Inside Oticon Reveal’s Dual AI: A Q&A with Thomas Behrens

When Oticon Reveal was introduced in August, a lot of attention centered on its “Dual AI” architecture: one AI system focused on speech and another designed to preserve meaningful surrounding sounds. But there’s a lot more to it than that: Oticon is arguing that better hearing in difficult environments should not be defined by simply removing as much background sound as possible. 

Reveal builds on Oticon Intent and retains its 4D Sensor technology, while adding new AI processing, upgraded feedback management, improved connectivity, and Auracast support. The goal, according to Oticon, is to make speech clear while preserving enough of the surrounding soundscape to help the brain understand context and maintain a more natural sense of place. Although HearingTracker detailed the new device in its product overview and our partner lab, HearAdvisor, is still testing it, we wanted more perspectives on Reveal's underlying technology. 

To do this, we turned to Thomas Behrens, MS, Oticon’s VP of Audiology and Research, who has been a longtime research leader at the company’s Centre for Applied Audiology Research in Smørum, Denmark. We spoke with him about what changed with Reveal, what Oticon means by a “60% stronger brain response” than Intent, and why he believes the industry may need to rethink its long-running focus on speech isolation. His comments below have been edited for clarity and brevity.

HearingTracker: If you had to identify the biggest change in Reveal for consumers, what would it be?

Thomas Behrens: I think the biggest change is really these new BrainHearing insights that have allowed us to take R&D in a new and sharper direction.

We’ve been chasing better speech understanding and SNRs [signal-to-noise ratios] for many years, and for good reason. It was—and to some extent still is—the number-one problem people face when they get hearing aids. We’re now actually seeing more people who are fitted by professionals who use best practices getting close to their normal-hearing peers in terms of some speech understanding in noise measures.

But at the same time people need more help and more clarity, so speech remains important. But the brain needs access to both speech and context in an optimal balance. That is where things get interesting. 

HearingTracker: Why is the surrounding context so important? Is it because hearing aids have become really good at isolating speech?

Behrens: Exactly—and that can create a problem if you take it too far.

What the brain is naturally doing is putting the conversation you’re part of at the highest level of attention. Then you have other sounds around you and sounds farther away. You might be in a bowling alley and hear people talking nearby, bowling pins, shoes, coffee brewing, popcorn being made, people cheering—all of these sounds are around us.

When we’re engaged in a conversation, we mostly notice what we’re part of. But if somebody suddenly starts talking about popcorn, that sound can become available to us and we can focus on it.

That’s really the new goal: We need to give the brain access to everything. If we only give people speech, they may not feel as immersed or embodied in the listening situation. The surroundings help the brain interpret what is happening. 

HearingTracker: So how does Reveal’s Dual AI create that balance?

Behrens: The Speech AI ensures that speech stands out and is crisp, clear, and available. The Context AI then sets the contrast between speech and the listening environment, so speech stands out, but the surrounding sounds are still available.

With Oticon Reveal, we have two AI systems integrated on the same chip. They talk together and share information all the time, so they know what each other is doing. That allows them to be much faster and more precise than what we’ve been able to do in the past.

For example, the new Speech AI is three times faster at detecting speech than our previous detector. It has been trained to recognize the structure of speech very rapidly—even when speech is embedded in noise. If there is no nearby speech, it doesn’t try to pull something forward just because the background has speech-like modulation. It will then  keep a more general focus on the surroundings. 

HearingTracker: Oticon says Reveal produces a 60% stronger brain response than Intent. What exactly does that mean?

Behrens: We’ve developed new methods using cortical auditory evoked potentials. These recordings have traditionally been used, for example, in pediatric fittings to verify that meaningful sound is reaching the brain when a child cannot provide a behavioral response.

We’re now using a similar method to look at how clearly speech is represented when it reaches the brain. You can think of the cortical response as an index of how sound is encoded and integrated from the cochlea all the way to the auditory cortex.

In quiet, you get a clear waveform showing that the brain has access to the speech signal. In noise, without effective help from the hearing aid, that response can become much weaker. When we turn on Reveal’s AI Clarifier, we see that response becomes much stronger again. And when we compare Intent and Reveal, we see a 60% stronger response with Reveal.

To me, that is thought-provoking because it helps explain why noisy environments can be so demanding: the brain may simply not be getting a readily available signal, but importantly, a well-fit hearing aid with effective technologies can make speech readily available to the brain. 

HearingTracker: That brings up a good point: How much control does the hearing care professional still have over all this AI?

Behrens: We’ve actually added more control with Reveal compared with Intent.

We still think it’s very important that hearing care professionals can shape the technology to the needs of the user. At the same time, we continue researching better defaults and diagnostics so the clinician has a stronger starting point.

Our Audible Contrast Threshold (ACT) test is one example; it’s a quick, language-independent test designed to measure a person’s ability to hear in noise, rather than simply their pure-tone hearing thresholds. You don’t have to use ACT, but it can provide a better starting point. We know ACT will update the prescribed settings for about 50% of the people who are tested—some will need less support in noise and others more.

The technology can be highly automatic, but that doesn’t mean the professional’s role becomes less important. The more relevant information you bring into the fitting, the better the starting point can be. 

HearingTracker: Reveal also introduces a new feedback system and Wind & Handling Stabilizer 2.0. Are these innovations more significant than they sound?

Behrens: Yes. With the wind system, there is noise—and then there is “noise noise,” if you like.

We may want to know that the wind is blowing. What we don’t want is the artifact of wind hitting the microphones, or hair rubbing across them. That kind of non-informative noise should be removed because it’s just annoying and distracting. The new system can reduce those artifacts by up to 40 dB, compared with 30 dB before.

The feedback system is also more than just stopping whistling. We’ve increased the precision from 16 to 64 bands and made it about 30% faster. That can help preserve stable amplification while giving the AI systems more room to work. 

HearingTracker: What other changes should users notice?

Behrens: We’ve added a new connectivity platform that improves the stability and range of the LE Audio connection. We’ve also added Streaming Comfort, which adapts the level of the streamed sound according to how noisy the surroundings are.

Reveal is Auracast-ready from the start, and we’ve kept the system energy efficient enough that the Dual AI can run all day. We also continue to use the 4D Sensors from Intent. The new Speech AI complements those sensors, and that sensing information helps inform the algorithms about what the user is doing and what is happening around them. 

HearingTracker: Stripping away all the technical language, what real-world situation best demonstrates what Reveal is trying to do better than Intent?

Behrens: I think Reveal is really about those difficult environments where people can have conversations while staying aware and “in place” with their acoustic surroundings.

Zeal just came out, so there are probably other things we want to do first. But, yes, you can be pretty sure that there will be an updated Zeal at some point.

HearingTracker: And what about expanding the Zeal concept itself? One limitation of the current design is that it uses single microphones.

Behrens: It was a big effort to develop Zeal, and we don’t have a two-microphone version today—although that’s something we’re working on. It would obviously require a new design to make what you might call a “big brother” to Zeal. But it is definitely another thing we’re looking at—again, unfortunately something I cannot comment on in any detail.