Hearing aid makers tackle the technology’s ‘number one problem’ with AI – STAT

For many people with hearing loss, a normal conversation at a busy restaurant is the holy grail.

[Background noise] is the number one problem that needs to be solved, said Abram Bailey, a former practicing audiologist who now runs Hearing Tracker, a popular online resource. Weve been trying to solve it in this industry for years and years and years. But we have only made these really tiny, incremental steps.

Since digital hearing aids became available a quarter century ago, their audio quality in complex situations like a cocktail party has inched forward thanks to developments like directional microphones and pre-programmed settings for common sound environments. But dueling new entries to the market one from established Danish brand Oticon and another from California startup Whisper are trying to take a bigger leap forward using artificial intelligence to make noisy environments clearer than ever.

advertisement

Though other companies employ AI in parts of their hearing aids or apps, Whisper and Oticons new products use deep neural networks trained on giant data sets to isolate and sharpen the right sounds for people with hearing loss. By all accounts, the technology has promise for hearing aids. Deep learning already powers technologies we use every day and in controlled lab settings, it can perform impressive feats of sound denoising. But progress may move more slowly in the world of hearing tech.

Everybody would love to have a breakthrough technology, but even breakthrough technologies tend to be based on incremental changes, said Todd A. Ricketts, a professor and vice chair at Vanderbilts department of hearing and speech studies. He added: What we might see from the first generation of deep neural networks is maybe small incremental improvements.

advertisement

Dont tell that to Dwight Crow, the CEO and co-founder of Whisper, which launched its hearing system last fall. The three-year-old company has its eyes on the kinds of quantum leaps that Ricketts says are incredibly rare.

Whispers approach is intriguing both because of its unconventional design and its Silicon Valley roots. Its founders worked at Facebook, its executive ranks are peppered with recruits from Google, and the company is fueled by funding from Sequoia and other well-known venture capital firms.

Where traditional hearing aids are constrained by the batteries and processors that can be stuffed in a tiny capsule tucked behind someones ear, Whispers hearing system uses a small external processing box the company calls the brain. Microphones on the hearing aids send audio wirelessly down to the brain hundreds of times a second, where its processed and sent back at a similar latency as other products.

According to Crow, skirting the size limitations of a traditional hearing aid allows the company to run algorithms that require far more computational power to complete in real time. The result, the company claims, is the ability to better isolate desirable sounds, like speech, and to elevate them more precisely than what was possible without deep learning.

So far, Whisper can only describe the benefits in terms of computing potential, claiming the Whisper brain can perform 300 billion operations per second, compared to around 1 billion in competitors.

While Whispers hearing aids work without the brain, one crucial question is whether people will invest top-dollar for a system that needs a box to work optimally.

People with hearing aids have a history of adopting accessories in order to hear better, said Crow, pointing to Bluetooth technology. Before it was built directly into hearing aids, many people wore an electronic necklace to make wireless connection with phones possible.

I would see at least a decade where having an auxiliary device is going to transform your ability to hear clearly, he said.

In another potential advantage, the box also makes it possible for the company to issue software updates. Whisper announced its first such update Tuesday, which the company claims improves the brains sound separation engine. For now the upgrade must be performed by an audiologist, but Crow said the company hopes future updates can be done by the user.

Shortly after Whisper started rolling out its system last fall, the Oticon More launched using a similar deep learning algorithm trained on 12 million sounds, the technology crammed into the hearing aids themselves.

The goal behind the [deep neural network] here is to take the sounds in and essentially make them balanced and accurate, so that they actually sound the way that theyre intended to sound within that environment, said Annette Mazevski, a technology assessment manager at Oticon.

In a clinical white paper, the company describes a research study with 18 subjects with hearing loss and an average age 68.5: The More improved speech understanding by 15% compared to the preceding model, the Opn S, when using the most common default fittings for each.

Tracy Hagan Winn, a clinical audiologist at Northwesterns Center for Audiology, Speech, Language, and Learning, started practicing around the time Oticons first digital hearing aid was released 25 years ago. She said popular features in recent years include Bluetooth and rechargeable batteries.

In the context of general consumer electronics, Bluetooth and rechargeable batteries might seem humdrum additions but theyre relatively novel developments in the slow-moving world of audiology where miniaturization is half of the art.

Winn said though hearing aid sound quality has come an incredibly long way and that the skys the limit with technology, she cautioned that the latest developments may have a more limited impact than marketing would suggest.

Each individuals hearing loss can be very different, she said. And so just because its a great new technology doesnt mean its going to work for everybody and that we do have to set realistic expectationsSometimes the marketing makes it sound like this brand new product is going to be perfect for everyone. And everyones going to hear like they were 18 years old again. And we just know that thats not the reality.

And that hints at a challenge for Whisper. Audiologists are a crucial piece of any high-end hearing aid system because they perform tuning, maintenance, and sales. Whisper has opted to lease its system through a monthly subscription. For $179 over three years, a customer gets the hardware plus all the regular service. The overall price of about $6,500 is similar to what a person might pay for a competitive product from an audiologist. Despite some loss protections and introductory pricing, the overall math may not add up for all users.

Moreover, though the concept of Whispers system might be alluring, its still only available through some 20 audiologist offices across the country, and the company will need to battle the inertia of established players like Oticon that are widely available. Audiologists and consumers alike grow to trust certain brands.

Winn said that given hearing aids are long-term investments people should plan to keep for five to seven years, going with an established company guards against the possibility of a company going out of business relatively soon after a consumer bought their device.

I would be reluctant to try a new manufacturer until I saw some pretty good stability, she said.

Crow acknowledges the risk and insisted that Whisper is here to stay as a force to be reckoned with.

In the end, what may sway consumers to the startup is Whispers promise of a learning hearing aid system that gets better and better over time, especially if it cracks the problem of trying to filter out background noise with AI.

Whether its possible, whether its actually happening right now with some of these products on the market, I dont know, said Bailey from Hearing Tracker. But if it does prove possible, he said, it would be an absolute game changer.

Read the original:

Hearing aid makers tackle the technology's 'number one problem' with AI - STAT