility or clarity of speech. This ability has for some time been the Holy Grail of hearing aid signal processing— highly desirable but, for well-understood scientific reasons, still unattainable.

Our aim is to provide readers with some basic scientific information that will help them to: (1) develop informed opinions about efficacy of hearing aid features, and (2) avoid creating unrealistic expectations in their customers or patients. We will examine the state of science in each of three areas of hearing aid signal processing: compression architecture, feedback cancellation, and noise management.

COMPRESSION ARCHITECTURE

Wishful thinking

Probably the most pervasive wishful thinking regarding hearing aid compression is that multichannel wide dynamic range compression (WDRC) can replace completely the function of the cochlear outer hair cells (OHCs).

Research published during the last two decades has shown the mammalian OHC system to be a necessary part of a biologically active process in the cochlea that amplifies, in a highly frequency-selective manner and with nearly instantaneous time constant, the mechanical stimulus to the inner (sensory) hair cells.3,4

At first consideration, this suggests that hearing aid signal processing should perhaps be designed to make up for the loss of OHCs by mimicking what the OHCs seem to do: provide very fast, very low-threshold compression in a large number of narrow frequency channels. Indeed, there are several hearing aids on the market that do fast compression in many channels, but they seem to work no better than hearing aids with slower compression in fewer channels. What, then, was wrong with this logic?

The answer lies in the mechanical tuning properties of the cochlea, and how they are different for low-level and high-level input signals. Figure 1 (after Johnstone et al.5) shows the mechanical displacement of the basilar membrane of a guinea pig at a single measurement location (corresponding to a best frequency of 18 KHz) for pure-tone inputs from low to high levels. For convenience, all responses have been normalized at 10 KHz.

Figure 1
Figure 1:
Amplitude of basilar membrane vibration as a function of stimulus frequency for input levels from 20 dB SPL to 80 dB SPL. From Johnstone et al.5

For the 20-dB input, the effects of the OHCs are seen in the very narrow tuning of this place on the basilar membrane: It responds only to frequencies close to 18 kHz. But, as the input level grows, the OHC system becomes less and less active. By the time the input reaches 80 dB, the OHC response is saturated, and the basilar membrane is tuned over a much broader region.

So there are two reasons that an impaired ear cannot have its narrow tuning restored by a hearing aid: (1) because the OHC system responsible for the tuning is compromised, and (2) because hearing aids must apply gain to input signals and therefore stimulate the cochlea at moderate and high levels where tuning is no longer narrow even in a normal ear. This means that even the narrowest band compression system will still produce broader-than-normal excitation in an impaired ear.

This state of affairs has led some researchers to propose signal processing schemes that actually sharpen the spectral peaks in the input signal to offset the wide spread of excitation in an ear without OHCs. Once again, the approach makes sense at first glance, but doesn't hold up on closer scrutiny.

This is best illustrated by looking at the output of a model of cochlear excitation that includes the role of the OHCs.7Figure 2 shows the response of the normal cochlea (black curves) and the response of a cochlea with 50% OHC loss (red curves) to the vowel /ae/ (bold curves). Also shown is the response of the normal and impaired cochlea to the same vowel that has been processed with a spectral sharpening algorithm (light curves). Notice that, although the spectral sharpening is preserved in the output of the normal cochlea, at the output of the cochlea with OHC damage the spectrum is smoothed out with essentially no difference between the unsharpened and sharpened signal. This occurs because even the sharpened input produces broader patterns of excitation in the cochlea with damaged OHCs. It is the reason that reported attempts to improve speech intelligibility via spectral sharpening have without exception failed.8,9

Figure 2
Figure 2:
Modeled excitation pattern for the synthetic vowel /ae/ (bold curves), and a “pre-sharpened” version of the same vowel (light curves) at the output of a normal cochlea (black curves) and a cochlea with 50% OHC loss (red curves). From Smoorenburg.6

The inescapable conclusion is that hearing aid signal processing cannot be considered a replacement for the complex function of the OHCs, no matter how attractively scientific that argument may at first seem.

What digital hearing aids can do

If WDRC is not designed to replace the OHCs, what is its purpose? As every hearing professional knows, the primary purpose of any hearing aid is to improve the audibility of speech and other signals of interest for persons with hearing impairment. This fundamental truth explains why WDRC is usually the compensation strategy of choice for people with mild-to-moderate sensorineural hearing loss caused by damage to the cochlear OHCs.10

Simply put, WDRC hearing aids provide more audibility than other compression strategies because they provide more gain for soft signals; that is, they do restore audibility for soft sounds, even though they cannot restore the exquisite frequency selectivity of the normal cochlea for those sounds.

Beyond that, the major aspects of WDRC design—number of compression channels and compression time constants—should be based on considerations of signal audibility and sound quality.

Number of compression channels

The optimal number of channels in a compression hearing aid is related not to cochlear physiology, but to the intelligibility and quality of sound, in addition to the ease with which a hearing aid can be fitted to individuals with different hearing loss configurations. For these purposes, what is the number of necessary channels?

Rickert and colleagues at Starkey Laboratories11 approached this problem by quantifying the number of independent signal processing channels required both to maximize the audibility of speech (as quantified by the Articulation Index, ANSI S3.512) and to match targets from the Cambridge non-linear fitting formula13 for a variety of representative audiograms.

Figure 3 shows a summary of our findings with respect to speech audibility. Notice that for each audiogram, performance improves dramatically from one to two channels, but has reached asymptotic levels by four or five channels. It is clear that three to four channels of compression will provide sufficient flexibility to match the vast majority of audiometric configurations encountered clinically, even without additional frequency shaping within the channels. Even fewer channels are required if some form of frequency shaping occurs independent of the channel gains.

Figure 3
Figure 3:
Audibility quantified by the Articulation Index as a function of the number of gain and compression channels. Each curve represents data from individual audiograms.

Compression time constants

We have already established that compression time constants need not be designed to mimic the function of the OHC system. But what should they be in order to maximize audibility and sound quality?

Fast and slow time constants have been shown to produce equally good speech audibility for a wide range of input levels.14 It is in the area of sound quality that time constants seem to be most important. Figure 4 (adapted from Woods et al.15) shows the results of a statistical analysis of the findings of a large-scale laboratory and field study conducted at Starkey to determine the preferences of both normally hearing and hearing-impaired listeners for various multichannel compression schemes.

Figure 4
Figure 4:
Correlation between gain variance and perceived sound quality. Top panel shows data from listeners with normal hearing. Bottom panel shows data from listeners with sensorineural hearing loss.

In this experiment, audibility of the speech signal in background noise was held constant across all processing conditions, and listeners were asked to rate the strength of their preference for the sound quality of a large number of different multichannel compression time constants and compression ratios. The x-axis of each panel shows the “preference distance” from the linear processing condition; the further to the right along this axis, the less the listeners liked the sound quality. On the y-axis is plotted the variance of the signal processing gain; this can be thought of as how often and how much the gain was changing. For fast time constants and high compression ratios, the gain variance would be greatest. For slow time constants and low compression ratios, the gain variance would be the least.

For both groups of listeners, preference was clearly related to the gain variance; the more the gain varied, the less the listeners liked the sound quality. This clearly indicates that slow WDRC time constants will give the best sound quality result across a range of compression ratios.

Reasonable expectations for compression architecture design

Starkey R&D groups designed our recently released digital hearing aid, Axent, to embody the known sound scientific evidence in the areas of psychoacoustics and signal processing. Where such evidence did not exist, we conducted the experiments reported above. Based on the current evidence, Starkey selected the following compression architecture characteristics for the purpose of optimizing speech audibility and sound quality:

  • WDRC compression
  • four adjustable channels of compression with capability for further adjustment of frequency response within channels
  • slow compression time constants
  • low-level expansion to minimize microphone and low-level ambient noise.

FEEDBACK CANCELLATION

Wishful thinking

Adaptive feedback cancellation is one of the most exciting and useful signal processing features available in some digital hearing aids. The wishful thinking about feedback cancellation generally centers on a belief that these DSP algorithms will eliminate all feedback, making obsolete the need for a well-fitted earmold or shell, or even eliminate the occlusion effect by allowing large vents in all fittings. Although we can look forward to steady improvements in adaptive feedback cancellation, we aren't there yet.

What digital hearing aids can do

Adaptive feedback cancellation refers to algorithms that estimate the feedback signal and subtract it from the hearing aid input signal without sacrificing any gain. These are sometimes referred to as “search and destroy” algorithms because they continuously adapt the characteristics of the digital feedback estimation filter to accommodate changes in the characteristics of the feedback. They differ significantly from so-called feedback “management” routines or notch filters that either sacrifice or limit available gain in the region of the feedback.

What are realistic expectations for adaptive feedback cancelers in today's digital hearing aids? Figure 5 shows a typical patient's maximum real-ear aided gain before feedback with and without the feedback canceler enabled in an Axent digital hearing aid. Notice that the maximum gain increased by about 10 dB to 15 dB. This increased gain margin is typical of what can be achieved with currently known methods for adaptive feedback cancellation.16,17 That is, feedback cancellation can provide an increase of 10 dB to 15 dB in available gain without feedback, but there will be a point (especially with high-gain aids and/or feedback-prone ears) where even a hearing aid with an adaptive feedback canceler will start to produce audible oscillatory feedback.

Figure 5
Figure 5:
Maximum stable real-ear aided gain from an individual subject with the Axent hearing aid with adaptive feedback cancellation enabled (red curve) and disabled (black curve).

What are the reasons for this limitation? They are not, as you might think, entirely attributable to the limited signal processing power available in today's digital hearing aids. Rather, a significant limitation seems to be in the nature of the feedback signal itself. Because acoustic room reflections almost always interact with and complicate the hearing aid feedback signal, the ability of even a complex adaptive feedback estimation filter to accurately model and subtract the feedback is limited.17 For this reason, significant increases in available gain margin beyond 10 dB to 15 dB in everyday use will have to await further advances in signal processing.

Reasonable expectations for treatment of feedback

Consumers should expect an adaptive feedback canceler in today's full-featured digital hearing aids, not just a feedback-management system. They should expect an additional gain margin of about 10 dB to 15 dB before feedback, and they should expect good sound quality while the canceler is at work. However, they should not expect that they will never again have to deal with feedback in any circumstance.

NOISE MANAGEMENT

Wishful thinking

Noise reduction or noise management is the single signal processing feature that creates the most confusion for hearing healthcare professionals and hearing aid users alike. The wishful thinking that has persisted throughout the years is that somehow an omnidirectional hearing aid can recognize noise and separate it from speech.

For the past several decades, achieving this capability has been one of the signal processing goals most sought after by researchers in many fields. As the hundreds of patents in this area attest, an astounding number of techniques has been developed and tested. But, so far, none of them has proven effective at selectively reducing the kind of environmental noise that hearing aid users would like to get rid of while preserving the loudness and clarity of speech.

There are two general single-microphone* signal-processing approaches that can be used to reduce noise.

If the characteristics of a noise signal are: (1) different from the speech, and (2) known by the signal processor, then the spectrum of the noise can simply be subtracted from the spectrum of the speech + noise signal, leaving a less noisy speech signal.18 This is generally how noise-reduction headsets for use on airplanes work: The airplane noise is steady, unchanging, very different from speech, and easy to characterize, so it is easy to suppress.

This technique doesn't work well with the kind of changeable background noise typical of the everyday environments of hearing-aid users.19 That is why a second approach to noise management has been used in hearing aids: In noisy environments the hearing aid gain is simply reduced when the estimated speech-to-noise ratio (SNR) gets too low. Because this technique turns the gain down for the speech and the noise, it cannot result in an improvement in SNR, although this is precisely where all the wishful thinking seems to have been applied.

The scientific literature is replete with examples of how noise management does not yield an improvement in speech recognition in noise.20–22 Until recently, the only publication reporting an improvement in aided speech recognition in noise with a noise-management algorithm was a paper published in 1984.23 The algorithm described in this paper eventually became available in wearable hearing aids as the “Zeta Noise Blocker.” The reported improvement in speech recognition, however, was attributable to experimental methodology: Listeners were allowed to adjust their volume controls in each experimental condition. Consequently, the listeners simply turned up their volume controls in the “noise management” condition and consequently improved the audibility of speech relative to the “no noise management” condition.

More recently, there has been another report of improved speech recognition in noise from an omni-directional hearing aid.24 This multi-site investigation showed a 1.5-dB S/N improvement on the Hearing in Noise Test25 between noise reduction “on” and “off” conditions in tests conducted at the manufacturer's facility, but showed no statistically significant improvement at the independent clinical site. Although the source of the discrepancy between the two sites is not clear, the results from the independent site are more consistent with the existing published literature on the topic.

What digital hearing aids can do

If we can abandon our wishful thinking about what single-microphone noise- management algorithms can do, we can understand better the considerable benefits that they do provide hearing aid users. Many of these algorithms employ sophisticated techniques to estimate speech-to-noise ratio in the user's sound environment and to reduce gain only in frequency regions where the noise is most intense. This reduction in the loudness of noisy signals can be perceived by the user as welcome relief; indeed, anyone who has listened to a hearing aid with one of these algorithms in place knows that they can easily give the illusion that speech is suddenly clearer with the noisy channels reduced.

Noise management is a feature that hearing aid users find useful. Figure 6 shows data from a field trial of noise management performed at Starkey with users of Axent digital hearing aids. At each of the three noise-management time constants tested, the majority of participants preferred some level of noise management over none.

Figure 6
Figure 6:
Proportion of Axent hearing aid users who preferred the minimum or maximum noise management (yellow) or no noise management (red) at each of the three time constants tested.

Reasonable expectations for single-microphone noise management

Users should not expect that their hearing aid noise management will make speech “pop out” from background noise, or that it will get rid of the background noise while preserving speech. Users should expect noise management in high-end digital hearing aids to estimate accurately the speech-to-noise ratio in the environment, reduce the gain unobtrusively in low SNR situations, and restore the gain when the SNR becomes more favorable.

Finally, as is clear from Figure 6, one size does not fit all when it comes to noise management or, probably, to other advanced digital hearing aid features. The hearing professional should expect the manufacturer to provide an easy way to fit noise management, which will mean helping the user to determine if this is a feature that he or she will find useful.

SUMMARY AND CONCLUSIONS

It is still wishful thinking to imagine that today's digital hearing aids can compensate completely for hearing damage, make feedback a thing of the past, or increase the clarity of speech in noise using a single omnidirectional microphone.

On the other hand, consider how much progress has been made in all three areas during the last decade alone:

  1. The benefits of WDRC in restoring audibility for low-level signals have been defined and understood.
  2. The first adaptive feedback cancelers have become available in digital hearing aids.
  3. Accurate, flexible noise-management algorithms are making hearing use more comfortable than ever before.

The next decade will bring further advances on these and other fronts. While industry R&D is working on those fronts, here is what hearing professionals should expect from manufacturers:

  • hearing aid design that is based on established scientific evidence of effectiveness
  • accurate and honest portrayal of the benefits of new technology
  • flexible fitting systems that will acknowledge potential differences among users in their desire for advanced features, and that will assist hearing professionals to achieve the best possible fit of advanced products to their customers.

REFERENCES

1. Larson VD, Williams DW, Henderson WG, et al.: Efficacy of 3 commonly used hearing aid circuits: A crossover trial. NIDCD/VA Hearing Aid Clinical Trial Group. JAMA 2000;284:1806–1813.
2. Humes LE, Christensen LA, Bess FH, Hedley-Williams A: A comparison of the benefit provided by well-fit linear hearing aids and instruments with automatic reductions of low-frequency gain. J Sp Lang Hear Res 1997;40:666–685.
3. Dallos P: The active cochlea. J Neurosci 1992;12:4575–4585.
4. Dallos P: Outer hair cells: The inside story. Ann Otol Rhinol Laryngol 1997; Suppl.168:16–22.
5. Johnstone BM, Patuzzi R, Yates GK: Basilar membrane measurements—and the traveling wave. Hear Res 1986;22:147–153 as cited in Pickles JO: An Introduction to the Physiology of Hearing, 2nd edition. New York: Academic Press, 1988: 42–43.
6. Smoorenburg G: Psychoacoustics of hearing loss. Paper presented at the European Hearing Instrument Manufacturers Association (EHIMA) meeting. Brussels, May 1999. Retrieved from
    7. Giguère C, Smoorenburg GF: Computational modeling of outer hair cell damage: Implications for hearing aid signal processing. In Dau T, Hohmann V, Kollmeier B, eds. Psychophysics, Physiology and Models of Hearing. River Edge, NJ: World Scientific, 1999: 155–164.
    8. Stone MA, Moore BCJ: Spectral feature enhancement for people with sensorineural hearing impairment: Effects of speech intelligibility and quality. J Rehab Res Dev 1992;29:39–56.
    9. Musch H, Pavlovic C: The effect of spectral contrast enhancement on speech perception in hearing-impaired listeners. J Acoust Soc Amer 2001;109:2440(A).
    10. Moore BCJ, Glasberg BA: A comparison of four methods of implementing automatic gain control (AGC) in hearing aids. Brit J Audiol 1988;22:93–104.
    11. Rickert M, Van Tasell DJ, Woods WS: Compression bands: How many are needed? Presentation at the American Auditory Society, April 2000, Phoenix.
    12. ANSI S3.5:Methods for the Calculation of the Articulation Index. New York: American National Standards Institute, 1969.
    13. Moore BCJ, Alcantara JI, Stone MA, Glasberg BR: Use of a loudness model for hearing aid fitting: II. Hearing aids with multi-channel compression. Brit J Audiol 1999;33:157–70.
    14. Stone MA, Moore BCJ, Alcantara JI, Glasberg BR: Comparison of different forms of compression using wearable digital hearing aids. J Acoust Soc Amer 1999;106:3603–3619.
    15. Woods WS, Frush RA, Van Tasell DJ: “Preference space” for parameters of two-channel speech compression. Proceedings of the 1999 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, New Paltz, NY, October 1999.
    16. Maxwell JA, Zurek PM: Reducing acoustic feedback in hearing aids. IEEE Trans Sp Audio Processing 1995;3:304–313.
    17. Kates JM: Room reverberation effects in hearing aid feedback cancellation. J Acoust Soc Amer 2001;109:367–378.
    18. Boll SF: Suppression of acoustic noise in speech using spectral subtraction. IEEE Trans Acoust Sp Signal Processing 1979;27:113–120.
    19. Elberling C, Ludvigsen C, Keidser G: The design and testing of a noise reduction algorithm based on spectral subtraction. Scand Audiol 1993;Suppl 38:39–49.
    20. Van Tasell DJ, Larsen SY, Fabry, DA: Effects of an adaptive filter hearing aid on speech recognition in noise by hearing-impaired subjects. Ear Hear 1998;9:15–21.
    21. Fabry DA, Van Tasell DJ: Evaluation of an articulation-index based model for predicting the effects of adaptive frequency response hearing aids. J Sp Hear Res 1990;33:676–689.
    22. Walden BE, Surr RK, Cord MT, et al.: Comparison of benefits provided by different hearing aid technologies. JAAA 2000;11:540–560.
    23. Stein LK, Dempesy-Hart D: Listener-assessed intelligibility of a hearing aid self-adaptive noise filter. Ear Hear 1984 5:199–204.
    24. Bray V, Valente M: Can omni-directional hearing aids improve speech understanding in noise? Audiology Online, September 24, 2001. Retrieved from
    25. Nilsson M, Soli SD, Sullivan JA: Development of the Hearing in Noise Test for the measurement of speech reception thresholds in quiet and in noise. J Acoust Soc Amer 1994;95(2):1085–1099.
    Copyright © 2002 Wolters Kluwer Health, Inc. All rights reserved.