Inside Signia MaX Hearing Aids: Four DNNs and a New Approach to Sound Processing

Hello everyone and welcome to This Week in Hearing, live from Munich at the Signia MaX pre-launch event. Yesterday we got all the details on what makes the MaX unique and we did an excellent headphone demo, which as it turns out was a pretty realistic imitation of what we would actually experience with the devices. But I’m getting ahead of myself. After the program and breakouts was finished, we went out for the evening and had a convivial and lively discussion. This morning I was fitted with pre-release devices and told to go out and live noisy. Okay, I’m up for that.

I’ll share my impressions of the MaX as we went about the day and compare them to the IXs I also brought along. But before that, let’s have a conversation with Berinda Samra. He’s a global senior commercial audiology manager. He did a great job breaking down the MaX’s unique architecture and describing how it benefits people with hearing loss like me. But give me a moment to finish up and then we’ll get on with it. Well, thanks a lot, Berinda.

It’s great to see you. Great to see you. Appreciate you spending some time with me. Yeah, no worries. Thanks for having me. You’re welcome.

Congratulations on the launch of the MaX. From what I saw, it’s a really impressive piece of kit. I can’t wait to try it out. I think of it as like the iX on steroids. Yeah, I mean, you can see it like that for sure. It’s definitely a— we’re very excited about it.

And it’s a completely new frontier in AI sound processing. So in hearing aids. So, what was your overall set of goals in designing the MaX? What were you looking to accomplish? Well, I think it all, you know, focuses on, yeah, what is your outcome that you’re searching for? What are you trying to do?

And I think ultimately what we wanted to do was have a full system intelligence, which is analyzing multiple dimensions of your surroundings and optimizing the hearing aid sound processing for that particular moment that you find yourself in all day. Okay, so multiple dimensions. Elaborate. So, what we have in Signia MaX is a 4-core brain called Acoustic Intelligence. Acoustic Intelligence is 4 deep neural networks all working together in a fusion to analyze your moment, your sound moment at that particular time, do an in-depth analysis, and then control 4 key features of the hearing aid, again, for that particular moment and that particular time. Okay, and I think it’s worth mentioning for everybody that when you talk about 4 DNNs, that’s really a software construct.

You don’t have 4 chips in there running things. You’ve got 1 chip design with 4 different DNNs inside the chip running simultaneously. Yes, so in order— if you want system intelligence, not just feature enhancement, then you need to embed the deep neural networks deep into your chip, right? So they can reach out and actually have direct control over the sound processing. So the 4 deep neural networks embedded in the Signia MaX chip are a deep neural network analyzing solely for speech. So it’s analyzing and isolating speech.

One that is analyzing and isolating your own voice. One that is analyzing for noise. And one that’s analyzing the scene. And it’s like I said, think of it as a 4-core brain. These are 4 DNNs, but they’re all working together in concert to get this multidimensional understanding of your surroundings. Okay, and some of this is evolutionary, right?

I mean, scene decoders have been around for a long, long time. Okay, so what, it’s been probably 20 years when you could identify scenes and adjust hearing aids in one form or another. What have you done to take it to the next level in the MaX? Yeah, I mean, we could also say that noise reduction had been around in one form or another for 20 years, but we introduced deep neural networks and AI for deeper analysis of that noise to then get better noise reduction, right? So we have a deep neural network trained on thousands of scenes analyzing the characteristics of different scenes. So exactly in the same way, like, yeah, you can have scene classifiers and you, you can classify scenes, but they use finite rule-based algorithms.

We have an AI powering our scene analysis that’s been trained on thousands of different scenes. Okay, and so that’s something different than the iX? It’s a whole new ball game, yeah. So it’s moving like exactly like with how we moved in noise reduction from machine learning algorithms to AI or deep neural networks to improve and enhance the noise reduction through this better analysis. That’s exactly what we’ve done in this particular one we’re talking about, scene, but then we have in every other dimension we’re talking about and working in concert together. So presumably then, tell me if I’m wrong, but as an end user, I’m gonna notice a much smoother experience as a scene classifier, you know, as I move in different environments, it’s gonna be more smooth and flawless and more optimized in any given scene.

I think I’ll actually flip that on you a little bit there actually. As an end user, if we’ve, done our jobs right, and I believe we have, you shouldn’t notice anything at all other than it sounds right all day. Okay, we’ll put that to the test. Please do. Right, because for example, I’ll notice with some devices, if say I’m walking down the street and it’s quiet on the street, open the door to a noisy pub, right? And at the very beginning you get a blast of noise and then you hear the Scene Classifier go to work.

Then it’s, you know, in the right mode. But this is going to be more seamless, so it should be less noticeable. Yeah, so we don’t we don’t generally believe in this idea of, oh, you’re in this mode, let’s, you know, change, now you’re in this mode. It’s, it’s much more fluid and adaptable and like this constant analysis because that’s life. Life flows and is constant. You don’t, you know, you don’t just suddenly end up in a different mode in life, do you?

So that’s how the Ambient Scene Adaptation is also designed to be, to have a flow and an ebb and adapt. It’s going to be fun to experience that as we go about the day wearing them today. Yeah, yeah, definitely. And you know, it’s funny, isn’t it? This is one of the things in audiology, like sometimes, you know, you could make something instant, it’d be easy to do it. You could instantly change a setting, but human beings don’t react well usually to instant changes.

So you have to bring in these slow, you know, bringing one thing down and bringing another thing up fast enough that it’s reacting, but adaptive enough that you don’t notice it. And that is one of the beautiful challenges of hearing aid sound processing. Okay, and one of the D&Ns is running your multi-voice tracking system, which is also in the iX. And I mean, it really works in the iX. I mean, I’ve had the iXs for a while, it works really well. What’s different or better about it in the MaX now?

So in iX, we had real-time conversation enhancement that is now evolved and becomes real-time conversation enhancement AI. So we go back to that same analogy of we always, we’ve had noise reduction in the past, we’ve had real-time conversation enhancement in the past, but now that machine learning algorithm that was in the background, and I’m not going to pretend to be an expert in the math there, but we know there’s these rules that you put in if it’s at this frequency or this type of fluctuation, that’s most likely speech, right? For example. Now instead, we’ve replaced a machine learning algorithm with a deep neural network trained specifically and only on speech, on analyzing, detecting, and then isolating that speech signal, right? And it’s that in-depth analysis given to us by the AI, just like the in-depth analysis given to the DNN noise reduction systems improves the noise reduction, that in-depth analysis on speech alone, that’s its entire focus, is constantly listening out for speech, allows to better isolate and then therefore enhance speech in multiple languages.

So, iX was running a machine learning model, really looking at acoustic signatures of speech and tracking them, whereas you now are actually, because of AI, you can actually identify speech and be more accurate about what you’re identifying as speech and following it as people move around. Exactly. And, you know, this acoustic intelligence, this 4-core brain with these 4 deep neural networks, is being trained in one of the most diverse training bases ever applied to AI in hearing aids, right? Because we’ve got speech signals, own voice signals, scene signals, right? Noise. It’s been trained on 40 million samples overall.

So, you know, it’s not just, I’m not, to be honest, if I’m gonna be, the number, okay, you get the right number, you need a number to train, but what sets it apart is the diversity of that training and the function of each one of those cores. So different voices, different languages. Yeah, exactly. Like 11 languages as well, you know, music. There’s all sorts that has been trained on to do these separate functions and then bring them together. So one of the more interesting parts of that is that you’ve now also, you’re now also able to identify own speech, right?

You can identify by the acoustic signature when the person themselves is speaking. So now you don’t have to train for it, right? Because I went through the training— sit in the room, count to 20 or something like that— to train the own voice processor in the IX. Now you can do it live without any training at all, correct? That’s absolutely correct. And You know, I love everything we’ve brought out and obviously bringing a new frontier of AI, sound processing, sound steering, system intelligence, all these things are so amazing, but I’m quite excited about OVP AI.

I’m quite excited about it. So no longer do you have to train. We have acoustic intelligence, we have that Own Voice DNN, and it has been trained on hundreds of voices, right? Thousands of voices, different languages. It can now on its own without any training pick up anybody’s voice out of the box, right? Yeah, so it doesn’t matter what your voice sounds like, what language you’re speaking, it can identify own voice instantly.

Exactly, and what we saw with OVP 2.0 in IX is that when wearers did go through the training, right, so OVP was then switched on, we looking at the data, we saw an up to 30% decrease in return rates from having this one feature switched on. And I’m so excited, happy, and proud that this is now there out of the box, day one, for everyone, baked in. So everyone can get this comfort of your own voice without affecting clarity of external talkers. Okay. And so, you know, one obvious benefit of not needing the training is it speeds up the workflow for the clinician. But from the end-user point of view, were there problems with the training version that you are now solving with this one?

Simply that, yeah, you know, Nobody likes to have to have an extra thing in their workflow, is what I think, really. And I think as well, probably psychologically, if something’s not baked in, maybe you don’t really fully feel its importance, right? If it’s an optional thing that you can train on. So I think now, just the idea that the HCP doesn’t have to worry about it, there’s no extra thing in their workflow, it’s baked into the hearing aid itself. I think it’s just about, it wasn’t necessarily like, oh, it’s so difficult to train, or so, you know, it’s an extra thing. And now that’s gone, and now everybody can just have it.

And that’s the exciting thing. I suppose if you’re doing, say, remote-first fitting, for example, for somebody, then you couldn’t do the training in that case, but now you don’t have to. Oh yeah, yeah, I mean, yeah, there’s probably logistical challenges that could come in, and Yeah, look, you, you’d have to make sure you’re in the correct type of room, for example, right? You could sometimes some people would do the training and they’d be next to a wall or something like that and the reverberation could change it. So yeah, you’re right. Even during the training, there could be issues and they’re just all gone now.

So, okay. And then the fourth DNN running is for noise reduction. Mm-hmm. Yeah. So we have a deep neural network for noise reduction and What that does is it not only is it like, oh, there’s noise, it’s also the, the type of noise as well. It can tell the difference between, between babble noise and, for example, traffic noise, because depending on the type of noise, you will need different amounts of noise reduction.

And if I may, I’ll just take it a step further. This is where acoustic intelligence comes in, right? Because, okay, we have a DNN, for noise reduction. We also have a DNN for your voice, and we have a DNN for external talkers. Now, I said they work together, right? So this 4-core brain, what it knows is, okay, you’re in noise, check.

Are you in conversation, or are you not in conversation? Because that’s going to change your moment. It’s going to change what you need and what you want from your hearing aid. Right? So for example, you walk into a bar, you’re waiting for your friends, they’re not here yet. You sat there, you’re not in conversation at all, right?

In maybe some, what, you know, one-dimensional hearing aids are only focusing on noise. Oh, noise limit hit, let’s go into this noise reduction mode. Woom. Okay, you’re comfortable, but you are kind of unnecessarily isolated from your surroundings, which is the exact reason you go out to meet your friends, right? Why should you just sit at home, right? Yeah.

So with Acoustic Intelligence, it knows, okay, you’re indoors, there’s babble restaurant noise behind you, it’s quite loud, but you’re not in conversation. Okay, let’s reduce it down, take the edge off. But let’s not go full bore. Let’s leave it. Let’s let some of that come in, let you enjoy and live in that moment, right? Then your friends turn up.

Again, if you’re one-dimensional and the biggest focus is noise only, well, okay, you’re still in noise, you’re still hitting that limit. Okay, we’ll still do noise reduction. Maybe we’ll give you a directional beam. Acoustic intelligence, Yes, you’re indoors, you’re in Bubble, it’s the same environment, nothing’s changed, but people are talking to you and you’re talking to them. The same environment, but your moment has changed. So your needs have changed and that’s when it kicks in.

Okay, let’s turn that noise reduction, you know, let’s make the noise down, let’s enhance that speech, let’s give them the contrast. This is like one example of what we mean by system intelligence and AI steering of the sound, not just individual feature enhancement. So yes, we have the individual feature enhancement, but the beauty is in how it kind of all comes together and flows in real time. Yeah, they’re all working in concert to give you the best sound depending on what you’re doing. Yeah. Now, is there any end user control over all this?

Obviously, they have that app. They can make changes in the app like they have been able to do in the past. But in terms of control, as in— So for example, my wife and I go to a café in Paris. The tables are on top of each other. There’s a couple of loud people having a very animated conversation next to me right here. I don’t want to hear them.

I want to hear her. Right? Other times I might be sitting in a group of 6 people and I want to hear the people over there. The hearing aid isn’t going to know if I don’t want to hear those people or if I do. So do I have the ability to influence how the hearing aid behaves in a situation like that? So you still have control in the app.

You could still have a control of going to a different program, etc. Yes, absolutely. It is designed, however, obviously to be as automatic as possible. and you know, the people that are talking to your left, by the way, you know, if they’re not facing you, that actually changes the acoustic properties of speech, by the way. Like, someone facing you it, it lands into the microphone differently than it does if they’re not facing you. So there is things in there like proximity detectors, reverberation detectors, to try and well, you know, figure out, is this person actually in conversation with you or not?

So ideally it would just work. But if you do, and I know it’s all about empowerment as well, you obviously do have control over your own hearing as well, and you’ve got the app. And so I can, for example, light up a beamformer and go forward if I want to. Yes. Right. Because again, like a table of 6 where there’s 2 people on either side of me at my table, they’re off to the side of me while I’m talking to the person here.

I still want to hear them, but other times I don’t. So I have enough manual control for situations like that. Yeah, absolutely. You still got like human involvement. It’s your hearing aids, it’s your hearing, and no one’s ever we’re gonna— we’re not quite at the moment yet where we can read your mind, right? And you need to have that control.

Yeah. And you’ve got it. Perfect. Well, it’s going to be a lot of fun to try that out. Now let’s talk about the basics of the device. What form factors are available right now at launch time?

It is the RIC device that’s available. Okay, so you’ve got one rechargeable RIC to start, the most popular form factor, the Pure. Yeah. Yeah. And then will the line expand over time? Absolutely.

This is the foundational platform, the foundational chip for the foreseeable future. And you know, at Signia, we really want to reach as many people as possible, right, with this life-changing thing of hearing healthcare, right? Getting that hearing, getting back into life. And one of the things we, do to do that is innovative design, right? To try and meet people where they are, to reduce any barrier that we possibly can to getting people to get in this life-changing benefit. So that’s not going to change.

That’s still our philosophy. And you will see us definitely expand the line and some of our innovative form factors will be coming as well. Okay, excellent. And so the RIC device is about the same size, if not exactly the same size as EIX. But you’ve got this powerful set of DNNs running in it. How’s that affect battery life?

Well, the battery life is still up to 30 hours. So of course, we’re running 4 deep neural networks all day, all the time, but you’re still getting 30 hours. So the chip, while it is 52 times more powerful in terms of processing power generation on generation, it’s also 50% more efficient in battery life because For us, we believe that for it to be truly intelligent and truly have an impact, you know, it needs to be available all day. You need to trust it will be there when you need it. And, you know, what’s a brain if it switches off? Okay, so running full-time, 30 hours battery life including some calls and streaming, so no problem, seamless all day.

Exactly. Okay. And speaking of phone calls and streaming, what are the connectivity options? Okay, so we have One Connect technology. So in one single form factor, one device, our RIC, we have Bluetooth Classic, LE audio, telecoil, and it’s Auracast ready as well. Okay, so you’ve smashed the BCT and the standard IX together because you’ve got Bluetooth Classic plus LE audio.

Yeah, so I, I think it’s the most comprehensive offering in terms of connectivity, and it’s, it’s all-in-one device. And really, the key here and why is that? We think connectivity is so important in reducing barriers that we just don’t want it to even be part of— it doesn’t need to be part of the conversation anymore. You know, you’re here to, you’re here to help people hear, right? You can do that with Signia, and you don’t have to ask what phone they have. So, and I’ll tell you, I’ve had people tell me connectivity is like an annoying thing that we’re forced to put in hearing aids, but It is absolutely essential.

Essential. I mean, so many people live half their life online. I mean, I’ll go home when I’m back home, I’ll, you know, I’ll have half a day of internet meetings. Well, I’m Bluetooth to the PC, right? I don’t hear very well even wearing the hearing aids if it’s coming out of the PC speakers. But if I’m direct streaming, it’s great.

Phone calls, listening to podcasts, all those things, and going to, you know, live theater or live concerts and being able to use the assistive listening system. All part of a person’s lifestyle. Now, the interesting thing about yours is, as far as I know, this is the first hearing aid to abandon the proprietary connection schemes. You’re not doing ASHA and you’re not doing MFI. You’re doing— with two, you’ve got the whole ground covered because you can connect with anything with Bluetooth Classic, but now you’re supporting LE audio. Like my Samsung phone, I would use LE audio.

Exactly. It’s those— with those two connectivity options, you can connect to all, so it becomes a bit redundant to start adding on these extra options when you’ve already covered it with the two, right? Well, then MFI is on its way out, right? Because, because US law, right? I think it’s the end of ’27. I’d have to go look that up, that proprietary schemes can’t be used anymore.

Exactly. We just, you know, what we wanted to do is get in that Bluetooth Classic so everyone can be confident that it can connect to anything. Today, get in LE Audio for the people today, but also then into the future as well as this becomes the, the, the new standard, right? So it’s just about trying to give— it’s kind of a bit weird, isn’t it? It’s a bit of a contradiction. Connectivity is so important that we wanted to be able to offer the connectivity option for everyone so we can stop talking about it.

Do you know what I mean? Yeah, right, right. Because it’s so important. It is important, and hats off for putting a telecoil in it too, because you walk into a venue like all the theaters in Chicago have FM system. Well, if you don’t have a telecoil, how do you interface to an FM system? Kind of, I love telecoil.

It’s kind of that, you know, it’s that old faithful. It’s that old reliable, isn’t it? Like all these new pretenders have come, but it’s still there because it’s so ubiquitous. It just works and it’s everywhere. And so of course we want to put it in, ’cause we want you to be able to hear wherever you go, right? So yeah, multiple options, and we’ve still got the telecoil there.

Yeah, that’s all terrific. That’s terrific. And how about waterproof? Yeah, it’s IP68, and it’s got a new plasma-enhanced vapor coating as well on the internals. So this is where you get this chemical mixture, right? It’s just there to protect, waterproof and protect, and you actually turn it into a vapor, a plasma vapor, and then kind of breathe it onto all the internals.

So, it just gets into every crevice and everything. And then on top, you’ve got the casing, right? So— Seal case, right? Exactly, yeah. Okay, so excellent. That’s another one of those things where I’ve had people tell me that’s kind of a gimmick.

Well, you go to the beach, you go out on the water, you want to be able to talk to your friends. Yeah, exactly. Look, it’s IP68, that is the standard, the same as the rest of the industry. So, there’s nothing taken away there. Yeah, so no, that’s great, that’s brilliant. So what else should people know about this device?

Oh goodness, where do I start? Where do I start? There’s, I think we’ve just put a lot of thought into it and a lot of effort. There’s a lot of kind of marquee features that we’ve just talked about today, marquee kind of new frontiers in sound processing. But there’s also some really cool, I guess, more, you know, on the ground features like, Sound Smoothing 2.0, right? That’s now high-resolution transient sudden sound reduction without affecting speech.

So, pens clicking, keyboard clacking, paper rustling, down, made comfortable, speech is still enhanced, no effect on speech. That’s gonna be a game changer for acclimatization, right? And I could go on and on. eWindscreen 2.0, the Signia Assistant, is now the next generation. You can just chat to it, just chat away. Like, think of the empowerment that that will give, right?

That you can just ask this assistant a question, and it’s not just on the hearing aid, it can be about the hearing acclimatization journey. So that support in the first week, getting some of those worries off the top of your head or whatever it is, so then you can go and when you meet the HCP, rather than asking them, I don’t know, How do I change the ear tip or whatever? You can ask the assistant. Yeah. And instead you can talk to HCP about your journey and your acclimatization journey and where you need help with the audiology. And you can get fine-tuning changes through it too, can’t you?

So for example, if I’m in a noisy restaurant and I say, I still wish I could hear my companions better, through the assistant, it’ll make changes and try them out, correct? Absolutely. And the key with that is, the good thing about that is that you can get optimizations that are within parameters and, and banded. It’s never going to go completely off-piste, right? Within, within parameters, you can get optimizations in the very moment, right, that you’re in. It’s going to analyze your surroundings and then suggest something.

And what I really like about that is that you don’t have that, maybe that to and fro and follow-up points. They’re trying to remember what the situation was, etc. You can get it right there and there. You do it while you’re in that situation, it’ll analyze the situation and say, try this. Yes, exactly. And it’ll do A/B.

And crucially, the HCP is in total control when they go to their appointment. The HCP, HCP can see, okay, this is what was done, this was done, and have the conversation with the, with the wearer. Okay, that’ll be interesting to see how many people actually use it. So when I look at statistics from the Market Track in the US or Eurotrack, you still have roughly half of the people not using the smartphone apps, but that’s changing rapidly. As with anything, right, with apps and smartphones, everything, you do get this first it’s slow and then all of a sudden— and I think the way I see it is, and the way I’ve always seen it, is it’s the Signia Assistant, our next-gen Signia Assistant. Yes, for the wearer, but really, and my passion is also an assistant for the HCP to free them to be able to focus on what really counts.

Yeah, right. So the, the hope is that more people use the assistant and not go to the HCP for, how do I change those wax guards again? Exactly. Instead, he can ask, hey, how do I change the wax guards? And it’ll just pick up the information from the guides, videos, everything you need. Right there, and you can just talk to it.

Oh, terrific. So really exciting. I like— I’m thrilled to try this device. I tried the headphone demo. The headphone demo was really well done, right? It was very obvious in the headphone demo what you’re doing.

And I have to say, it’s actually the best demo I’ve seen because, you know, you see on the screen the beam’s tracking and so on. So it’s nicely done. And after having listened to the headphone demo, I’m really looking forward to trying the devices, swapping them in and out with EIXs, and, you know, sharing my impressions. So happy to hear that. We’re so excited to get it out there and get people getting the benefits of the Signia MaX. Well, thanks a lot.

I really appreciate you spending some time with me. Cheers. As I mentioned at the beginning, we were fitted with pre-release MAXs and sent out for the afternoon. This was in the form of an organized Munich city tour. I had an ace in the hole, my set of BCTiXs. To borrow WSA’s terminology, let’s do a sound preference test.

My first surprise was right at the beginning. We took several taxis to the meeting point, the fish fountain in Marienplatz. Those of us who got there early waited for the others while having casual conversation. I had no problem understanding everyone, even with all the plaza noise. This was in automatic mode. Out of curiosity, I went to Omni mode to reduce the DNN processing.

The noise level increased, and most surprisingly, I could hear the bubbling fountain. Back to automatic mode and the fountain was gone. I mean gone. It was clear right then how the MaX was curating my experience to emphasize conversation focus. As we went about the walk, I switched back and forth several times. In each case, I had a clear preference for automatic mode where the D&Ns were fully engaged.

At one point, we stopped in a beer hall to have a bite. There were a dozen of us or so at a long table. It was only moderately loud, not enough for a true test. Even so, I was more comfortable listening in automatic mode and and able to understand everyone without strain. Then I swapped out the MaX for my iX. I always thought the iX was a very competent device, but in the restaurant and afterwards, I definitely preferred the MaX.

Listening carefully, it became apparent that the iX sounded more processed and the MaX sounded more natural and easygoing. I thought afterwards with a smile that this is going to throw a monkey wrench into the sound preference test. I’ll be very interested to see how that plays out. I’m curious to do more control trials, including in loud noise, to compare the two. I am also curious just how curated the sound becomes in different situations. For example, the fish fountain experience made me wonder if my wife and I are walking in the woods and talking, will I still hear the nearby brook or the birds singing in the background?

Fortunately, with thanks to WSA, I have the MaX coming so I can play. Stay tuned for that. For now, I’ll say this: the MaX did an excellent job on speech awareness and comprehension, including in some complex situations. Nice job, Signia. You’ve made clear, meaningful improvements I’m sure others will also appreciate.