Exploring Starkey’s Edge AI Hearing Aids: Always-On DNN, LE Audio, Health Features and more

Hello, everyone, and welcome to This Week in Hearing. And welcome to Starkey Day, where I had the pleasure to interview Achin Bhomik and Dave Fabry to discuss their latest device, the Edge AI thanks for joining me, Dave, and congratulations on the Edge AI launch. Thank you so much. It’s great to see you again, Andrew. I’d like to actually break down the Edge AI in terms of its audiological improvement. Of course, the Genesis was a leap for Starkey. Yes. And what’s further improved with Edge AI over Genesis? Certainly. Well, as you know, when we launched Genesis, we were fiercely proud of the fact that we had 118 decibels input dynamic range. That still is the largest dynamic range in the industry because we know that in terms of intensities, being able to reach and provide gain and amplification to patients with minimal hearing loss all the way up to severe to profound hearing loss requires that large input dynamic range. We continue to offer as broad a frequency response as possible, but where we’ve seen some improvements is in particular with our DNN based feature that began as Edge mode. And you and I have talked about that in the past as a situational feature where if someone. All hearing aids now use machine learning classification systems that enable the patient to go throughout their day and identify whether it’s a quiet, noisy, musical, windy, whatever environment, and then apply only those features necessary in those environments. And I’ve seen across generations that’s gotten better and better. With my older hearing hearing aids, I still would have to fiddle around and create more custom modes. And with 2024 hearing aids, it’s working much better. It’s great that you say that, because what we found was what was designed initially as a situational application Edge mode, where I would double tap or push a button in the app to do an acoustic snapshot in a challenging environment. Now, when we launched this in 2020, we talked about it primarily as speech in noise, because that’s the number one challenge that first time hearing aid users. Have, and experienced hearing aid users, too. Everyone, speech and noise is the brass ring to show my, okay, Boomer moment for them from the merry go round and reaching for the brass ring. So speech and noise is always the biggest challenge. But we also know that when we launched that in January of 2020, 3 months later, everyone was in lockdown in the US and people were wearing face masks. And I had all my patients who said, hey, you told me to use that in noise. But now I’m finding when I encounter someone behind a plexiglass or with a mask on, it’s helping make sounds more audible for me. And so we began to see, okay, challenging environments come in a large variety of difficulties, quiet, noisy, windy, et cetera. And so we started to expand that feature, and many patients started saying, hey, if Edge mode is so good, why don’t you use it all the time, rather than the personal, what we call our personal program, our acoustic environmental classifier program. And we said, well, you don’t always want to have things as sharp as when you’re doing an interview or doing a presentation and wanting to listen at a conference. Some other times, like later this evening, if I’m visiting with some of the people that came and attended this event in the bar, and it’s noisy and I’m a little more tired, then I may want to reduce and be more aggressive on the noise management settings. So even within the same person, your intent across different environments changes. And so we said, well, we’ll still use it, use the personal program. Maybe we can fit up a restaurant or a crowd or an outdoor program and then use Edge mode just in those challenging environments. But we found, I’ve got a number of patients that said, I’m hearing great with the personal program, and I don’t want to have to tap in and tap out of edge mode based on the environment, because Edge mode, initially when we developed it, didn’t continue to adapt over time. They had to go in and go out and go back to the personal program. Then we said, now we’ll give you more granularity to enhance audibility or reduce noise. People like that. But then my patients, who just want seamless technology said, I don’t want to fiddle with it. I don’t want to engage with my devices that much. And I’ve said for a long time, you don’t know your patient until you know your patient and understand, and you don’t know from their education level, from their background, from whatever you got to learn. I’ve got patients that are smarter than, way smarter than I am that say I just want to put my hearing aids in and just use them. So when we continue to explore this track of using situational DNN to now where we have always on DNN, when you ask about where did we go from Genesis into edge AI, now we have this always on DNN with 100 times the DNN processing capability from what we had even in Genesis. And it enables that to continue to adapt throughout the day. So now my patient, who wants seamless put their devices in and say, double tap or button press into improved clarity, and it’ll just continue to adapt throughout the day. So you could see people actually using edge mode plus on continuous basis. I have patients that do just that, and they’re using it continuously. They may start out. We have three modes. So best sound is sort of more aggressive offsets than our personal AEC program. But it’s a balance between more aggressive noise management and more aggressive clarity. Based on the acoustic snapshot, then others, we just can go to enhance clarity or reduce noise. And I have patients that set it in one of those three and go throughout the day and just leave it. And just leave it. But then maybe as they go in the, hopefully during the sessions today at the conference, they wanted to be on enhanced clarity as the social event, the dinner. Then we want to reduce the noise a little bit more in the same person, because even within the same person throughout the day. My intent may change, but to have it continuously adapt is one of the big innovations with this product. That’s one. The second relates to the Auracast omnicast, which you’re an expert on. And, you know, I’ve heard you present eloquently on this many times, and I’m really excited about the future for Auracast. And I shared with me the story of you know, my enthusiasm and excitement is being able to walk into a sports bar and be able to use a QR code and listen to my team, even if I’m living or visiting somewhere where they don’t like my team. This was actually the best example I ever heard. You walk in and the Vikings are on the big screen, and you’re a Packer fan, and you’d rather watch the Bears beat the packers on the small screen over there and hear the sound of it. It was really close to being correct, except I’d flip the Bears and the packers knowing. But, yeah, I mean, to me, that is a phenomenal application. Movies, places of worship, all of those areas are ways that you’ve talked about this, and you’ve taught me a lot about the applications from your presentations, and I just think it’s a game changer. They’re talking two and a half million places in the US alone by 2030 that are going to have. That’s still a few years away. It’s closer than we think. As I get older, the clock ticks faster. But really, just a few years from now, as movie theaters, as bars, as places of worship upgrade their systems, we’re going to see more and more compatibility with the hearing aids that have Auracast, omnicast with the Bluetooth standard, the BLE standard. And I think the reason that I’m more optimistic about this change, because we’ve lived in a space where, with telecoils, and you go to a place of worship, and you say, okay, well, they installed the loop, but it’s primarily used by a small number of people within a congregation, and there really isn’t a widespread utilization. But I think with Auracast omnicast, there is an opportunity to be used by people with hearing loss and people who just happen to have wireless headsets that are compatible with that and also want to use that same feature in the bar. That’s why I think the sports bar is a great analogy, because it’s a place where people gather, it’s noisy, and they may have different interests. Someone wants to watch hockey, someone wants to watch basketball, someone wants to watch football, and those are all known to go in and have the sound to go along. Yeah. So you and I are very like mine, then, because many of us are living as much in the virtual world as in the real world, with Internet meetings, with consuming audio or video, with audio, going to public places, and to be able to seamlessly connect all those is part of the living. With hearing loss experience as well. It’s not just in person communication, but it’s how do I hear well when I’m having virtual meetings? Yes, 100%. So, really, that Edge mode application that can now be continuously adapting is one thing, the Auracast omnicast capability, and that future proofing. We know that the infrastructure is going to have to develop, and so people have to be a little patient, but we’re trying to future proof the devices as much as possible. And then the third has really been our commitment since 2018 to health. And one of the things, we were the first in our industry to incorporate inertial measurement unit sensors to enable physical activity to be tracked. Because of the link between hearing loss and cardiovascular risk factors, cognition. We’ve engaged in our My Starkey app with the ability to track social engagement, not monitoring what people are saying, but looking to see how social they are. Looking to see when they’re talking and when they’re listening. Is that what you’re. And we’re getting more and more granular with that capability to be able to differentiate my voice from your voice. And so that social engagement and the physical activity have been central. But then, as we discussed in the launch today, we’ve had a fall detection feature since 2019. Excuse me. And so that’s a fantastic feature in the standpoint of if we know that hearing loss, even a mild degree of hearing loss, elevates the risk of falling by three times and that increase with increasing loss that people who want to live. I just became a senior citizen and Medicare card holder this year. You’re ahead of me. by one year. One year. Well, I’ve got lots of advice for you on Medicare, but the issue is I want to live independently. My wife and I live near my granddaughter and soon to be grandson next year. And we want to be as engaged as possible throughout every stage of life. But I find that even though I try to stay fit, I try to stay active. I’m not as good with balance gait and strength as I was 20 years ago, even ten years ago. And so one of the things that we thought is that while a fall detection feature is great, being able to alert up to three people via text message, if I suffered a fall when wearing devices, an even better feature would be to try to use those inertial measurement units to assess what is typically a clinical evaluation, to assess the elements of the steady protocol developed by the CDC. STEADI is an acronym for STopping Elderly Accidents, Deaths and Injuries. And typically in the clinic there’s a series of four tests that are done that are assessing your balance while you’re standing with your feet next to each other, instep, serial or standing on one leg for a certain period of time, and that’s scored by a professional. Second one is standing and sitting in a chair without using the sidearms to do that. And the third is just sitting in a chair, standing up, walking 10ft. Turn around, sit down again. We’ve got the sensors embedded in the hearing aids. We thought we could do that and so we partnered with Stanford and we continue to have a rich partnership with Stanford to have them. We built a product I think it’s leading to. My question is, how well do you correlate the assessment in the hearing device with a professional assessment? Really well. And I tell you, I’m going to put this to the test because I’m in the middle of PT for balance at this very moment and I’m very curious to see after I get fitted with the Edge AI and see. We have very high correlation. We have a study that. updated, right? And we just got accepted for publication Otology & Neurology. I’ll get you the reference. Literally, we just got the word that our publication in a peer reviewed journal, Otology journal, was accepted last Thursday. So is the author proof online now then? Not yet. I think it will be any day now, but I’ll find out. And authored by the primary investigators out of Stanford and the Starkey team. So it’s truly a collaborative effort. And the very first phase was to see with three judges, human observers in the room, working through that balance, gait and strength testing and having them score it, and then compare to the automated scoring using the IMU and the app. And then we went to the professional monitoring remotely. You talked about looking at screens. We wanted to take this in iterative fashion where then we had excellent agreement between the in person observers and the automated scoring. Then we went to a remote telehealth kind of condition. And then the third phase is unsupervised because that’s really the goal, just to your point, going through a PT event and then working on strength, gait, or balance, then seeing if that could be combined with balance training exercises, that if you identify that you have an issue with strength or balance or gait, you could have just one or two of those. So do you actually have the exercises built into the app as well? Coming. Right now, we have the STEADI protocol in the app that enables you anytime you want to go through the. Yeah. You can assess, and if you’re concerned about your balance as a result of the assessment, you’ll go to PT. Today we’ll partner with a PT or an audiologist who’s working with vestibular, but more likely a PT, OT, and then work to identify balance, strength, a gait, and then specifically address that and then see if there’s improvement. I can tell you focus group of one. On my case, my scores improved after I went through PT to strengthen my legs and balance. I had a strength issue and a balance issue, still not as great on balance as I used to be, but I can stand on the bosu ball now and bring one leg around and not fall off of well, and I’m. Early enough in it that I’m seeing it. It varies day by day, actually. Right. There’s some got to get sleep, some worse days. But it’ll be really interesting as I go through more of the PT, but even more so when I actually run the balance assignment. Yeah. I want to hear from you as to how easy it is to run through the test. And then also in combination, it’s really, you’re the prototype of the next phase of this study is then combining that with a PT or at home exercises that are focused on those areas. How do you encourage the hearing aid wearers to actually do the balance assessment? Well, by trying to make the user experience and the user interface with the app as easy and seamless as possible. Is that built into the hearing care professionals training regimen? To make sure that people understand that that feature is even in there. It’s a great question, and in my opinion, at least, we work with dispensers, hearing instrument specialists, and audiologists. And you can argue I work with many hearing instrument specialists and many audiologists, and I think I’ve worked with some outstanding friends that are in both campsite. I will say that audiologists, by virtue of their AuD degree, are working with hearing and balance within their scope of practice. Some states will limit the scope of practice for dispensers more narrowly on hearing aids, whereas the audiologist has a scope of practice that includes hearing and balance. So, perhaps in an umbrella standpoint, I know a lot of dispensers that are very keenly interested in working with patients with a balance, even if it’s. Outside their legal scope of practice, according to their state, they can still point out to people that this feature is exact and that they can perform evaluation. Exactly. So I wonder if that’s part of your training program. That HCPs are aware that that’s. In there have to be, because it’s a tool in their tool belt. And then, as I said, it’s really up to their comfort. What I talked about today in my presentation was, if you’re not comfortable working with that element, be aware of the fact that hearing loss comes with an elevated risk of falling balance problems. Remember that hearing loss rarely occurs in a vacuum. Hearing loss has high comorbidity with cardiovascular risk factors, cognitive issues, as we’ve seen, falls vision. As you get older, you’re more likely to lose both hearing and vision. So we have to think about that when we’re designing apps to make sure that we’re using bigger fonts, or the capability of bigger fonts and controls. I talked about the fact that with, with Edge AI, we’ve got a new Auracast compatible remote control. We’ve got voice control and voice. Control, which is really fun because that’s really the other element that’s brand new in this as well. We’ve had elements of the smart assistant in the past, but then really the Gen AI virtual assistant that Achin talked about, and he gave a demo of that today. The reason that we’re excited about it is it truly is intelligent in the sense that it can triage. Whether you’re asking a question about how to use your hearing aids, and it’ll go to the manual and populate in the app, we have a learn section where it’ll populate videos that shows you how to clean your hearing aids and really augment what the hearing care professional did when they fit you, just to remind you. And so it’ll automatically triage you into the app to show you or instruct you about your question. But to be able to get assistance on your devices or to be able to control your device if, for example, you’re mobility impaired or you just happen to have your hands full of, get your hands full. I mean, the same technology that kind of brought people to be able to open their, their trunk of their car when their hands are full, we want to be able to say, hey, change my volume. Not only to say, you can ask it and say, hey, how do I change the volume on my hearing aids? And it’ll show you in the learn section how to change the volume of your, say, raise my volume. You saw that in the demo. I use that feature. You can say, change to the outdoor program. Now, if the professional didn’t program an outdoor program, then it won’t know how to change that. But if it’s one of the programs that you have embedded in there, it will know how to do that. And then you can ask about the weather. You can ask, what is there to do in Minneapolis? You can do all of that. And it does remind me just one brief segue you mentioned earlier that with iterations on your generations of hearing aids and your hearing aid experience, the thing that I’m finding is what we call our personal program. And then those manual programs, situational ones, for outdoors for restaurant or crowd, increasingly, I’m migrating my patients from multiple manual programs to the personal all around program and Edge and in that case, edge mode can now be applied situationally or left on if they want it, like we discussed earlier. But I’m seeing that even when the hearing care professional applies a specific transportation program, Edge mode adapts, optimizes in that situation within that program. And so we’re pretty excited about that too, because it’s all the user interface, whether it’s the balance evaluation, whether it’s looking at using the intelligent assistant to be able to command or control your hearing aids, to ask about the weather. For those who are, like you said, I’m with you. If I’m outside, I don’t need to ask about the weather. But if I want to say, hey what is there to do in Minneapolis? It can have that conversation. It’s not going to be for everyone. But increasingly, a lot of people are using Siri and Amazon. I use Google. I mean, I’ve got Google Assistant in my ears. I use them. Yeah, I use it very often, actually. Right. And the user interface for us is what’s so important that we can have the patient stay within the My Starkey app and use that feature so that it will triage automatically without having to think about going to use a different application. And it will just show how to use their hearing aids, command and control their hearing aids. Deal with prompts. Make a reminder. They can do an audible reminder to say, hey, don’t forget I’m meeting with Andrew later today at 3:30. And it’ll do a daily, weekly reminder medication. This really helps us into our health presentation. Remind me to take my Medication, Lipitor. Or whatever at 10:00 p.m. tonight. We know in the aging population that adherence to chronic medications is poor, often less than 50%. And just by using natural language processing to be able to remind them, as long as they’re still wearing their hearing aids, the risk is, if they’re not wearing their hearing aids, they don’t get the reminder. They’ll see it, but we want them to wear their hearing aids. Another motivator to wear your hearing aids from when you put them in, in the morning until the end of the day. Yeah, I mean, there’s so much around that. We could have a whole podcast around that, right? Yeah. And how the brain works and why you need to wear them full time to adapt. But I want to go back to the balance for one other question that is. So, right now, you run a discrete balance assessment. Correct. But of course, the machine learning capability is getting so much better now. Do you foresee a time when actually it would do balance assessment on the fly? It’s a very good question, because, remember, we built that IMU from the start. We already. I’ll just leave it with this. Already we know that we can automatically, some might say auto-magically, detect whether a person is walking, running, doing aerobics, or riding a bike. You even said stationary bike. Yeah, stationary bike. The signature of stationary biking has got to be pretty subtle. It is. Because you’re not moving. Right. And your head’s not moving. Correct. And you’re able to detect stationary biking versus something else. And I can tell you, I use that all the time where I don’t have to say, all I say is I’m doing a workout and I do that between my watch, but it will automatically start aggregating the activity in biking. And I’ll try to switch between a bike, an elliptical go on the treadmill, and it will give me credit in each of those different buckets for the different activities without meaning to intervene. And hence the question, because right now, for example, you’re doing a running engagement score. Yep. When do you start doing a running balance score? TBD. That’s all you get from me today. Now, we talked about the audiological benefits a little bit, but let’s get a little bit specific because it was said that Edge AI offers up to 13dB SNR improvement in diffuse noise. Is that versus nothing at all? Yeah, this is an issue, and I’m glad you bring this up, because I believe very strongly that when we on the manufacturing side and in the hearing care professional side, we always want to show that we’re getting better. And we absolutely are getting better, but we want to provide, I think, an analogy to automotive industry, where they provide city and highway miles. And I have no problem with manufacturers reporting their best scenario, their best case scenario of looking at a single noise in a specific location and comparing that to how a person would do unaided versus. In that. But that 13 decibels. Versus unaided. Right. And against diffuse. Yeah. Right. And then diffuse. And again, this is a very important point, is that we see, in many cases, apples to oranges, comparisons made between a manufacturer that, that recommends open dome tips or open fitted molds, versus those that want to ensure that on standard testing that their devices will perform better with directional microphones and noise management, you’re going to always get better performance when you’re not allowing sound to come in through the vents. So we’re very specific about the testing that we do in terms of closed domes and that that 13 is with closed domes in that situation against unaided. Okay, got it. Is there published data showing the 13 dB? We have internal studies right now. Internal studies. There will be published data that Michelle Hicks talked about in terms of their team. But the issue is, I don’t want to create the impression that in every situation, for every hearing loss, that you’re going to see that order of magnitude. But that’s where I think city and highway mile analogies of looking at comparing between Omni and directional or vented and unvented dome tips. And I’m really thinking that we need to be very deliberate about the types of comparisons, even some of the comparisons that are done the independent labs that are doing that, to always ensure that when they’re reporting specific findings or specific results, that they’re using the same setup, occluded or unoccluded setup for everyone, because then it’s a level playing field. Well, let me ask the question a little bit differently. Right. And that is across the generations between Genesis and Edge. How many more dB improvement with edge? I will have to look, because I’m not just going to give a swag here on this, but we compared both. What I’ll tell you is we’ve compared, when we have a level playing field between previous generation and competitors products that, regardless of the number, whether people are thrown around, 13 and ten and seven and three that we are comparing very favorably in our testing that we’re doing of our products and competitive products couple dB, in many situations, it’s going to always vary as Achin said in his talk today, when you use open dome tips, you’re going to see that benefit attenuated. What I’m getting at, I guess, is that the big difference between Genesis. And Edge, we will see, notice an Improvement, much more DNN. Right. The DNN capability is vastly improved in Edge. And so I would expect to see some dbs of improvement between Genesis and Edge. Yes. And the big thing is, as I tried to make the case today, is we know the four top drivers of expectations for hearing aid performance is sound quality, preventing loudness, discomfort, speech understanding in noise and spatial awareness. Oftentimes when people in particular are going for the best signal to noise ratio benefit, beam forming microphones notoriously provide the best signal to noise ratio benefit, but at the expense, in many cases, of spatial awareness. And so we want. And the other thing is, everything we learn from psychophysical research that has gone on in the past that are applied to machine learning algorithms is considering and building upon past experience. DNN models, rather than being rule based the way machine learning models are, they aren’t encumbered by those same biases that came inherent biased. As long as the training data is unbiased. Exactly. And that’s the tricky part. You got to get lots and lots and lots of different environments and then make sure that you’re testing different than your training days. Training database. It’s easy to get a good result if you’re using the same training database as your testing database. Yeah, right. It’d be like having all the answers to a multiple choice test and say, well, I got a perfect score. But you trained on those very same questions. It’s not. Yeah, no, that’s a great analogy. And the last feature we haven’t talked about at all is waterproof. Now you’ve got waterproof across the line. That’s right. And on all of the rechargeable products? On the rechargeable, that’s right. And so we’re really aiming to get IP 68, where there is a standard, and with the IEC standard and IP68 rating, the same is that maxes the score that you can get on an electronic device, whether it’s a phone or hearing aids. Now, meaning you can submerge our devices in a column of water a meter deep for 30 minutes. And the difference is there are various ways that people can test that to that IEC standard. They could take the receivers off and just test the body of the device and say, we passed for that. But who’s going to remember to take their receivers off and put just the body of the device when they jump in the pool. I’ve been sitting here laughing because I was there. And part of that when we started dropping receivers in columns of water at Knowles. Yeah. And that is often a vulnerable. Both the point of the connection point and the receiver itself. So we test the full array. And then the other thing that we do in addition the dirty little secret with the IEC standard is that it’s fresh water. And as you heard from Lou Ferrigno’s testimony his testimonial that he works out a lot and he perspires a lot when he does. And his concern is that people don’t sweat fresh water and salt water is highly corrosive and can be more damaging You could pass an IP 68 standard, but still fail in real world settings because of perspiration. Somebody that lives in a very hot climate and perspires and works out and is active. So we want to exceed or max out the standards where they exist. But then we build. And I’d love to have the opportunity to show you around a little bit of our quality labs of what we combine with our machine and our design team that is capable of building our own little chamber of horrors for hearing aids to evaluate where there isn’t a standard, using saline putting them in 100% humidity environment for days on end. And our whole goal is then to really do post mortem if we can test to failure, if we can get them to fail after many days in a 100% humidity environment and improve it. Why does Starkey see this robust level of waterproof as being important? Well, people live in environments where there’s dust, where there’s perspiration, where there’s precipitation, they’re real world. I’ve said for a long time the ear is a hostile work environment. And we want to be able to have patients have confidence when they use rechargeable batteries, that today and five years in the future that their hearing aids are going to last all day without having to have range anxiety. We want them to have confidence that when they put their hearing aids in and it’s pouring rain outside or they’re going to go golfing and they’re going to be in the elements, that they. I don’t want patients to have to think, ooh I’m going golfing today, and I need to hear my buddies when I’m golfing. But I’m afraid that there’s a 50% chance of rain. I don’t want to damage them. We want to give them the confidence to make them everyday proof. And the one other thing, I’ve done. It across all technology levels too. We have. Absolutely excellent. Absolutely. And then the one other thing that I was going to think of, people will say, well, how do I know when to change wax guards? And it’s a very common question that my patients will ask me. We developed a feature called self check. That is a dashboard that they can run anytime that is convenient for them. All they need to do is take their hearing aids out, send them on the counter, we don’t in their ears. And then it takes about 5 seconds and it will give a status check as to whether the microphones, the receivers and the circuit are all functioning properly. Well, that’s interesting. You’re actually running a feedback loop. We’re running a test that looks and evaluates the microphones and receivers and the circuit. And so even after putting them in water, I will take them out, dry them off a little bit. And if it says the microphone or the receiver is yellow, it’s green, yellow, red. Just simple like you would use any dashboard and dry them off a little bit more and then they’ll be up to snuff. But the most common application for that is patients will say, well, how often do I need to replace a wax guard? They said, when it’s necessary, receiver goes yellow, run it every day, and when it turns yellow, then it’s time to replace them. Some people can go months. Other people may have a week or so if they make a lot of wax. That is something that we even incorporate into an intelligent reminder that can be enabled by the professional or the patient to say every week, remind them to run the self check and to clean their instruments. And when that happens, if you start reminding them to use that, we went from a feature that many patients forgot about. And just repetition will start to foster that habit, that good habit of running their own dashboard so that they’re always ensuring that the hearing aid is running up to top speed, whether you’re out on the gulf fishing, whether you’re just wearing the devices on a day to day basis and you’re wondering, are all systems go the same? When you get in the car, you check that you’ve got a charge or enough gas and you wouldn’t dream of leaving without knowing how far your range was. We want to give that kind of convenience, flexibility, and seamless application for patients to have confidence in the quality and the reliability and the performance of our devices. Very interesting. I really like that feature. Yeah, I do too. Anything else you want to add about the performance of the Edge AI? It’s been a pretty complete discussion. No, I think we’ve had, as usual you know, you’re a very engaging host, and I appreciate the questions. Well, thank you very much and thanks for spending some time here in the midst of all the launch activities. My pleasure, Andrew. Thank you. Welcome. Achin and thanks for spending some time with me today. First off, congratulations on the launch of the Edge AI. It’s exciting. New development for Starkey, for sure. Thank you, Andrew. It’s great to talk to you again. I remember our meeting in Germany during EUHA last year. Yeah, that’s right. As we record this one week from today. Yes. Let’s talk about the DNN event. You had described how it’s 100 times more powerful than its predecessor, which is the Genesis AI. And in what terms is that 100% more powerful? Yeah. So, before I even get into DNN, today is a special day. This morning the Nobel committee awarded the Nobel Prize in chemistry to Demis Hassabis, who used deep neural network for protein folding problem. And yesterday, the physics Nobel Prize was given to Professor Jeffrey Hinton for some of his inventions of neural network work. So we are sort of like now on the cusp of this breakthrough, where it’s pretty clear that the best way to process sensory perceptual information, whether it is your computer vision system that allows you to drive the car autonomously better than humans do, or it’s your hearing aid that will have to do the difficult work of enhancing speech and suppressing noise, which has always been the goal for engineers working on hearing aids. But what we have uncovered, not just us, it seems like almost every player in the hearing industry now we have realized this deep neural, neural network based signal processing is the way to go. No surprise, because that is how, as I was explaining this morning, the human brain works. The reason you have such an amazing ability to understand speech and reduce noise, the cacophony of sound in your own brain, is because you have a deep neural network in your brain. So let’s just copy a little bit of that. So we had some success with Genesis with that. And it was pretty clear that the more deep neural network computation we do, the better results we’re going to get. So Edge AI, I’m going to withhold some of the architectural details. We are all competing with each other in the industry. I don’t want to make my competitors any smarter than they already are, but we wanted to take a very significant jump up in the deep neural network computation in Edge AI line of products, compared to Genesis line of products. Guess what? We’re not stopping at this. So there’s going to be even a bigger jump. So where we are, I feel very comfortable with the signal to noise ratio that we get, the amazing fast processing that we get, and more important, hopefully I get to explain this. What you’re hearing from our patients as a result of this technology, just like hearing more. So I’m gathering that the DNN is doing a couple of things. Okay. So, and it’s just kind of want to, at a high level, understand what the DNN is trained to do. So you’re doing speech enhancement in noise. Are you doing noise separation? Speech and noise separation. Are you also using it for your classical techniques? For example, the directional microphones with multi talkers, you’re trained to handle multi talker environments and so on as well. And then what’s been done for a long time, sound scene analysis and so on as well. Correct. Those are the three things that. So you are very knowledgeable about all of the ways that DNN could be used. Thank you for I don’t get a chance to, to talk about these details with too many people. So I’m, again, going to be a little cryptic, but you’ll appreciate the lay of the land here. I think the ways that you are discussing what all could be done with deep neural network, various different things like enhanced speech, reduce noise separate the sources of sound and split the cacophony into, classify them into different things. Yes, you’re completely right. Each one of them could be accomplished with deep neural networks. But where I’m going to be cryptic about is, I think that is not how we want to architect a sound processing system in a hearing aid. That would be, it’s inspired by the old and classic ways of processing sound. Various different stages. You do this, you do that. It reminds me of, let’s say, how we used to work on computer vision back in, prior to 2012, where we would work on first let’s detect the edges and then let’s use this algorithm to find out where there might be two eyes. And then what’s the signature difference between your two eyes placement compared to an average face. Classic algorithm that is separated by stages. Good thing was every stage was very explainable and understandable. Enter deep neural network. We threw all of that out, and here you are. You have a goal of accomplishing certain tasks, and then you have the data set carefully picked in order to train your deep neural network model. At the end, you have the result that comes out of it, you subtract from the intended result, you have an error. Use that error to retrain this by doing a backpropagation. The good thing is that it does an amazingly better job than the dual classical stage by stage computation. The bad thing about it, good luck explaining exactly what’s happening in the zillions of neurons inside the neural network. It’s like human brain. You can’t really go and pinpoint exactly what neuron is doing what. but we know the end result is this Deep neural network in your brain is helping you recognize face better than your spouse can, or enhance speech better than any older hearing aids would. That’s where I’m going to stop, and I’m not going to go into the details of how you architected it, but I, we would not want to have a serial approach, just replacing the older stages with deep neural network. Then I have to have very power hungry system. And the way we want to architect it is a smart system that is able to do more energy efficient computation. Yeah. And that’s a great explanation of how DNNs and the training works. And of course, a lot of resources. I have to acknowledge that a lot of the resources you put into us go into the training data set and the training process. Right, because garbage in, garbage out. Yeah. You need training data dataset, and also the architecture of the neural network as well. They are not all made the same. You would appreciate, again, from what’s out in the common place is convolutional neural network for recognizing face transformer neural network for doing a chat with the Internet. So they’re all neural nets, but they’re not all made the same. So we architected our very special architecture that, again, we have not publishing all the details of it, but it’s extremely energy efficient. You heard us maintaining 51 hours of battery life with always on deep neural network processing. Very different approach than some other products out there that are trying to do. Suffering from battery life issue. Bigger hearing aid because they put bigger batteries. We came up with a very clever DNN architecture that maintains the battery life. Small devices within increase the size at the same time, amazing signal to noise ratio enhancement. So you have all this going on, and then you have Edge mode +. Yeah. So Edge mode plus is a feature that I turn on specifically in a particular environment. So with all this DNN capability running continuously, why do I need what. You find is this. And I go back to the intuition coming from patients with identical audiogram. Yet some would complain a lot more about difficulty in understanding speech in a noisy environment. Yeah, some people, normal audiograms, complain about speech and noise, too. Yeah. And there is simply no way for us to probe it. You can go have conversations with your audiology professional, and they will do the best they can. Tune, they’re not with you all the time. So in specific environments where you’re having particular difficulty, you might want a very aggressive mode for speech enhancement and noise reduction, which you do not want on for most of the times. So, for example, I call this a slider bar. Ideally, if it was possible to easily have a slider bar that I am in a quiet environment. I’m just having a chat with Andrew one on one. Right now. I’m going to prioritize the quality of speech because there’s not much going on here. But the moment I walk into a bar with you, I’ll put the slider bar on the other side. I’ll say I will handle a little distortion in the speech, but I want to beef up the speech intelligibility improvement. Will be more than enough to make up for slight distortion in the speech. So for Edge mode. Have three modes there within Edge mode + Once you bring up edge mode, you can do that with tapping the button, double tap if you want, or through the app, you get three options. Modes affect the battery life where you’re still going to get more than all day. Our 51 hours is with DNN running all the time. Any Edge mode, mode. Edge mode +is doing the same computation. It just different sets of parameters to tweaking in the same neural nets. So it does not consume more power. Edge mode + does not consume more power than regular operation. Because the way we architected Edge AI devices, which we introduced today, is different from genesis and from what I read, different from other companies hearing aids. That deep neural network is in the main audio path. It’s always processing, always on. It’s not like you’re offloading certain, you classified certain environment offloaded to the DNN engine and bringing it back and combining again. We need to do that because it results in distortion, results in losing cues in the sound environment. We wanted it in the main audio path. So that’s the way that Edge AI is different. DNN always on. So even if you’re using Edge mode plus, it just invokes a different parameter set that’s informed by the acoustic snapshot that you’re going to take, but it doesn’t consume any more power. And does the Edge mode + train the system? In other words, if I go in a particular environment, like the social we were in last night, and I invoke Edge mode plus, and I get it, just so did it learn from me? And the next time I go in that environment, will it automatically go to that sort of set? Great feedback, and I would love to do that, but we don’t do it. Okay, so we wanted to, if you’re going to use Edge mode, we want to put that option in your fingertip that I want it now or I don’t want it now. So we want you to invoke it. But here’s the new thing, though, in the Edge mode plus, I don’t think I have all the time to explain this. In the first incarnation of edge mode, when you had it, you had to change manually. You have to invoke it if the acoustic environment changed. So right now your acoustic environment is very different from, again when you walk out and a lot of people then we require you to go ask for edge mode again. Now, if you want to initiate it, it’s automatically deciding if there has been a substantial change in the acoustic environment. It’s changing the snapshot. And hence you could call that a learning because it’s changing. It’s not like the static edge mode you used to have before and you have to invoke it again to calibrate. So I invoke Edge mode + in a bar. I walk out of the bar and quiet. It will revert. It will not revert back to the normal because you are still in Edge mode unless you want to get out of it and go to your personal memory. It will stay in Edge mode and it will constantly recalibrate, learn from the environment and decide what the parameter sets that’s needed for that environment. You could call that a learning. But that said, the neural network model that we have, which we train for certain environments and the parameters get influenced by the input, but then parameters themselves, the network architecture is pre trained. We don’t want to have a runaway issue. Yeah, right. Absolutely. But I wondered if you, and maybe even if it’s possible in the future. I’m curious about that. If I consistently go into a bar and I’m consistently selecting Edge mode plus in this mode, when does it start to recognize I’ve gone into a bar and just go there for me? Yeah. I can always write it, right? Yeah. As I said, I’m quite intrigued about it because what happens is sometimes I think I’m in the same acoustic environment because, hey, this is the bar here, the same barista, the people around me. It seems to me like it’s the same acoustic scene. Yet when you disconnect yourself as your human ability from it and all you have is the audio files recorded from it. Listen to it, something completely different. Completely different because of the directionality of the sound, the mixtures. So no two acoustic environments are the same. The best way to decipher is through a neural network because it’s able to deal with it. Okay. No, thanks for explaining that. The other thing I wanna talk about actually is connectivity because in the worst kept secret ever I knew Starkey was working on LE audio for a long time. Right. And you obviously had a lot to do with that. My old boss told it off. Yeah. Right. So what was it a year and a half ago now where Pat Gelsinger demonstrated a Starkey hearing aid connected to a PC? And so now you actually have LE Audio and Auracast capability. What does that implementation look like today and how do you roll it out in the future? In the Edge AI. Yeah. So first, Pat is a great mentor for me. He’s still the biggest critic of hearing devices. Because he himself uses hearing aids. So when he uses our latest and greatest products he’s power user for my product. So back then we obviously had the devices already working. We showed a live demo on the stage streaming from Windows laptops but we wanted to take it to a place where we are proud launching it as a product. I don’t want to get into all the gory details but working with Intel, AMD, Microsoft we are still fixing those low level bugs between us. Some of them were to do with Windows and they work diligently on rolling out fixes to the point that we felt the system’s robust now that it is ready for consumer. We didn’t just want to go and launch something and be forced to market with the technology when it’s still. The edges are still rough. Right? Yeah, no, I can appreciate that. And I’ll tell you, I’ve got my two year old PC that’s in that bag right over there. It runs LE audio. Nice. It doesn’t have the hearing aid control features because I’m waiting with beta for the next Microsoft update to be pushed to me. Right. But LE audio on the PC is brilliant and I mean really le audio connectivity in Auracast with all the different things that people do today is very much the future. So if I walked into a place that was broadcasting Auracast like let’s say next week, at EUHA where there’s going to be 38 Auracast transmitters going would I be able to take the Edge AI and tune Auracast? I think for auracast you still need to wait a little bit for this random connections because we are collectively not just us with the big ecosystem players for consumer electronic devices we’re still deciding on the UIs that are going to be but I think the first thing that you’ll notice is like the proliferation of unicast one to one on one bi-directional audio streaming. Because for business users being able to stream from your PC in a noisy airport bi directional audio from a Zoom call on your PC to your hearing aids is going to be great. They would love to have that menu that will come up and say I know how to select gate 73 to my hearing aids. It’s in the works. We feel very comfortable with the spec but I think it’s a wave of adoption that you’ll have to see. But in the meantime you’ll see proprietary devices like your tv streamer that you heard about today. 50ft range, amazingly better sound quality with LC3 codec, smaller designs, lower power. So I think this is going to have years of rollout. Apple still hasn’t rolled out their early audio version in the phone, so obviously they will do that. I’ll let them announce when they do that. But we are going to see a wave of new devices come to market. But instantly, we feel like there is significant benefit now to have this out. We’re not taking out anything, though, like in all of the this radio that we have in there. In a way, it’s unique, is that it’s backward compatible with all the hard work. Okay. So you can still run an ASHA mode or MFi. MFi too. And you can run in LE audio mode. And here’s the thing, you know, some of the patients I. These days, I fit patients because even as an engineer, I got licensed in California to fit people with hearing aids. I wanted to really get into the details of understanding patient feedback and all of that stuff. Some of the patients tell me, I didn’t think Apple has released LE audio in the phones yet, because the new radio that we have in here, even with MFi and Leah two, you see twice the range and much better sound quality because we also took up opportunity to re architect. I was explaining this morning about how the streaming sound is processed. It has a separate engine now compared to all the other environmental sound that gets processed. And the way we provide the streaming sound differently has a fuller, better experience, even for Lea two with MFI, just a better sound quality so people think did you do something with LE here? No, not for that product. I do a lot of streaming of various kinds. I’m really looking forward to. I’d love for you to because streaming music quality varies widely amongst differ hearing devices. And perhaps you’ll appreciate this. Youd never create the boomy, low frequency bass sound. Also, you’re not occluding your ear. I hate it when my ears are occluded. I love my AirPods pro 2 when I’m in a plane canceling noise. But as soon as the plane landed I take it out. So you don’t want to occlude people’s ears, and as a result, you need to, you know, it’s for always listening. It’s not your replacement for Bose over the headset or airpods pro 2’s occluded, fit for music, but it should be good enough that for most cases, you don’t have to take it out. Yeah, that’s the thing. It should be good enough. It should be good enough. I’m very much looking forward to trying that. And I was fascinated that you built a voice assistant into it. I’ve been following the voice world since the very beginning. You’ve done mobile voice now, and I’m fascinated by that. How do you see that actually being adopted by people? Take me through it. Will everybody ignore it at first? Do you think it’ll catch on? Will it add value to a hearing aid user? Tell me how you feel. Yeah, my view again, I’m like you, very enthusiastic about this future. As long as the technology is good enough, we would do it now. Suddenly it will not happen overnight. It’s like tomorrow. Suddenly we’re not going to go out and just start talking to our devices, but we’re going to realize, wow, this is really good enough. I gave you three examples this morning. You might want to just change the device and no, I don’t want to bring out my app right now. Increase volume and you should be able to do it. Set a reminder for my I need to pick up my daughter at 06:00 p.m. it just does it. Or, hey, I’m thinking of where I go to Europe for my vacation. Thinking Italy or friends. What do you think? Yeah, it’ll be really interesting to see how people actually adopt and use the voice feature going forward. Yeah. So, listen, I really appreciate you spending some time with me. It’s great to dive into more of the details on how the DNN is working and the advantages that it brings to people. Thats great talking to you, Andrew. You’re so knowledgeable in this ecosystem and about devices and where they could go. I really enjoyed our conversation. Thank you very much.