Special Amplification for OTC Hearing Aids?

Note: This article is a continuation of the article How to develop a good hearing aid. In this series, I compile my day-to-day observations about the future of the hearing-aid market, practical questions about device design, mostly related to the self-fitting OTC devices.

Is OTC Hearing aid the same old Hearing aid, only cheaper?

The proposed FDA Rule for OTC hearing aids specifies, among other things, their required electro-acoustical performance. The document implies in several paragraphs that the safety and effectiveness of these devices can be assured by complying with the provided values.?

Remarkably, the OTC rule doesn’t introduce a strict separation towards the traditional category of Air-Conduction Hearing Aids, except for excluding the highest output levels. (Allowing the outputs of up to 120 dB SPL, relating them curiously to the fortissimo parts during a symphonic performance - that might be, statistically, one of the rarest acoustical situations in the Universe.) This lack of differentiation between the categories could lead to understanding that all hearing aids are substantially equivalent, providing more or less the same kind of hearing-loss compensation.

But this is not true. The type of amplification in modern hearing aids can differ significantly, with many qualities that cannot be revealed using standardized measurements. Moreover, OTC hearing aids will inevitably address new market segments, with future users that are likely to expect special benefits. These new users will be more sensitive to the variations between the products – and will call for new perception-enhancing features and new signal processing schemes.

Here I will try to outline the principles behind signal processing for hearing aids and present some of the most promising schemes for the OTC market.?

Traditional hearing aid amplification

It is no surprise that the FDA’s OTC specification primarily focuses on level-arithmetic, where SPL-magnitude of amplification is directly linked to the effectiveness of the device: If you control the SPL values the rest will follow.?This approach was very successful for many decades.

In the hearing aid industry, the definition of hearing loss as elevation of hearing thresholds is still dominant. The diagnostics and most of the solutions for hearing difficulties are modeled around the mapping of dynamic ranges: the “healthy” whole of the acoustical environment is mapped into the “impaired” residual dynamic range. The mapping used a simple linear projection.

Most commercial devices utilized multi-channel WDRC (Wide dynamic range compression). WDRC applies input-controlled gain to the sound captured at the microphone, acting as a funnel that keeps the acoustical stream from spilling outside of the target sink. As I will later try to explain, a funnel that is a sensible solution for fluids might be less practical for solids.

Problems with Hearing Problems?

Hearing loss is a complex condition that differs from person to person – even from ear to ear.?

Our understanding of healthy hearing is still not complete. Human hearing involves several counter-intuitive processes. For example, in the inner ear, the interplay of inner and outer hair cells introduces dynamic compression. This internal compression depends on the level and type of the sound and helps us to intentionally focus on a particular sound. This is very practical for listening, but it complicates our model of hearing. When the acoustical excitation is coded into neural impulse it has already been exposed to several mysterious modifications.?

The shaping and re-shaping of the sound representation in the inner ear demonstrate challenges faced by developers of signal processing: For restoring hearing, it might not be enough to restore the acoustical part of the signal path.??

Take for example the so-called Hidden Hearing Loss (HHL) – a condition where the impairment in perception cannot be attributed to a measurable hearing loss (as diagnosed by a pure-tone hearing test). We might as well joke, that a person with HLL can indeed?hear?well, she only has certain problems with?listening. What are the consequences for the design of hearing aids?

Having a healthy audiogram, a person with HLL doesn’t need a restoration of dynamic range or level sensitivity, she needs a restoration of auditory perception.

The design of an audio prosthesis should not focus only on the restoration of ‘healthy SPL sensitivity’, but respect the totality of the auditory perception, including cognitive processes.

Speaking of Speech

With limited information about individual hearing loss, the design of a hearing aid can include the understanding of the most important of the audio signals – speech.

Speech is indeed considered in the design of hearing aids. Time constants for WDRC and noise reduction algorithms follow the typical energy envelope of speech. The term “syllabic WDRC” proves this well. Another example: The negative effects of WDRC on vital speech patterns are known since the 1980s. They are the source of different trade-offs and disputes in the scientific community: more channels vs. fewer channels, long compression times vs. short compression times, etc.?

Coming back to the notion of WDRC as a funnel that squeezes the dynamics into the narrow target, you can intuitively grasp the negative effects on fine energy fluctuations in speech. Speech is a dance, a sequence of tiny steps and intense jumps, the message coded into the difference between the highs and lows. Those can be lost if squeezed through a rigid funnel.

Most of the attempts to bring speech and hearing aids together concentrate on the fidelity of amplification – on retaining the vital features of speech during the compression. But speech features can be further augmented to improve peception. Here are some examples from the field:

• Algorithms that try to linearize amplification around the most likely levels of speech: A combination of a slow and fast compressor, floating linear gain, histogram-controlled linearized amplification, compression curves with multiple compression knees.??
• Algorithms that attempt at spectral augmentation of speech features: formant booster, spectral contrast enhancement, consonant tilt, etc.
• Algorithms that modify the duration of phonetic elements of speech – most often extending the duration of consonants.??

In the hearing aid market, you can find marketing claims of improving ease of language understanding, of diminishing cognitive load. Those benefits are not expressed in terms of acoustical properties but quantified using cognitive factors. I believe that this is a promising development.

The motivation for purchasing an OTC product arises directly from a cognitive difficulty. The valuation of the benefit should ideally avoid any intermediate criteria, such as appropriateness of sound levels, and –if possible– stick to the cognitive realm.??

Time-frequency duality

Signal processing for hearing was long constrained by available processing power, but there are pure algorithmic barriers unimpressed by the progress in semiconductors. Users of OTC devices will be particularly sensitive to the latency of signal processing, so there is a good reason to keep the latency well below the noticeable values of about 10ms, even better in the lowest single-digits.

The time footprint and frequency resolution are interdependent parameters of audio signal processing. It is important to understand that denominations “zero-latency” and “channel-free” WDRC are marketing labels that do not directly relate to physical values. Note also that signal processing with a non-linear phase will provide for a favorable latency in a limited spectral area, but allow for much longer delays outside of that sweet spot. In the field of audio processing, it applies too: There is no such thing as a free lunch.?

Here are some algorithmic approaches to domesticate the time-frequency duality:

• Separate analysis and amplification routines, whereas the analysis routine allows for higher latency and the low-latency amplification sacrifices resolution and/or linearity of the phase
• Combination of frequency-domain and time-domain processing
• Partial signal-prediction during re-synthesis

Amplification in OTC hearing aids could in general have lesser constraints for the frequency resolution and open so some headroom for improving the latency.

The concern about latency becomes more pronounced in sophisticated de-noising algorithms.

Better speech perception in noisy environments?

For me, despite my healthy audiogram, using a hearing aid is fully justified in cocktail-party situations (acoustical environment with multiple competing speakers). Directivity features can improve my speech perception significantly and make my dinner more pleasurable, but in very difficult environments additional de-noising or source separation could bring me even larger benefits. When such a device arrives, I will be a typical OTC customer.

Advanced de-noising and source-separation algorithms of today are relying on artificial intelligence. Such processing comes with a computational footprint and a latency that are both one order of magnitude above the specs of today’s hearing aids. More computation power will soon be available, but the latency will still require a trade-off, limiting the use of such algorithms to most demanding situations.

Assuming that OTC devices will often be purchased for situational use, maybe in a cocktail-party environment, advanced de-noising techniques could soon become a common OTC specification.

Individual signal processing in each OTC device?

It is safe to assume that future OTC buyers will bring a broad variety of problems with auditory perception: some of those related to the elevation of hearing threshold, some of them related to diminishing cognitive ability.?

Today, the hearing aid is fitted by altering the parameters of a fixed signal processing scheme. In the future, diverse causes of hearing difficulty might be served by variable signal processing schemes.?

The notion of an OTC hearing aid as an individual solution for cognitive difficulties, rather than a universal acoustical amplifier, can bring us to the idea that each OTC customer could individualize the signal processing scheme.?

Artificial intelligence can provide means: by collecting the cognitive preferences of the user, the signal processing itself should be personalized. If not invented by AI, then combined of known blocks, including variations of nonlinearities that might be counter-intuitive according to the traditional practices of electro-acoustical processing, but effective for a particular user.?

This brings us to the matter of self-fitting...

To be continued

This series of articles is likely to be continued. The times are interesting for OTC hearing aids and deserve every commentary.