The Convergence of AI, Audio, and Hearing Devices

All right, everybody, and welcome to another episode of This Week in Hearing. Today, I’m joined by Aernout and Boltzmann two guys with very interesting companies that we’re gonna get into today. So without further ado, let’s do some introductions. We’ll start with you, Boltzmann. You wanna share a little bit about yourself and your company, Aizip? Yes. So my name is Zengyi Li. You can also call me Boltzmann for convenience. So I graduated from UC Berkeley PhD in 2022. And we actually started Aizip a little bit before that, in 2020 during the pandemic. So AIZIP is a company that build tiny AI models for on-device usage. So the topic we have today is mostly around hearing aids. So we are making the AI model that runs on hearing aid in AAL’s framework. Awesome. Thanks for being here, Boltzmann. How about you, Aernout? Yeah. My name is Aernout Arends I’m the founder and CEO of Absolute Audio Labs. I am a product designer, but I’ve been struggling all my life finding the right balance between playing music live on stage across the globe and still having a professional career. That changed when I when I damaged my hearing and I got hearing aids, and they worked well for me but I was really disappointed with the performance on music. And since music was such a big part of my life I went looking for a solution and when I couldn’t found it, find it I actually developed it myself. Now we’re talking about 2012, so that’s quite some time ago. I founded the company in 2016 because I found a solution that really works well for enjoying music again when you have hearing impairment much like myself. We brought that to market and throughout the years, the team grew and I added a great development team that was having their own product line on speech intelligibility. And these guys joined me in 2018, and since then we’ve been expanding all these algorithms for music, for speech intelligibility, and whatnot destined for hearing-impaired people to really enjoy music but also get some great audio. There you go. That’s awesome. Well, cool. So I wanted to have you two on today and I kind of want to frame this conversation for the audience because I think that it’s, it’s a really interesting time right now that we’re in, something that’s kind near and dear to my heart because this is a topic that, you know, I’ve kind of been covering for going on about 10 years, probably right around 2016, 2015, when I was really starting to, you know, blog about, you know, the, the innovation that’s happening in these ear-worn devices, right? And one of the kind of macro trends that has been top of mind for me is sort of the, call it the ‘peace dividends of the smartphone war’ to, to coin, you know, Chris Anderson’s term, but it’s the proliferation of all of the components within the smartphone ecosystem that have proliferated so massively into every consumer electronic that we have today, and then it’s the innovation that’s sort of now underway in cascading sort of throughout that supply chain. And so one of those is, you know, the chip level itself. And I think that if you’ve kind of been following what’s going on with smartphones and just with all of these different, you know, pieces of technology, you’ve got these new architectures of how these chips are constructed, right? I mean, it wasn’t that long ago that the system on the chip was sort of introduced, and now you’ve got all of these very intricate complex ways that you can layer on different types of hardware within the chip that can support different kinds of functionality. And, you know, I think for, in plain English, like what this is all kind of amounting to is kind of a a commoditization and a consumerization of just about every kind of application, it seems like that had maybe previously been restricted and relegated to very hyper-specific use cases. And so the hearing aid’s a really good example of that where the, the hearing aid is predominantly constructed with the ASIC chip, right? And these ASIC chips are really, really good at one application, and the downside of that is then you sort of have to engineer for that specific application, and you can engineer in a lot of ways and, and create a lot of really cool things for that. But from an economic standpoint, it, it really kind of creates a bottleneck from a supply chain, you know, standpoint where, you know, you, you really only have a fixed amount of supply based upon the companies that are willing to put the capital investment toward the ASIC chips that are required, that are the standard for this. And so what’s kind of happened over the last I would say 10 years or going all the way back into the introduction of the iPhone and the smartphone and everything even that preceded that, is this microtization of Moore’s Law and, you know, everything’s getting faster and getting smaller and microtized and all that. And so here we are in 2026 where I think you two kind of, this, this partnership that you’ve created, and I’d like to kind of segue into that, is a really good encapsulation of this, which is this whole platform that supports, you know, AAL’s platform that can support multiple plug-ins, multiple different applications that can feed into then these different consumer-grade chips and provid… features and capabilities that, again, were previously limited to these hyper-ly, you know, sort of designed products in the, in the ASIC era. So you’ve got an era now where you can enable, I think, a, a much broader supply through some of these interesting integrations at kind of that consumer-grade level. And for me, like that, when we talk about OTC hearing aids and we talk about, you know, kind of this, like, new dynamic of the non-prescription hearing aid world and everything else that falls into this market, this market’s so nascent over here and it, it seems like we’re kind of at the cusp of this beginning of this mass proliferation of consumer devices that are far more capable than devices even a few years ago. So I’ll I’ll pull up there and I’ll pass it over to you. All right. No. I mean, I mean, you’re spot on. There is a lot of convergence going on in the market, and it’s been going on for some time. And you see the hearing aid features being added to AirPods. You see the, the hearing aids itself trying to provide proper audio streaming and other connectivity and features. You know, they, they, they, it’s, they come from a time where the ASIC was necessary in order to provide the battery life, and that’s been a key feature. And why do I mention that? Because over the past couple of years, hearing enhancement has been possible on audio chips, the generic ones that you’ve been talking about. But what we’re seeing right now is we’re sort of at the sweet spot, at the crossing point, where your battery life actually enables you to run full-blown hearing aid for a full day on a generic audio chip, and we’re proving that we can do that. And it, it takes a lot more than just a chip, chip of course, but it also brings in a lot of added benefits that the audio chips bring: easy connectivity, ANC proper audio support, streaming, whatnot, all of that is available while at the same time you can run a full day on the, on, on one single charge. So yeah. That’s, we’re definitely at, at breakpoint in time where this is now becoming possible, and the way we’ve been indeed is in a platform where we have all our algorithms running so you can actually run a, a full-blown hearing aid but you can also use it just for hearing support or OTC devices or on TWS as, as a hearing feature. And, and we’ve opened that for, for third parties from our perspective. So with, with the focus on getting innovation to market, I mean, there’s now all of these opportunities are opening up, like you mentioned. Now, how do we get these to market? I mean, we have our own algorithms on hearing but want to accelerate that, and companies like izzet, they have great AI noise suppression that we want to enable. So we’ve opened it up and, and we’ve started this partnership so we can help them go to market quickly on this platform in a proper full-blown product. Right. And I’m gonna kick it over to you here, Boltzmann, but I just wanna kind of dovetail on that and say, you know, I think that’s such an important point, right? We’re talking about now platforms, like almost as if these are app stores that are, you’re able to plug in something like Boltzmann’s technology. And again, in our, in the world of hearing healthcare, the holy grail for a long time has been a really good hearing aid that can really, you know, delineate noise from speech and, and all of the different speech and noise type features that have been sought. And again, it’s like, I think that there was one trajectory where it was like you kind of keep doubling down on the existing technology and then you’ve got this, you know, this fork in the road that’s sort of presented itself that now you’ve got this other track where you can sort of use the more platform ecosystem plugin, you know, type concept which I think is, is just fascinating. So Boltzmann, why don’t you share a little bit about your technology and how it works from a, you know, like the de-noising and all that? Sure. Yeah. Maybe let me explain the, the difference between AI de-noising versus traditional de-noising first, right? So the goal here is, roughly speaking, separating speech from noise. of course, you don’t want to suppress all the non-speech signal because some of them are important for you and your safety. But traditionally people rely on very simple assumptions, basically handwritten rules to distinguish between speech and noise. For example, if if something doesn’t change over time, you, maybe most likely it’s noise. If something is fluctuating over time, it’s speech. So based on those simple rules, you can remove stationary noise pretty well. But if you have things like hand clapping, you have water flowing, those kind of non-stationary noise traditional algorithm really struggles a lot. So what AI or deep learning noise reduction brings to the table is that it’s trained on data. So internally, we have you know, thousands of hours of huge data sets that covers all the different type of languages, human speech all different speaking styles, as well as all the different kind of noise you will … in the real world, as long as it’s not, not a human speech. Right? And then our model is basically optimized over this entire distribution of noise and speech, so it can handle almost any non-speech noise stationary or non-stationary. Yeah. So if you really test it you will find a big difference between the AI denoising versus traditional denoising for non-stationary noise cases. So that’s the basic of speech denoising. Just want to mention a few things about our partnership with AAL and why this platform is important to us. So Aizip is actually a horizontal company. We have a huge number of applications but we cannot go super deep on any of them. Right? So for hearing aid, you have, besides noise suppression, you have a lot of other functions like DRC, you have EQ, you need to match the audiograms, you have connectivity, all, the things like that. And we are not the expert to develop all of those. Right? So the partnership with AAL really fits our need because we need a, a way to go into the market without doing all this extra work that we’re not good at. And AAL’s platform is perfect for us because they have everything else and they just need one you know, AI denoising function, which we provide. Yeah. So, yeah. Yeah. It’s been a very good partnership and very valuable to us as well. Yeah. Yeah, indeed. And maybe it’s worth mentioning that we, we did our first demo together a couple of months ago at a trade show. And we had a setup in a, in a very noisy environment with traditional hearing aid running on our platform and the Aizip plugin, and you could clearly hear the difference that as soon as you kicked in the Aizip plugin that, you know, it went completely calm. You could easily have a conversation. So it really showed in a real life environment, environment running at ear level with real speakers you could really yeah, experience the benefit of, of the, the Aizip plugin compared to all the traditional algorithms. And so, just so I understand, would the would the premise here be that if I’m a company that’s sort of looking to, you know, produce my own hearing aid retail my own hearing aid I would, I… You know, sort of the sourcing process, I would, I would basically choose AAL and as, as the platform that my chip is gonna run on? Is that how this would work? And then I’m just trying to understand the commercialization of this, right? Because I think it’s a fascinating sort of model touching on Boltzmann’s point where it’s like, you know, you’ve got these these horizontal companies that they don’t have the incentive to drill down deeper but it would be perfect if they could sort of extend that capability to a company like yours that’s sort of this very specific platform for this application, right? So I think this makes sense of like the ecosystem of it. Is that how that would work though from like just a Mm-hmm. … standpoint? Oh, absolutely. I mean you’re spot on. The, the, the reason why we chose this approach is really because we want to offer this kind of standardization in order to make it work with partners like Aizip, but also make it work for the hearing aid companies who, want to run our algorithms. So we have implemented the platform on various chipsets and that, that brings a series of advantages that bring to the, the hearing aid manufacturer, they can really choose the AAL platform, which is called Pure Audio and then decide on a chip to run it on. And if throughout the process they want to run through or, or on the different devices, wanna run with different chips, for us it doesn’t make a difference. But also for them, it doesn’t make a difference because the interface for them is the same. It connects to the same fitting engine, it connects to the same retail heart software. So that kind of standardization layer really is the, the, yeah, the connecting point of, of the suppliers like Aizip, but also to the hearing aid manufacturers who want to decide on how they build their, their devices. That makes total sense. So I guess one question I would have, you know, and this would be more so speaking to the people that are like Boltzmann that are developing applications that might be appropriate for something like this, you know, what’s the pitch for why they should work with AAL and is… and vice versa? I Mm-hmm. is there any reason why AAL wouldn’t sort of create these applications on their own? So can you walk me through that? Oh, yeah. I mean it’s, it’s an easy pitch from our perspective. You know, AAL is, is the platform that covers different chipsets. There is no other platform that, that does that and has the openness so they can implement it. So the way in order to reach the market, the different hearing aid manufacturers at different levels is through the AAL platform. That’s by far the quickest way. And the standardization is to the interest of both the algorithm providers and the chip manufacturers and the retail. So if at all levels everybody wants that standardization and we offer that standardization, that’s why you come to us. That makes a lot of sense. Anything you want to add there, Boltzmann? Yeah. I just want to mention that so TinyML, development is not a easy process. So it took us quite a some time to figure out how to make a model so small and make it still perform well and how to, you know-… deploy them to a real world, a tiny device, and still run fast enough. So that’s why we are a horizontal company. We have a same tool set that solve the same problem across many, many applications. Right? But, for each application, we just provide that one core function that enables the AI and you know, not the whole solution. So, And let me ask we have seen a lot Yeah, sorry. Let me just add one thing. Yeah. Go for it. So we have seen a lot of customer who try to develop model themselves. Even you know, much bigger company, billion-dollar company they have a team of two, three people, they still cannot do it. So that’s why Aizip can still, you know stay in the market until today. And because we have our unique solution. That makes total sense. I was just gonna say, you know, as somebody that’s working in the heart of AI right now does it feel as if things are accelerating, you know, in, in terms of what you can do? Is there that sense? I mean, it seems like it from afar, but I’m just curious within this specific domain of AI, what’s, what’s today feel like versus maybe two years ago? Yeah. Certainly the computation power on the… You know, we, we all know that computation on the cloud is, is exponentially expanding, and everybody’s buying GPUs and stuff. But computation on the edge is also expanding exponentially in a slightly different way because all these new ASICs, well, we don’t call it ASIC, but SOC’s, being designed today all have NPU capability. So they don’t run at a very high clock, but at the same clock, they, they are like tens times or a hundred times more powerful than chips only a couple years ago. Wow. Yeah, it’s a very useful thing for us as well, because for application, especially things like noise reduction, the performance really scales with the compute rather than the, the model size. So with more compute, we get you know, better and better separation between speech and non-speech, for example, and is a benefit for the consumer eventually. Yeah. Now, I can only fully subscribe to that. Our experiences is that a couple of years back, it was impossible to what we’re seeing today, and Aizip has been a great support in really getting the right application for hearing aids going because noise suppression, as such, you need the technology, you need the understanding like Aizip has. But application on different yeah, different applications actually mandates that you, you get the right level. And for hearing aids, you don’t wanna cancel out everything. You really want the right balance between speech intelligibility and still be in an acoustic contact with your environment. And yeah, they really, they really got to the bottom of that and really helped us to the, to the best level that that’s currently possible. And I’m sure that going forward, it will only improve. Yeah. No, I think it’s super fascinating. So as we kind of wrap up here, I mean, closing thoughts on you know, kind of what 2026 will look … again, for the viewers, you know, how, how can they sort of expect this space to increasingly become more and more relevant to them? Yeah. I mean, you know, restoring music for hearing-impaired, that’s really what got me going. But as we added speech, and, and looking at what’s happening now, it’s, it’s really very exciting times. We’re seeing a lot of traction, a lot of companies actually starting to standardize on the platform, which means that in 2026 and going forward, we will see products hitting the market that actually yeah, bring the performance that we want, and not just into the Western world, but also across the globe because the price point is much more interesting. So also markets that are currently not being addressed actually can have access to hearing support. Very cool. Yeah. Just wanna mention that during CES, I, I was walking around and I was, you know, asking those hearing aid company who claim to have AI, you know, “What… Do you have AI denoising in there?” And many of them are being honest and told me that, “Oh, actually it’s not truly AI denoising yet.” And they’re still concerned about the power consumption because they’re using an older generation of chips et cetera. So know, from what I saw the market is wide open and all of them are actually pretty interested in engaging with us or a partner like AAL, to work together on bringing true AI denoising to their devices. And of course, I’m talking about the, the OTC manufacturers not the big five. Right? So I think it’s 2026 could be the year where most of the OTC actually put those function if not in, in their real product, but at least in their next design. Yeah. So excited to on this. Very cool. That’s awesome. Well, Aernout, Boltzmann, thank you so much for coming on today. It’s been a pleasure talking to you, too, getting to know a little bit more about your business, about this partnership between your two companies. I think it’s gonna be a really exciting space to watch here in 2026 and beyond. So thanks for everybody who tuned in here to the today. We’ll talk with you next time. Cheers.