Starkey Omega AI+, how does quad DNN computing help you hear better?
Updated: 1 hour ago
Starkey Omega AI+ offers a 2.1 dB SNR improvement, which means the same listening effort when the noise level is 56% higher! What does that mean? How does the G4.0 processor in Starkey Omega AI+ with 4 DNN computing centres help you hear better in noise?
Starkey Omega AI+ is similar to Starkey Omega AI in terms of look, structure, battery life and connectivity, but not from a processing point of view. Instead of a full article on this new product, in this article, I discuss the improvements in the sound processing and why Starkey believes that the AI+ model is superior to the one before. For a full review of the Starkey Omega AI, please see this article.
How does Starkey Omega AI+ make it unique?
Starkey claims Omega AI+ takes hearing aids another step closer to natural hearing. Omega AI+ mimics the sound processing the way it happens in your brain, perhaps closer than any hearing aid has done it before. The result is a 2.1 incremental improvement in signal-to-noise ratio for the wearer.
How is Starkey Omega AI+ different from Starkey Omega AI hearing aids?
The approach that Starkey is taking in sound processing is more in line with Signia's Pure C&G MaX, where multiple DNN processes are hired to model the way your brain processes the sound. The Gen 4.0 processing architecture entails 4 DNN processors that mimic the sound processing in the human auditory pathway. The Omega AI is equipped with G 3.0 generation that does only single DNN processing. These 4 DNN processors in the AI+ model process:
DNN for enhancing speech in noise - The first one mimics Wernicke's region, which enhances speech and suppresses noise in a loud noise environment.
DNN to enhance directionality
DNN to enhance spatial nd sound cues
DNN to optimise the dynamic range of the speech versus background sound
This is how the Starkey Omega AI+ achieves an incremental 2.1 in signal-to-noise ratio for the wearer, which means the same level of effort in hearing speech if the noise is 56% louder.
I would encourage you to watch the video that I have shared in this article, where Starkey's CTO explains how the 4 DNN processors work in Starkey Omega AI+. For a full review of the Starkey Omega AI, please see this article.
What are other unique features of Starkey Omega AI+ that no other hearing aid has?
Besides Starkey's unique approach in Starkey Omega AI+ in mimicking the human brain in that sense, there are several health monitoring and reporting features that you can only find in Starkey Omega AI+ hearing aids. Some of those features are:
Fall detector - You can set up the myStarkey app to send SMS notifications to 3 of your contacts if the wearer falls down. This comes in handy, for example, if the wearer is frail or has a medical condition with a risk of falling down. This feature alone could be a lifesaver for many elderly and senior users.
Reminders - You can activate your AI assistant on Starkey Omega AI+ and ask for a reminder set up to remind you about taking your medication at such and such time every day. Or to remind you to pick up your kid from school at 5:00 pm every weekday.
Sleep pattern - IN some rare cases the patient is advised to avoid sleeping on sides on on the back. Starkey Omega AI+ can provide a sleep pattern report through myStarkey app.
Hearing aid left behind - The wearer of the Starkey Omega AI+ will receive a notification when one of the hearing aids is left behind and outside the Bluetooth range (about 10-15 meters).
How good is the Starkey Omega AI+ at hearing speech in noise?
There is no independent lab test done on the Starkey Omega AI+ yet. The only reference available is the lab test done by HearAdvisor on Starkey Omega AI (and not the AI+). The Starkey Omega AI received 2.9 out of 5.0 for speech-in-noise score.
Please see a comparison of the best AI hearing aids in 2026 in this article.
Independent lab test of Starkey Omega AI by HearAdvisor (source)
Tell me more about the sound processing in Starkey Omega AI+
This is a very interesting interview with the chief technology officer of Starkey, so I thought the transcript of the interview might be of interest for those of you more into the advanced technology in AI hearing aids.
Achin, I'm so delighted to meet you. Thank you so much for your time. I know you're an incredibly busy man to meet you. Here we are at EUHA 2026. And I have been inundated with new tech. Um, including from Starkey. So tell me why I should be excited about the Omega AI+. What's, what's the deal with it?
Yeah. So, I am really excited about Omega AI+. Let me start with the core of the product.
To start, I start with the technology, but then I first layer on neuroscience and get into the technology. It's got a brain-inspired architecture that includes an ensemble of deep neural networks. Okay, tell me more. So how does the human brain help us here?
Of course, the ear plays an important role. It transduces sound into neural pulses, and then we say it goes into the auditory cortex. But exactly what's happening? Luckily, there's a lot that is known in neuroscience and neurobiology about exactly what processing a healthy human hearing system performs, and then the goal for us is to copy those functions in hearing devices to help people hear a forecast. Let me now kind of get into the details, and not the progression of sound processing, but I'm going to discuss the hierarchy of importance of the brain's sound processing, and then I'll translate that into technology, and that's in Omega class. Once you are trying to listen to somebody, of course, one of the most important things to do is to understand conversations in a noisy environment.
As you know, that's the biggest complaint people with hearing impairment have.
Everyone coming through my door. It's the number one biggest challenge that people experience.
Yeah. Have you, have you battled that hearing aids work well in quiet? But I really am challenged in a noisy environment. Well, how does healthy human brain with healthy hearing, what processing does it do? What does it do?
It turns out, of course, we know now that there's a dedicated neural network. There's a dedicated processing centre in the brain right next to the auditory cortex in the posterior temporal gyrus. We call it Wernicke's region, named after the scientist who discovered it. It's got a singular function of enhancing speech and suppressing noise when that's what I'm trying to do, right? So it just makes sense that we will have a dedicated neural network that will do that in the chip of the hearing aid because a neural network is the best way to perform that function. Why do I know that? Because biology has evolved after millions of years of experiments, and it has come up with that computational architecture. Well, of course, this is now what every hearing aid is trying to do it. Have a neural network that's going to do speech enhancement and noise reduction.
What else does the human brain do for hearing? It turns out, even if I'm trying to listen to you in a noisy environment, the sense of directionality is extremely important. I really want to know if the sound is coming from here or over there. This ability to preserve the sense of direction for sensory signals is extremely important. So I must not compromise the directionality to enhance speech. Otherwise, my perception system is not going to be natural, normal or comfortable, which I imagine would be very easy to do if you were a tech that you could manipulate it so much that it takes away some of those spatial cues, important functions. Yes. So we have now a dedicated neural network that drives the directional system in a way that the brain. So where is the brain's directional system? Well, it turns out it's in the superior olivary complex. That's where you have the auditory directions processed and analysed. Guess what? With a neural network. So, our experiment showed that, rather than the classic signal statistics-based approach for controlling the microphone's directional system, a neural network approach is predictive, fast and responsive.
All right. So, you got the directional system with the second deep neural network. What's the third most important thing in the hierarchy of auditory processing? It's spatial awareness. The idea is, yep, I'm right now intent on listening to you. I got my sense of direction figured out, but I must be aware of everything that's going on around me, instantly, without losing the signal. So I must not suppress spatial cues of sound in trying to do job number one: speech enhancement, noise deduction. Job number two: directional. We have a third dedicated neural network that mimics the functions of the dorsal stream in the parietal lobe, right above the auditory cortex, where you have the spatial awareness preservation neural network processing. So it makes sense then to have a dedicated neural network to do that in the hearing. So even if I'm focused on you, if there's an important sound somewhere else, it will not be suppressed, but it's presented to me without attenuation. That's the third neural network. Well, we got a quad DNN architecture. So you got number four. Yeah. And the fourth one is what I'm most proud of because I believe it's innovative. It's going to set a new direction
in our industry. We've been working on it for a while. And so, to talk about this,
let me first bring out the brain again. So what else is the brain doing? Well, the brain's doing an important loudness mapping function. So, to explain to your audience.
Tell me more. Tell me more.
So the natural world has a very wide dynamic range of sound. Let's say from the whispers, 20, 25 decibels, maybe even fainter sounds than that. Maybe it's 0 decibels, 10 decibels, very low levels of sound. But then you have loud sounds. You might be in a rock concert, and there's 110 dB of sound, right? So this wide dynamic range of sound, it's 100 to 110 decibels. And yet the brain's mapping that into a narrower range. For somebody with hearing impairment, your hearing thresholds might be at 50 dB, but your loudness levels of comfort are still 90, 100 dB. So you got to the brain has to map along with the, with the cortical, the cortical function collaborating with the clear processing. But where is the loudness map mapping happening? It turns out most of that processing happens in the inferior colliculus. So the job of mapping the large environmental range into a shorter range in your brain is a complex function. The way a hearing aid does that, or used to do it, is through compression, right?
Makes sense. So you have compressors, but it turns out the traditional way of doing compression in hearing aids, you have to compromise between speech signals and environmental sound signals, and they're different. As you know, speech is a fast-moving, dynamically changing sound. To handle that, to compress it, you need fast compression. But for slowly varying background sound, you need a slow compressor. So for hearing aids, it's a compromise of the two. So you might benefit in one area, but it will cost you in another area and vice versa. And to explain what I heard from, you know, ears from patients is that you have to prioritise either speech recognition accuracy or sound quality. By sound quality here, they mean a subtle pumping artefact. If the fast compressor persists even after speech disappears. So, for example, you might be typing, and the keystrokes have similar to speech profiles, fast-acting sound.
Interesting. At the end of it, though, the leftover will create a little pumping artefact,
and people used to say, oh, just this is a price you pay because you're using hearing aids.
All right. So we looked at how the brain does it without pumping artefact? Right.
So it turns out the time constant of the brain is determined in the inferior colliculus.
So that's a dynamic way of dealing with different types of sound without compromising
one or the other. So this, we call it now the DNN dual compressor. It's an, you have, so the, to explain this, you have the, you have the slow compressor that's running all the time. It's taking care of your audiogram, the mapping to bring it to the patient's range.
But then the instant the deep neural network compressor detects speech, it invokes, on top of this, the slope compression is always there, a faster time constant to deal
with speech. It disappears as soon as the speech is gone and then comes back when speech is needed again. So this DNN dual compressor completes the fourth of the quad DNN architecture.
That's new in Omega AI+, and we're really proud of it. There are other things that I'll tell you, but I've learned a lot today already.
I've been sitting with you for 5 minutes.
Um, so from an end user's point of view, because anybody watching this is, um,
has probably received a lot of information with what you've just said as well, what does that mean in terms of their day-to-day experience when wearing Stark, Starkey Omega AI hearing aids?
What will they experience versus, versus previous generations of Starkey or your competitors?
Yep. So at the end, that's what matters most. Yeah. What does it mean to me that I'm, I'm
wearing hearing aids? Well, so first of all, there are two things. One is the, um, ability to understand speech in a noisy environment. That's one metric. The second one, that is the, the comfort of, of hearing, the listening effort. How low is it? So I'll give you a metric, um, for your audience. I'll first explain how do you quantify and measure what this listening effort is. Well, one way to think about that is to think of a controlled environment where I can play with the signal-to-noise ratio. So I got, let's say, speech; I'm trying to talk to you, but then I can vary the different levels of noise to play with the signal-to-noise ratio, and then I, let's say, I'm measuring your effort. How much listening effort are you putting in to understand that speech content? And now it makes sense logically: the poorer the signal-to-noise ratio or, given the same level of speech, the higher the noise, the listening effort would be higher.
With Omega AI+, we measure a 2.1 decibel advantage. That means even if the noise levels are 2.1 decibels higher, your listening effort is the same, and this is an aggregate effort of all of the neural network systems working in parallel. And 2.1 doesn't sound like much, but that's massive, isn't it?
Massive amounts, because as you know, in signal-to-noise ratio, it's a logarithmic scale.
Absolutely. So, actually, I did the calculation in the linear scale. Okay, it's 56%. Okay.
I mean, so your noise will be 56% higher and yet your listening effort is the same. It's,
it's magical for the hearing experience. You know, I've told this story to others,
but I don't know if your audience is aware. I have normal hearing. I have regular, normal hearing, no hearing loss yet, and yet I am using my devices all the time with a closed, with a custom tip on as well. I know. Yes. And I know you're a strong proponent of, personalised custom tip.
We say our devices provide you with superhuman hearing. For me, it's not a buzzword. It's not a marketing term. I have my normal hearing, yet in an environment like this, sitting in the loud, for anyone watching this. This is, this is noisy. So I would rather have the 2.1 decibel advantage, interesting, in listening effort than not. So it makes me superhuman with it. [laughter] So that's the patient advantage. This is, and I think that paints a really clear picture versus what we were talking about at the beginning of today as well. I love the science element, but I love putting it into language people will understand.
Um, one more question, and I know you're a busy man, so we've got to wrap up. You've got one extra feature which we've not seen in any other tech before from Starkey or any other manufacturer when it comes to sleep. Yes. Tell me a bit about that.
Right. So, again, back to the technology strategy we started in 2017, when I joined, when I joined Starkey, we said we're going to focus on three areas. They're not all of equal importance. The first one is intelligent hearing. That's job number one. Every hearing aid has to do a great job of helping me hear better, but once we do that, the second is health features. I want to be healthier. I want to live a healthier life. We have fall detection because older patients will fall. We want the loved ones to be alerted. We have balance assessment for them to know the risk of falling and improve on that metric. Last year we introduced respiratory rate monitoring. It's important; if it goes out of whack, you want to know.
This year, we're introducing a sleep tracking feature. Well, you might say, why do we care? You got your wristwatch. You got your rings that can track sleep. We built a unique sleep tracker, unlike anything else. It can also tell besides the quality of sleep, it can tell what side I slept on for how long. Did I sleep on my back, on my left side, or on my right side? It's important because if you have sleep apnea, your doctor will say don't sleep on your back. It make you, it will make, make the condition worse. But there was no way to track it unless you went to a sleep lab. And your smartwatch won't do that. A ring won't do that because built into the things. So, again, I don't expect everybody to start sleeping with their hearing aids. For those individuals who have sleep disorders, they will benefit from it. And our goal is not that every health feature needs to be used or be important for everyone. But, you know, fall risk is not for everybody, not for me.
But for my older parents, it's important, right? For those individuals whose lives we
saved because they fell and their loved ones found them unconscious. We have those examples. It's worth it. Same thing with sleep tracking. If you are the ones that will benefit from it, we're proud of the work that you will help them, besides helping them hear better.
I've heard this before from Starkey. So, it's not that everybody necessarily needs all of the features activated. It might be that you benefit from the sleep tracker, maybe from the fall detection, but they're all there, and so you're going to cater to as many people as possible.
And then, before I go, you, one other feature, because we have the third, I said three.
The third aspect of it is to make the device smart for you. Help you with reminders.
Take medication when you need to. Okay, the device reminds you. Tell me more. Talk to,
You know, you could simply talk to the device with our voice assistant and say; Remind me of, so what would you like to be reminded of? You could say 10:00 p.m. every night. Remind me I need to take my pill. It will remind you. My younger friends who are still using hearing aids, they're using them to remind themselves of having to pick up their daughter from school that day, right? And the new feature we are introducing with Omega AI+ is hearing aids left behind. I have a colleague at Stanford who left his hearing aids at the back of the, at the seat of the plane and walked out. So now, the moment the Bluetooth connection's broken because of the distance, you'll get an alert on your phone. Your hearing aids are left behind.
So we are making the device smarter and smarter and smarter. So fully automatic,
help you hear better, help you live better, and let it be smart to ease your efforts as you go about your things. That's our goal. Make it superhuman.
Hey, Achin, I need to get my hands on these hearing aids. I need to put them through their paces, and I need to put a video together, I think, so that I can share all of this
with the rest of the world. I'm so grateful for your time today. Thank you so much. Um,
anything else you want to share with the audience? I think you've covered everything.
Well, to your audience, I have to say, I wish I could talk to Matthew today about what's in our pipeline, because this product that's in the market in your hands is something I stopped working on a year ago, and then it's all about getting it through the production line, getting in your hands and getting to the patients. I hope to sit down with you soon again and talk about the next product.
You tease because we're just getting started.
Great to meet you. Thank you. Thank you for your time.
Comments