¶ Gig Gab 524 - Monday, March 9th, 2026
Gig Gab, episode 524 for Monday, March 9th, National Meatball Day, 2026.
Greetings folks and welcome to gig gab the show by for and about working musicians our sponsors today include squarespace.com slash gig gab where you can go and save 10 off your first purchase of a website or domain using code gig gab and then also claude.ai slash gig gab where you can sign up for claude today it includes claude co-work which is a great thing uh we'll talk more about both of those in a little bit for now here in durham new hampshire i'm
dave hamilton and today my guest co-host uh is the lead engineer and founder of alpha sound and alpha alpha labs they are the creators of the d feedback plugin that we have recently talked about a few times um our guest co-host has been a pianist for decades i think uh and is involved with the salem orchestra he's equally at home behind a console behind a keyboard or behind uh managing these crazy products that they are creating he's building these tools for musicians while also running real
world production so you get to see firsthand how things break down between the stage the business and the tech we're going to dig into all of that devin sheets thank you for taking your time with gig gab today hello
Hello nice to talk to you.
It's nice to talk to you too yeah uh this is you know we we talked about d feedback on the show and it it's obviously been making great waves in the audio industry. Uh, and so I, when we got to meet it at NAMM and you demoed it for me, which I also talked about on the show, I, it just, it just works. There's not, you know, there's not a whole lot of fanfare with this.
So I, I would, I want to dig into your whole story, but now that we're kind of on the D feedback thing here, how did that come about?
¶ Let’s Grow this Legacy Family Business
What, like, what, what, what was the evolution of this?
Well, the long story short is that, I mean, so I grew up in a family business doing sound for concerts and events and doing installations. So Alpha Sound is, you know, legacy family business based in Salem, Oregon. And so that was just kind of part of my milieu, like just doing sound. But I was also a musician, and I went to college for music at Azusa Pacific
University in Southern California. moved back for a while and actually went and lived in Europe for a while, then came back and said, let's grow this family business. And so we were a big Nexo Yamaha house. We had Nexo STM system. And so it was around 2016, 17 that we really started to do, we really kicked it into of high gear and we were doing super high-end corporate work. We started doing large-scale installations and large Catholic churches and stuff like this.
I think around 2019 or so, I just, I went to NAMM for the first time that year. And I remember walking around, looking at everything, just going, it's all so amazing. Like everything is so amazing.
Yes. It's overwhelming. It's such cool stuff. Yeah.
Yeah. There's like every, everyone, everything's coming out. It's just incredible. We can do all these cool things. And yet, I don't know, it just, it struck me.
¶ We haven’t “just solved” this feedback problem
The most basic, most fundamental problem with live sound systems, like that you get a speaker and a mic in the same room, and what can happen? They can feed back. I realize we haven't just solved this problem. There's no simple, elegant, turnkey way of solving that problem. There's all these interesting attempts. There's lots of automated EQ-based systems or gating threshold, intelligent gates and stuff like this. And everyone's got their bag of tricks. I certainly had mine.
I had all the standard tools that you would use to try to get gain before feedback. But I just – PSE works great.
Right?
Oh, everything. Yeah. 50-45, you know.
Oh, yeah, right.
Whatever, you know, waves. Now they've got like the – they had X Feedback and there's Feedback Hunter. And they're all – they can work, you know what I mean?
Sure. They're better than nothing. Way better than nothing. Yeah.
Yeah, it's not like good engineers aren't capable of getting gain-before-feedback, even in tough situations. I certainly was able to. Right. But it's just... I thought, why isn't there like the balanced audio type of solution for,
¶ Why isn’t there a “balanced audio”-type solution for Feedback
like that fixed hums and buzzes just because by its nature, it's this simple thing that is now everywhere, right? It's in every console, every piece of equipment is just uses balanced audio. But imagine at the time when it came out, like, you know, when balanced audio was invented, people must have just thought, this is insane. How is this possible? It just by its nature just gets rid of the possibility of hums and buzzes and so.
Yeah and and i think it's important because i i'm seeing where this is going here but i think it's important to take a minute and talk about how balanced audio and things like it work they're doing pattern matching machine learning neural networks really what it's what we currently call ai right but but you know for years we called this machine learning where you would feed an engine Many, many examples of a thing and then have it and then you and then the the
the very powerful computers do the crunching and find the patterns. Right. And so you feed these very powerful computers, you know, hundreds of thousands of examples of in this case, you know, buzzing and hum and they find the patterns in them. And then they can distill that down into something that can be used very efficiently. And that's a that's the the neural network there and it just pattern matches and finds it and can eliminate it with very low cpu overhead and very low latency right
Well so that's that would be an ai approach to getting rid of hums and buzzes sure um but obviously just like even earlier than that like going back you know like just the literal i mean just literally balanced audio itself, the fact that you just take the two signals and you have an inverted one that carries down the same line. And that itself is just a simple way of reducing hums and buzzes.
But what you're getting at, I think, is the principle behind it, which is you're taking the one signal, and in the case of balanced audio traditionally, you're creating an inverted copy of it to travel down the same line. And then they both pick up whatever noise and hum and buzz they will in common. And then when they arrive at the end, you just take and you flip the other one
back in phase, in phase. It was out of phase, but you flip it, combine them together, and then the intended part of the signal sums together, right? It's maintained, and then all the hums and buzzes cancel each other out because of the phase flip that happens.
I thought you were talking about plugins. I'm glad you explained that.
No, but exactly. That's how you do it. Let's say you were to make – let's say – let's create an alternate universe where they had never invented analog methodology for hums and buzzes like balanced audio. And then someone came along and wanted to make a plug-in using AI to finally do it. Let's say that's the universe we're living in. The way you described it is exactly how you would do it.
And we do have that for like wind noise and things like that.
Yeah, yeah, yeah. This exists.
Oh, it for sure exists. I just misunderstood you. And I'm glad you corrected me. That's amazing. Yeah, yeah.
But it actually sets things up nicely because in a sense, that is what we're
¶ NAMM is a great inspiration…and it inspired Devin and his team to seek a feedback plugin solution
trying to do with D-feedback. We're trying to basically figure out how can we create a polarity-inverted copy of whatever we want to get rid of in the signal in real time. This may be jumping ahead a little bit. No, we're here. We're not jumping anywhere. Yeah, yeah. But back to NAMM, though, this is exactly what I thought. I thought, why hasn't anybody sort of come out with the...
Not only just in terms of the simplicity and the sophistication of that simply elegant solution for hums and buzzes in the analog sense, why haven't we figured that out for feedback? Even in an analog sense, forget digital, forget AI, forget neural nets, how come no one's invented a way to just get rid of feedback in that simple of a manner? And so that was where I started. COVID hit, and then, of course, everyone's just sitting around doing nothing.
And so I actually took the time to contact people that worked in R&D departments at these companies that I have ends with because I either sell the brands. And I would pepper these guys with questions like, hey, why aren't we doing this? Come on, you're a big company. You've got all this money. Why don't you fix this problem? and I've got some ideas and whatever. And they're like, hey, kid, it's harder than you think. Otherwise, it would have been done, right?
Yeah, but like that, what you're describing in this moment here is what happens now Many times, I don't want to say every time, but it happens often when someone has a business idea and asks, why hasn't anyone else done this before or just proposes it and sort of puts the idea out there that this hasn't been done before. I think I can do it right or I think it can be done. And immediately you get people saying to you that will be impossible.
You won't be able to do it. What they're really saying is, I believe it's impossible and therefore I'm not going to do it. And it's hard to separate those two things, but I'm glad you did.
Well, and I also just think that people have been so entrenched in the industry for so long in their ways of thinking about what feedback is and how to get rid of it. Whether it's EQ-based methodology or whether it's intelligent gating or whether it's – there's all these different things to try to do. And I just think what happened is that nobody – I may not be the first to think to do this, but it seems to be the case that I may have been the first to productize it.
To try to say, why don't we use the sort of inverse, in this case, inverse IR methodology to try to stop the potential of feedback. Now, these types of systems do exist to some degree already in telecommunications, but they're different. And also, they're not in any way low latency. You can't use them for live sound, and they have a lot of artifacts. They kind of like destroy the sound. You know, your typical like, you know, like video conferencing type of noise
reduction and echo cancellation and feedback reduction type of techniques. They sound terrible.
They do sound terrible. Yeah, yeah, yeah.
Yeah, but the idea of trying to use that method, and again, basically, this method is you're trying to construct an impulse response on the fly, and we use a neural net to do that. Got it.
Okay.
Just like, for example, if you have – Yeah.
You give an example. Actually, let me give an example because it's relevant. I want to hear you say this. I am currently, you are in a hotel room, and I am using a plug-in, a Waves plug-in, called NS1, their noise suppressor, simply to get rid of the noise of what I presume is your hotel fan, like the air conditioning fan or whatever it is in the room that I can hear in the background.
¶ Training the AI to listen for three things: human voice, reverb, and feedback
And NS1 does a wonderful job of doing this, and it's doing it in a similar way to what you're describing. So I'm going to turn it off for a second and let you talk and, and for as long as I can stand it. And then I'm going to turn it back on because I don't like to hear all that wind noise. But so now it's off, I believe. Yeah, now it's off. And if you're listening on headphones, you'll probably hear Devin with some fan noise in the background. But go ahead and explain, please, your example.
Yeah well you know it's it's it's um it'd be like if you had um an eq based system where you know you you can make all these eq dips to try to you know ring out the the the feedback and and get rid of it um and then you were to have an ai grab a hold of all the knobs and do it for you and it could do it super fast and you know grab way more knobs than you could possibly grab as human being, even if you were highly experienced and that's all you were doing.
Basically what we're doing is we're creating a series of parameters to create an impulse response. And then what we do is we give the AI access to those parameters and we tell it to listen. We've trained it to listen for the human voice and the voice.
Reverb and feedback that's really the unique part of what we've done is we've we've you know taken what would have been a classical dereverberation algorithm because those already exist and we basically tweaked it to actually think about getting rid of audio feedback and it's it's um it's basically what it's doing is it's listening to micro cues in the audio signal that a sufficiently. Advanced audio engineer listening to the same thing would probably think in their head or at a gut level.
If you've got a microphone and you're turning it up and you can just tell, it's like a lav out on a stage and you can just tell, if I go a little bit further here, it's about to ring. I mean, it's not ringing right now, but you just know it. You know it.
Oh, you definitely know it.
Because you've done 20,000 shows and you just, it's like, you can't even put your finger on it. You couldn't really even technically explain it, but you just know it. You are certain. Because you hear it.
And you are correct. Yeah.
And that's exactly what the AI is trained to listen for, is those micro-queues. It basically runs a probability calculation, and it says... If we left this alone, it is a certain probability that in the future, this will turn into feedback. And if that probability threshold gets crossed in various ways, then it takes action now to bake an inverted copy of those portions of the signal into the IR that it's constructing in real time.
And then in real time at zero latency, no added latency, it's just zipping that IR together with the incoming audio from the microphone. So it's just canceling out anything that you can target, anything it can figure out how to identify and fit into that IR is just canceled on the spot in real time.
You're doing this, you're essentially using the past, the very recent past, but the past as it were, to come up with the impulse response. And I want to talk about what an impulse response is in a minute, but you're coming up with that impulse response and feeding that in alongside.
¶ Truly zero latency for the plugin
That's how you're doing this with no latency is you are using the past but not delaying the audio in the process. You're just you're coming up with this. OK, I think this is the magic answer for what's about to happen. And then you feed it in, you know, you slip it in the stream real time and it tunes it out.
Yeah. And, you know, how this came about kind of makes sense.
OK, I was just thank you, because, you know, I I've heard about this. I've seen it in real time, but it wasn't until this moment that I understood everybody says zero latency. And a lot of times they mean, you know, single digit milliseconds. Right. You know, right. And and you you are truly doing this with zero latency for the plug in your workflow to get to and from the plug in may or may not add some latency, depending on how you're integrating it.
But that's just natural with any plugin in workflows like that. You're adding some latency that's there. But the plugin itself is operating without latency because it's essentially doing two things in parallel. I know it's doing way more than two things in parallel, but it's doing two things in parallel. One, it's letting the audio through. It's just adding this little thing based on the past. I get it now.
Thank you. Yeah, and it could, I mean, all it has to work with is the buffer size of the DAW. So when you set your buffer to, whatever, 32 samples or 64 samples or whatever, it's fine with that. It doesn't need to add any more than that. So if you don't have the plug-in inserted or you do have the plug-in inserted, it's not going to change the overall latency of the audio signal. It's just using the buffer size. And our program happens to be buffer-independent anyway.
So, I mean, if your CPU is fast enough, it could operate on a buffer size of one. It doesn't have any minimum buffer size that... There's the DAW buffer size, and then there's the on and off traffic for the I.O. Device. And so between all of that, you're going to have some amount of latency. There's no hardware that's zero latency. That's not possible. You have some amount of, I mean, even analog hardware takes literally some time for the electricity to flow in and out of the circuit.
Less than with digital, for sure.
Yeah, of course, the speed of light in that case.
Yeah, yeah, yeah. But it is, yeah, yeah.
The point is that our plugin does not add any additional latency, and that's another very, very unique part of what we're doing. There's other plugins that, for example, on the dereverb front, there's other dereverbs out there. And our plugin is a pretty good dereverb. If that's all you wanted to use it for and you didn't care about the feedback, let's say it was like broadcast or a podcast, you only want the dereverb. Ours actually works pretty well.
I mean, that's not like what we put it out there for, but it is a part of what it does. And it's actually better than most other D reverbs on the market, we think. And it runs at zero latency. Most other D reverbs are... 30 milliseconds 40 milliseconds plus some of them are a lot higher than.
That i can't i can't run a d reverb plugin on my setup here that that works well i mean ns1 the one that i'm using uh it's very low latency and it does a decent job of picking of picking out the reverb of your hotel room but it's not perfect right i mean not even close um unfortunately at the moment anyway i can't run i would need external hardware i think to run uh d feedback on my
¶ I don’t have any coding or AI background, but I have a gut feeling AI will fix this feedback problem
setup here because i'm on a mac and and i i'm understanding that your plugin isn't quite ready to run real time on a mac you need to run it on outboard gear is that right that's
A longer story but.
Yeah for sure yeah but
But yeah i'll here i'll mention that in a second there's some other things to be said there. But just to complete the story, so to go back, so it's 2019. I'm at NAMM going, everything's amazing. Why aren't we doing this? Through COVID, I was asking people, why aren't we doing this? And they're like, it's harder than you think. So I basically at some point just said, I have no idea what I'm doing. I don't have a background in AI. I don't have a background in coding.
I barely know how to run a computer. I've been a Mac laptop guy my whole life. I've never even touched a Windows computer before. And so that was me. But I'm a sound engineer, I have this gut feeling that AI, whatever this means, is going to fix this problem. And I had started to mess around with things like ChatGPT, and I was like, okay, AI is for real, and it's very powerful, and we're going to use AI to fix feedback.
I knew that it needed to happen, and so I put out a lot of feelers on the internet trying to find anybody who would even give me the time of day. Imagine that you're actually some high-end AI coder somewhere in the world,
¶ Finding an AI programmer who was interested in doing this work
and here I am reaching out to you. Hey, I'm a sound engineer from Salem, Oregon. Can I take a few minutes of your time to ask you about solving the problem of audio feedback using AI?
¶ Social Media to the rescue!
It's amazing that anybody even talked to me, really.
Yeah, but I'm guessing the right AI, and I mean, clearly it exists, so you found the right one, but the right AI programmer would rise to this challenge, I would think.
Well, I paid some people some money to try some things that didn't work out. And years went by. I was out a lot of personal money. But late 2023, I actually found a group of coders that aren't in the audio space. They come from legitimate AI programming space. The projects they're working on were extremely high-end, and I just happened to grab their attention actually on social media. Easy. They just messaged me on social media, and so now we just text message each other. I've never met them.
They're from China and places like this, so I demanded one time that we do a video chat just so I could see they're real people. Yeah, yeah, yeah. Before I started making bank transfers. Yeah, that's fair.
That's fair. Sure. Yeah, yeah.
So I literally, we had one video chat and then the rest has just been text message based. And there's no like work hours or whatever. We just kind of text each other whenever and all the time, all day. And I just started making bank transfers all through 2024. Yeah. I had no idea. I gave the project a 4% chance of success. I was like, this thing probably will fail. This will go nowhere.
But you did it anyway. That's what I love.
I can't wait around for the big companies because they're obviously not going to do this anytime soon. I don't even have wind that anyone's even working on this. So I got one life to live here and we need to solve this problem. It's ridiculous. So yeah, I just started paying these guys way more than I could probably actually afford. But at some point, I think my wife was like, you know, if this doesn't work, we're going to be poor for a long time.
Yeah, I've had those conversations. Yes, yes. It was risky.
My father stepped in to help me financially, too. You know, I mean, he's watched me. He's put a lot of money into my crazy projects through the years. But this was significantly more than the other ones. And I think that even he was a little bit like, this whole thing is so weird. You're just sending large amounts of money suddenly to places like China to just have coders you've met through basically text message who you've never met in person do high-end AI coding
work that you have no background in. I mean, the whole thing is just, it's absurd. It was so absurd, my bank actually shut me down.
I was going to ask.
I had to, I had to move to a different bank for a little bit there.
Yeah. I had, I had a, um, we're going back, you know, 25 years, something like that. We were in the, with, with Mac observer, which is a website I started and ran, we ran for a long time. Um, we were in the web ads business and we started expanding the business and managing ads for other companies. And the. The idea of having a networkable ad server did not exist, right? There were like little ones. And so we worked with this programming team in
Russia at the time. They were in Siberia up in Novosibirsk. And I mean, I sent them money for a decade before I ever even had a – I mean, the first time I had a voice chat with them was in person in New York. And that was 10 years later. Like it was all text messages.
And I had an issue with my bank. they were like you can't be sending money to russia i'm like i have to oh yeah people work there for for me and it was really hard 25 years ago to prove to the bank that like this was legitimate yeah yeah it's like they're like this this is this year the next thing you're gonna do is is meet a nigerian prince and like it's gonna be like oh no they're real people i take look at what they've made i had to
literally sit down with the bank and show them like nope here's what it is and they're like okay you know i think they just wanted to beef up their file for when i sued them for letting me send money to russia i've
Had to do that i've definitely had to direct uh certain individuals in the financial world to the uh d feedback the alpha labs website to prove that this is the thing this is the thing they go they look at it and they go oh okay yeah we get it now yeah um yeah but you know at the beginning when i had nothing to show for it just imagine this Like, this is crazy.
Right. Oh, same. Yeah, yeah. You know, I mean,
Yeah, I didn't go for as long as you did, but, you know, I went for a solid six months before we had anything. And, you know, we tried many things and they all failed and things just weren't working. And I remember it was like –, It must have been like 3 a.m. their time. Yeah. And I got a message, and my main coder was like, I think I've done it. Because the biggest problem is we were trying to get it to work at zero latency, and it just couldn't.
¶ At 3am, a text message: I think I’ve done it.
It wasn't sounding right. And, you know, things were. So anyway, that was, I think, yeah, we, we, We got it working at zero latency, and then I decided to just start using it for my summer shows and concerts and stuff and events. And then I put it in my first installation in August of that year at a Catholic church in my town. And so it seemed to work.
Yeah, yeah.
Because I had no intention of actually selling this publicly. My only intention to begin with was I just wanted to use it for my work and my installations. In fact, so right now, I'm in North Carolina. Two years ago, it was actually literally around this same time that I was just
¶ Installing De-Feedback in Churches
starting to make the first bank transfers to the coders. And they were beginning work on this. And I was in the middle of installing a big sound system at a Catholic church here in the Raleigh area. And so I installed that system. And again, it's 50-45s. It's all the stuff to try to get gain-before-feedback out of these lobs the priests want to wear out in the middle of this huge reverb. It's a 300-foot front-to-back, massively reverberant Catholic church.
And they want to wear lobs right in the middle of the whole place with speakers everywhere. And my intention was I want to try to get to a point where I can do an installation like this completely headache-free. I just want to just turn on whatever processing it is and not have to spend a week trying to fight for gain-before-feedback and I can do it okay. Like, it's fine. It's just, I remember two years ago sitting in the back of that church just thinking,
I've got to get this. I can't do another install like this. It's just I'm sweating bullets every time the priest goes out there wondering if it's going to feed back.
This is the time, yeah.
Yeah, yeah, yeah. And so it's just so interesting. Unfortunately, I just didn't have the time to – we couldn't get it working fast enough for you to put it into that installation. So here I am two years later coming back to install D-Feedback now. Oh, I love this. In that church. And today, actually, after we're done here, I'm going to drive to South Carolina to do the same thing for a church. I'd done a year before that, and now I'm going to put D-Feedback in that church.
Amazing. And so I've been running around basically installing D-Feedback in all of my legacy installations. But again, that was my only intention with it. I didn't have any plans to sell it publicly. But the problem is I started using it and posting about it on social media. And everyone's like, well, how do I get that? And I'm like, I don't know how you get that. I mean, maybe I should sell it.
There's these programmers in China you should talk with. Yeah, yeah, yeah. talk to your bank
Yeah so uh amazing oh no i decided i was like let's just make a website let's just like it's a plug-in like that's how we've been testing it sure plug-in so why don't we just i don't know anything about this i don't know anything about e-commerce i've never sold anything before why don't we just make a website and however you do that and then let's like sell it and however you do that i don't know any of these things you know just uh so yeah we we threw that together and,
that fall, November 11th.
We launched November 11th of, of, of 20, 24, 24. Okay. So it's, it's been on the market for about a year and a half now.
Yeah. Yeah. Yeah.
Amazing. I'm going to take this opportunity. Well, you asked some questions, right? How do I get this done? How do I put a website out there? All that stuff. It, it, I can't help but be reminded of the two sponsors we have for this episode. So, and I really appreciate this every time I'm able to do this. And I thank you folks for going with me for a minute and listening to what our sponsors have to offer because they are the ones who make it possible for GigGab to exist.
We all make it possible. You're listening, guests like Devin, all the work that my team does, and our sponsors help kind of round that out. So thanks to everybody. And yes, I want to take a minute and tell you about our sponsors here. And Claude, our sponsor, has quickly become my all-in-one backstage crew AI
¶ SPONSOR: Claude.ai – Ready to tackle bigger problems? Sign up for Claude today, which includes access to Claude Cowork, too, when you visit Claude.ai/giggab
for my creative life and beyond, right? I use it to turn every GigGab episode into clean, reliable show notes, brainstorm content, and come up with those punchy episode titles that make you actually want to tap play, right? It helps me plan our calendar and social media. It helps me process my email. It connects the dots between ideas. And even throws out new segment concepts when my brain's still stuck at soundcheck. Claude is the AI for minds that don't stop at good enough.
It's the collaborator that actually understands your entire workflow and thinks with you, whether you're debugging code at midnight or strategizing your next business move. Claude extends your thinking to tackle the problems that matter. Claude Cowork brings Claude Code's agentic power to your desktop, right? No terminal required, so anyone can use and have Claude do real work. You point it to a folder on your computer or add connectors like Google Drive and Gmail.
You describe what you need, and it handles the rest, like organizing files, building spreadsheets, drafting reports from scattered notes, right? You queue up your tasks and come back to finished work. So, ready to tackle those bigger problems? Get started with Claude today at claude.ai slash giggab. That's claude.ai slash giggab. And check out Claude Pro, which includes access to all the features mentioned here in the episode today.
Claude.ai slash giggab. Go check it out. I think you're going to like it. I sure do. And my thanks to Claude for sponsoring this episode. And, you know, as a working musician or a band leader or the friend who never
¶ SPONSOR: Squarespace. Check out https://www.squarespace.com/GIGGAB to save 10% off your first purchase of a website or domain using code GIGGAB.
misses a live show, you need more than just a social feed, right? You need a real home base online. Our sponsor, Squarespace, is the all-in-one platform where you can build a killer website, sell tickets or lessons, right, and actually get paid all in one place. I love that you can start with their cutting-edge templates, then use their blueprint AI tools to shape the site around your vibe, right?
Your genre, your personality, your story, your quirks, all those things get baked right in without needing to know a lick of code. You can drop in your live dates, embed your videos, sell VIP bundles or lesson packages, and hook up email campaigns so that your fans never miss a show. Their built-in scheduling and payment tools mean you spend less time chasing invoices and more time doing what we talk about here every week. You can always be performing.
So check out squarespace.com slash giggab for a free trial. And when you're ready to launch, use offer code giggab to save 10% off your first purchase of a website or domain. Check it out. I've been using Squarespace for a lot of different things for years, and I love it. You're going to love it, too. And my thanks to Squarespace for sponsoring this episode. All right. So this is fantastic, man. I'm really appreciating this conversation here.
There's a question I wanted to get to because we've talked about impulse responses
¶ What is an impulse response?
many times, and they are at the core of what you're doing with D-Feedback. Can you – and if you can't, it's okay. We'll figure it out together. Can you explain what an impulse response is for people who don't know?
Yes, and I'm actually, I think I'm well-suited for this because, to be honest, I don't really understand them. I'm not a coder, I'm not a mathematician, but I work with them, and I can explain how they function, I mean, what they do. Great. So, you know, okay, so let's say that you have a room, like this room that I'm in, okay? And you, an impulse response is kind of like taking an audio picture of how the room sounds.
Like at this, you know, at this point in time, like as long as we don't change the walls or, you know, put a bunch of new furniture in here, put some acoustic treatment or open a window like the room. If you make a sound in the room, then the room will do its thing. The sound bounces around and it decays and it has a certain sonic signature that is persistent. Like it doesn't change.
It doesn't change unless you change the nature of the room. Yeah.
Unless you were to change the room.
By putting bodies in it, more bodies other than yours, right? You know, that would change things. Or if you move the bed out of the room, if it's a hotel room or that kind of thing. Yep.
Okay. You can change the impulse response of the room by changing something about the room. You can make it bigger or you can remove a wall or you can put a bunch of stuffing up on the ceiling or add a bunch of people in it. You're changing the impulse response. But as long as you don't change anything about the room, the impulse response of the room stays the same.
And it doesn't matter what audio you impulse into the room, like me talking, or I clap my hands, or I do whatever, Whatever sound gets emitted into the room as an impulse, the response of the room will be what it is. And so that's an impulse response.
Got it.
Got it.
Okay.
Yeah. Yeah. So basically what you can do is there's a certain type of mathematical algorithm that can express that. And so you can create a filter, for example, that represents the sound of that room. And then what you can do is you can use a process called convolution where you can take any arbitrary input, like a vocal signal or, for example, anything, a guitar, a drum, and you can convolve it in real time with that static impulse response.
And then it will make that guitar or vocal sound like it was in that room. So that's when you insert a reverb plug-in into your audio chain that uses an impulse. If it uses an impulse response algorithm, that's what it's doing. And then you can load other IRs into the program to simulate other rooms. And you can go select from a menu of different rooms. And those are all recordings.
And typically, for example, if you actually go listen to the IR itself, if you go into the IR file or folder menu and listen to it itself, it just sounds like a . Usually they create these by popping balloons. They just take a balloon and pop it, and then record the sound of that. That's an IR. Oh, okay.
Thank you. Yeah, yeah, yeah. Yeah.
So that's basically what happens is that you can take a literal audio recording of a room. That is the actual IR. That's like the impulse response is the broadband frequency response over time of what the room did with this impulse sound. And then you can create a mathematical algorithm that represents that. And then you can use convolution to make any other sound sort of sound like it was in that space and produce a new audio file that has the room baked into that.
Okay, so once you can do that, Well, then the question is, what if you don't actually have to go in and you don't have to pop a balloon in a real room and use real microphones to record real audio? What if you could just create a program that can arbitrarily change things about
¶ De-Feedback is an on-the-fly IR generator
that math equation, about that IR expression, that you could create the sound of a new room just by tweaking some parameters? You don't actually have to go find a room and pop a balloon in it.
Sure, you're just creating. You can just, yeah. So you're creating the impulse responses on the fly and then feeding them to the algorithm to see if there's feedback in that picture and reverb and other things.
Yeah. I mean, you can think of it like you are creating an AI. So let's say it this way. There's an IR generator. Think of it like a gigantic switchboard with all these knobs and stuff that if you start twisting them, you can create whatever IR of anything. It doesn't have to be just like a room. You can create an IR of any static sound that you can possibly think of. And as long as you twist the knobs just the right way.
And that could be anything from the sound of a room to the sound of a car engine to the sound of background noise, chatter, whatever you could possibly think of this IR generator could create. Okay, then the question becomes, well, that's a lot of knobs. It's literally, in some cases, millions of parameters if the model is complex enough. So you have all these parameters at your disposal to play with, to create an IR, okay? The potential is unlimited. You can create an IR for anything. Sure.
Now what we do is we have an AI that we have trained.
To listen for the sound of human voice and things like room reverb and background noise and the potential for feedback and if the if the ai is sufficiently trained and is doing its job right we give it access to those parameters and we say here have at it try to figure out how to twist these knobs in just the right way in real time to constantly be updating the IR that we are convolving with the incoming audio signal to try to create an IR that is the opposite of whatever we're targeting.
Like we're telling you, leave the voice alone.
Don't change the human voice in this signal but if you hear room reverb we want you to create an ir that's the opposite of that room if you think there's going to be feedback because of this and this and this and this about the audio signal create an ir that has the opposite of that in it if you think that there's some background noise that needs to be removed create an ir that expresses the opposite of that so whatever it can figure out how to cram in that ir right now right
it's constantly updating to stay on top of reality, then anything that's in that IR is going to just cancel out what we're targeting and leave everything else alone.
This is brilliant. This is why no one's done this before. I think you deserve a lot of credit for this, and I think you deserve more credit than you give yourself because you are asking
¶ Finding the right programmers was the key
the question, why has nobody done this before? Nobody sat and thought about it this way before. At least not combined with the drive and willingness to push forward and fail and figure out how this could actually come to be. You knew that this could come to be. You didn't have the programming skills to make it happen. But you had the bullheaded persistence. People ask me, you know, you have a little bit of success in business and people
come to you with ideas, right? And what they're asking is, can you sprinkle your magic fairy dust on my idea? And the reality is, as anybody who's failed and or succeeded in any venture knows, there's no magic fairy dust. The magic fairy dust is bullheaded persistence. You're willing to just keep pushing forward because you know it's going to work. And sometimes you're proven wrong, right? I've had plenty of failures, you know, and I'm sure I'll have more.
But when you make it and you're right, that's when people are like, oh my gosh, this is amazing. But, and it is amazing, but I really think, thank you for doing this and pushing forward and like you had the vision and stuck with it, despite everyone telling you this is harder than you think. Maybe it was, but you still got there.
Well, and you know, If there have been other people that have been entertaining trying to do this, I think that the roadblock for them was that they just could not find the right people to actually make it work, especially at zero latency. Because I think that from what I've learned from interacting with the R&D departments of a lot of these large household name brands and companies, in the audio space at least, is that their engineers come from the audio space primarily.
And what's happening is that a lot of these companies are having basically what are audio guys who learn to code, and they're like trying to learn AI stuff so they can do AI work in the audio space. It's like that is not the right approach. That's not going to get you. You can do some things. Of course, there's going to be a lot of cool AI apps and programs that come out for the audio world. But something like this is so technically complicated and challenging that you
needed – it's not like I really even intended this. It just sort of happened this way. I found –. Guys who are not, they're not from the audio space. They're from the AI world. And I grabbed their attention for 20 minutes and had them think about something in my world from the audio space. And we had to do a lot of, actually, it's funny because, you know,
¶ Mind-melding was necessary
you know, talk about using Claude, like using these large language models in the beginning was extremely critical for the success of this project because I don't speak these guys' language and they don't.
Like I was just going to say, quite literally, yes, on many levels.
There's multiple other languages involved here, you know. But then also, I don't speak their technical language, like talking about high-level mathematics and neural nets. And then when I talk about the audio world, and there's a lot of stuff that just goes right over their head. They don't know what I'm talking about. I actually used a lot of early large language models, including Claude, actually. So I would do this thing. I told them this is what I was doing.
I would do this thing where I took what they would talk amongst themselves about what we were doing technically. Large, huge blocks of math and equations and code. And I would take that and I would put it into whatever it may be, Claude, ChatGT.
Yeah, yeah, whatever LLM you were using.
I was actually, I was using them all at the time, you know, kind of experimenting.
And firmly believe as someone who uses AI a lot, I firmly believe that having access to multiple LLMs and having them fact check each other. And like that, that is in today's world, that is necessary. And you can do it with it. Like chat GPT doesn't, I guess they do make it fairly easy, but Claude makes it fairly easy to like switch LLMs. Right. Cause Opus is one, Sonnet is another.
Right. very very simple having them fact check each other is like that's how like when you when you hear about people who are using agents and mass and doing all this stuff it you're just they're just tapping into different llms it to to and feeding them the same prompt or or you know similar prompts and and doing that that's what those agents are doing and you can you can do that on your own too It makes, it really, really helps, yeah.
Yeah, well, so I would basically say, I'd go paste all this in there, and then I would say, I would have this prompt where I'd say, this is what we're doing, is, you know, this is my whole backstory and everything.
Yeah, you gotta give it context, right.
Right, and so then I would basically say, here's what I need you to do, is explain this to me like I'm five. Like, just, you know, pretend I'm an idiot and I don't know anything. And then I would have a conversation with it about, what needs to happen, you know, from an audio standpoint. And, you know, this would include my real world tests and stuff like this. And then what I would do is I would basically ask it to convert what we had concluded by just me conversing with it.
And I would ask it to convert that back to smart people talk. I would send that back to the coders and they're like, this actually works. Is actually pretty productive. And so it allowed someone like me to interface with coders like them all over the world. I think that if I had tried to do this even just a year or two earlier, it would have been impossible. There's no way. We couldn't have communicated. And so that was very unique.
That's fast and really smart. What a great use case. For ai i mean like literally this this project helped humans that the ai in this in the context of this project help humans collaborate right like you wouldn't have been able you just said it you wouldn't have been able to do this collaboration without being able to translate what you needed into language they spoke efficiently so that well yeah i
Think it's an interesting this is kind of a side point, but if an LLM manufacturer
¶ Using AI to make it possible to collaborate with other humans
were to want to do a case study on this kind of success story, I would actually put this project as a high candidate for the fact that look at the potential here. Look at the ability to have people that are from disparate backgrounds who have nothing to do with each other, cannot communicate in any other way.
All we've done, all our tool did was allow them to talk to each other over the internet and look at what they've done they've solved the problem of feedback for everybody like this is an amazing example of what's possible if people can just communicate with each other that's.
It yeah i mean
You know you.
Know to your point where you have um coders who be or sorry already audio people who became coders uh They think a certain way, right? And you needed programmers that thought a different way. But the problem with that is they think a different way. And so the LLM becomes that translation collaborator, if you will. Really, really smart use of it. Man.
I've used it for so many things. Same. It's my early contracts. Like, you know, all the contracts we made between ourselves were made by AI. Of course. We have, I use it. And so here's another thing I used it for in the beginning.
¶ Using an LLM to rate the work of your contractors and employees
I was, again, I was like, this could be a scam. There's always the potential that this is just a big scam somehow. I don't know how, but maybe I'm sending all this money overseas, and this is just, I'm getting taken somehow. Well, so what I would do is when they would send me prototypes and stuff of code and of the plugins and stuff like this, I would take sections of it, and I would put it into these LLMs, and I would ask it to rate their work.
Say, look, give me a rating. Is this intermediary-level code? Are these guys doing mid-level work? Is this super high-end, cutting-edge type of stuff? Very consistently, almost everything they sent me. In this particular case, I was using ChatGPT, and it would tell me that these guys are operating in like the 90th percentile when it comes to cutting-edge AI work. This is extremely high-level, extremely cutting-edge, and this is not a scam.
And then, of course, the proof is in the pudding. It actually just works.
It works. Yeah, yeah, but you couldn't know that at that point in the project. Right, exactly.
I could know until literally the summer. It was like June before we had our first functional version I could actually use. And so for months and months, I was sweating bullets going, maybe this is just a big scam somehow. But that was reassuring to have these LLMs. All of them that I tried were consistently telling me, Claude, ChatGPT, there were a couple others that were like, this is high level. This is extremely cutting edge stuff.
Amazing.
That's awesome. That was another thing that was interesting.
¶ How do we get De-Feedback into the hands of working musicians
How do people, thank you for this, and thank you for this little detour. I know it's not really music stuff, folks, but it's the world we live in, and I'm fascinated by it.
Well, I think I know where you're going with the next question, and I think it does have to do with musicians. And here's why. Because let's say that you're a singer-songwriter, and let's say you show up to your gig. How often is it that you're gonna show up and the sound engineer just doesn't know what they're doing, obviously, or won't listen to you, and there's feedback, and there's feedback in your monitor, there's feedback through the speakers,
and you can't do anything about it. You're powerless. Yes. Let's say that you were to go on our website, dfeedback.ai, and you were to get the tiny little computer that we have up there and the little Focusrite interface. Anybody knows how to run those. You just plug them in, and we've tried to make it as turnkey as possible. You carry that in your gig bag, and you show up, and it's like, okay, I've got my microphone on the stand here.
I see where the cable goes down and plugs into wherever in the floor pocket. Well, I'm going to just take that. I'm going to plug my mic into this, and then I'm going to come out of this and go into that floor pocket. Now you've just inserted D-feedback through your vocal mic. And yeah, the sound engineer might give you a little grief for that up front because they don't like it when people do that. But as soon as they hear what it does, I doubt that they're going to,
in fact, by now, they'll know what it is. And they'll be like, yeah, of course, obviously you can do that. So that's a great way for musicians.
What's the cost of that to acquire what is needed to do exactly what you just described? What is like the all-in cost of – because there's a little piece of hardware, which is essentially the computer that's doing the processing, a Focusrite interface to interface with the computer, and then obviously the cost of the software, the plug-in itself.
Yeah, so the fundamental is the license. You gotta buy a one-time license to be able to own it and use it forever, this version of it. Sure. Which is, we're gonna come out with better versions, of course, but this version is great. And many, many people could just live with this version forever. They never have to update if they don't want to. It's a one-time cost of $4.99 US dollars to just own a license to use the plugin. It's an audio plugin, okay?
Then you have to find a machine to run it on. It runs on any machine that will run like a VST3 plugin for Windows or for Mac. It'll do VST3 or an AU.component for Mac. We recommend people run it on Windows, though, because there are reasons why Mac is a little bit more difficult to get to work. We don't have access to the Mac OS in the same way we do for Windows. Mac is more of a closed system.
Yeah windows for people who also listen to mac geek gab uh you heard you've heard my trials and tribulations with core audio over the years uh and apple keeps changing things to core audio they
Change things you don't have access to the kernel no you can't get in there they so with windows we're we can we've got a lot more control um and so we can do things to.
Make it work if someone has a windows machine of a certain caliber could with their 499 with their $500 license forgive me for the dollar uh they in theory they could and and they could get their mic in and out of that in whatever way they currently have that would be enough uh in most cases to do a single instance of of this is that right
Most likely. I mean, yeah, many.
Many people do that.
You have to test it. But so a lot of people are in a position where they don't have a high-end machine to try to run this on. They don't want to deal with it. They just want something that's approved, that just works, that they can just order. It shows up at their doorstep. They just unbox it, and a couple button clicks later, it just works.
Um and so that's what we and also like it's they don't have the money for like a high-end machine yeah yeah yeah five hundred dollars was already a lot to spend on just a license on just a plugin so yeah so we um what we did was we worked with the company um to have these little tiny like square computers they're called headless because they don't have like a screen or keyboard or mouse or anything in them and we worked with them to create a set of os settings
and configurations on those computers that's highly, highly specific.
We worked for
Almost a year trying to get those things into shape so where they would run dfeedback flawlessly and at very low latency. And we tried a bunch of different interfaces because this is the other question I get a lot. Well, can I just use my own interface I've got laying around or whatever? I mean, you're welcome to try, but we tried like 30 some different interfaces. Maybe they're a great interface. Maybe they're expensive and it works fine otherwise.
But when you run D-feedback on a computer, because it is so unique, it's so not only just in general, it's very, very heavy on the computation side of things. But it's also the kind of code that it's wanting to run through the CPU is very, very unique. And that doesn't play very nicely with things like audio drivers in a lot of
cases, unfortunately. So you may have a great audio driver for a great interface that you've had forever that works fine, and you run d-feedback on your computer, and then you get audio glitches. And so we tried a bunch of different interfaces, and it just so happened to be the case that the Focusrite Scarlett series with a very, very specific driver that we got from their R&D team, that works well. We know that that's solid on these computers from our website.
And so we tried to set it up to where we're barely making any money on the sale of the computers. The goal was just we don't make any money on the Focusrite sales. We're just directing people to Sweetwater's website. You can buy the Focusrite there if you need it. You can order the computer from this other third-party computer company. The computer that will run one instance, which is what most people need anyway, because you can also subgroup vocals together if you want.
So if you have a panel discussion and you've got a bunch of people talking, but they're taking turns, well, you can group all those together and send them through the one instance. You don't need an instance for each person.
For each microphone. Oh, interesting.
Maybe you could do that.
Sure, sure, sure.
So a lot of people buy that first. We're calling the option one computer, which is just that little computer with the single-channel Focusrite interface. That hardware combo is like $600. It's just a little less than that because it's like a $450 computer with like $130 interface or something. And... You know, you get that. Yeah, option one. And so that's that little computer with the little Focusrite interface down there.
Yep.
And then you plug that in, and we've configured it to where once you set it up initially, you have to enter your license and everything and go online and do that. Do basic config. But for the most part, I mean, it's easy. It takes, you know, 10 minutes to set up. And then after that, you don't ever have to connect a screen or a keyboard or mouse to it again. You can just show up to your gig. You can just turn it on, and it'll just start passing audio. It's like a standalone unit.
And then the other computer has four channels. And then, of course, now we've got some higher-end units that we've been working with, like Waves, to make sure that it runs on their piece, the live box.
Okay, so, and that's where we met, actually, in person, anyway, at NAMM, was when you were in the Waves sort of umbrella there, demoing this, and that was on a live box, you said, right?
Yeah, because we had been testing it. I bought one of the first live boxes out, like, on the market and was testing the plug-in on that for a long time, and Waves got wind of that just because I was very public on social media about this.
Yeah, yeah.
And it so happened to be the only unit, in fact, today it still is the only unit that we know of like that, those types of units that will run our plugin successfully. So I was perfectly happy to tell people, hey, it runs really well in the live box. And they're like... Hey, you just want to come to NAMM and demo in our booth? So I was like, sure.
That's amazing.
Why not? So yeah, that's what that was about. And so it's great. And there will probably be others soon that will also be able to run it. So we'll just keep expanding the list of approved hardware out there.
Is there, maybe I should ask this differently. Can I make any introductions necessary? It would be great if this were an option baked into the kind of all-in-one mixers that, like, you know, Allen & Heath sell and, you know, Behringer, Midas, all of that stuff. Like, it seems like that's where one of the places where this should be. Just baked in. It's an option. You can pay your $500 and unlock it. You know, Allen & Heath has a lot of those options, so Midas & Behringer.
So those would be, I feel like, kind of obvious expansion opportunities for what you're doing.
Yeah, I mean, well, there's many expansion opportunities. Yes. And one of the difficulties right now is, you know, I mean, the plugin has been
¶ Where do we expand?
out there for, like you said, it's a year and a half. And I would say in the last six months, and really particularly the last maybe four months, we've been on this sort of upward trajectory of, you know, adoption. Like it's starting to be used at very, very high levels in the industry. Oh, yeah. Although we had some early successes. I kid you not, within 48 hours after launch, the guy who mixes for Mariah Carey, Paul Falcone, he messaged me and was like, hey, is this for real?
I've kind of been following this. Someone told me about it, that it's public, and I want to use this for Mariah Carey's Christmas tour. They just started. So out of the gate, it was used for Mariah Carey's Christmas. I was sitting there November 10th, 2024, and I was like, oh, is anyone going to buy this? Yeah, yeah, yeah. And then, you know, here we got, you know, Mariah Carey touring with it. So that was very cool.
We had some, you know, but then once you get guys like Robert Scoville, you know, picked it up, Pat Baltel, you know, Drew Aldridge. Randy lane todd windmill i mean there's like this collection of people that, uh really i think them just organically picking it up oh yeah no sending me pictures scoville.
Told me that he heard about it somebody on his weekly mostly weekly back lounge uh yeah sort of zoom call asked him do you know anything about this and he was like no you know yeah so yeah dug in and then it turns out you guys live like a couple hours each
Other we well we live like an hour and a half from each other so we we met up a couple times for coffee and yeah you know chit chatted and uh i sent him a demo unit and like the next weekend he was using it for rock hall for rock yeah.
He came on the show and talked about that yeah exactly the
Whole thing was it just yeah really it it it really um helped uh sort of get the word out and and of course then at nam with the waves thing and so um there are many conversations that are being had now um Um, and there's a lot of, there's, I think the, the, the difficulty is that there's so much potential for this type of product.
Yeah.
Um, yeah. It's hard to, first of all, it's hard to know what direction, because we can't do everything at once. I mean, we're a small team. We're not some big company with departments, and here, I'll have my people handle that. It's like, I am.
The 24-hour tech support for the product. Yeah, you are people.
My coders are all working on stuff we have to fix. Like, I don't have time to do other stuff right now, and I don't have that much money. I mean, the company is profitable now, which is nice, because we're not sweating bullets anymore financially.
Yeah, yeah, that's great.
But no one's getting rich yet. It's enough to keep the doors open, to pay the guys, to continue to work on stuff, and it's more comfortable. But we don't have a bunch of extra money to just throw at other totally different side ventures. We just bought a new training computer to try to train V2 of the software. That's where we're spending most of our time right now.
Right. So going back, you know, earlier in the in our conversation where I said, you know, you take all these examples, you train it and then distill that down. And that's what the real time computer does. But that training takes a very powerful computer, as I said, or series of very powerful computers. And you have to have those. You you are you are managing them, too. That's right. Yeah. Yeah. Yeah.
Yeah, well, as opposed to a lot of other companies that do AI work,
¶ Homebrew AI!
a lot of other companies are using off-the-shelf AI products. Sure. They're just renting compute space online. They're using... Training data that is kind of like common data sets and stuff like this. We are the equivalent of homebrew AI, is what I could call it. One of my first tasks in this whole project, I'm asking my guys late at night, what do we need? What are we going to need?
And they're like, well, we're going to need a very, very large data set of vocal data and IR room data that has to be done in a really specific way or it won't work for what we're doing. Um, cause there were, um, like some publicly available data sets that you could have tried to use. Um, we've, you know, could have paid a lot of money to have some made, but I'm like, well, you know what? I'll just do it.
So I called up all of the venues I'd installed sound systems in and people I knew around town and asked janitors and church worship leaders, like, Hey, can you let me into your building at like 3am? I just need to make some IRs of your room and I need there to be no traffic and no one making noise and stuff. and I'll lock the doors behind me when I leave. So I did that for like maybe three or four months in a row.
Just all I did was run around with a speaker and some microphones and an amp in the back of my car. And anytime I could, on the fly, I'd get a text at 11 a.m. someday. Like, yeah, you know, like, come on by. There's nothing going on. I would, I'd go and I would make IRs. And I collected just so, so many, you know, thousands and thousands of IR files from all these rooms. I did every room in my house, including the bathroom.
Of course.
You know, and then vocal data, we collected a lot of vocal data, and some of it I paid to have made because it was just too hard to figure out how to, you know, either find it or get it otherwise. And, um, so, and then, so that's the data set. So it's a much of it is stuff that we made ourselves or paid for.
¶ Creating your own AI model
And, um, then there's the model. The actual AI model itself is one of our own making from scratch. So we own the base model. This is very, very unique. Wow. Um, this is just in the AI space in general, not just in audio.
Most people don't have their own model. You're using, I mean, you know, we talked about like Opus or Sonnet or Llama. Or, you know, all the other ones.
Splitter and the AI, like audio world. I mean, there's many off-the-shelf models. And we looked at all of them. We're very familiar with all of them and none of them were going to work. So we just started from scratch and we built our own.
Amazing.
And then, of course, the machine, you know. So I wanted to, in the future, I just thought if we are to get contacted by companies that want to do an acquisition
¶ What’s the future look like?
or have demands for security, if we start, I don't know, I'm just throwing this out there, but let's say we were to work with financial institutions or government institutions that required a certain level of security to our product. They want to know the history of what we've done. I want to be able to say, this is our own model.
It's all in-house.
We trained on our machine that is behind a commercial-grade firewall at all times. So stuff like that.
I was trying to think ahead about. It's very interesting. Most entrepreneurs wouldn't go that far. I realize... Some of it was enforced by what you were trying to do, like in that the model, the right model didn't exist for you. But you took it a step farther in that. I mean, you've created a moat around your, you know, your business here. Obviously, someone could compete with you. I'm presuming you are either in the process of or have filed some patents and things like that.
But, yeah, that's great, man. Thank you for telling your story here. This is this has been amazing. Yeah.
Yeah. And, you know, I think that definitely when it comes to,
¶ You can get this and bring it with you to gigs where someone else is doing sound
you know, musicians, you know, singers, this is not something that I think the thing I keep wanting to stress to people is you don't have to wait around for your sound engineer to figure it out and like go get this. you can just go get this. This is something we designed so that quite literally, you can buy what we're calling the option one machine on our website. If you want to get that, and you just want to have your personal D feedback rig that you carry around to your shows and events.
And like I say, maybe a year or so ago, you would have had a little back and forth with the sound engineer to make that happen. But nowadays, I mean.
We talk a lot about back and forth with with sound engineers here and and so our listeners are are fairly savvy now we you know it because there's always something you're going to want to inform the engineer of or present to the engineer and there's you know there's ways you don't just walk in the room and say i have deep feedback you know or i have this need you yeah you you know you develop a rapport first and hey i've
got this thing it works really well i've tested it other where i'd love to be able to use Let me know what you think, like, you know, the approach matters, and you know, it's, I always say, even if you're just meeting your front of house engineer, Your front of house engineer is a band member for the evening. And if you go in with that approach, most of the time they will adopt that approach with you.
Sometimes, you know, people, they've had a bad day or whatever it is, and it's not quite as simple as that. But oftentimes it really is that simple. Like, hey, you're in the band for the night. Like, we're all working together. Same common goal. You know, we're going to support what you need from us. You're going to support what we need from you. And when you go in with that attitude, showing up with D feedback, like you said, most engineers now are aware of this thing, which is great,
makes life much easier. But yeah, it's- Well,
And you think about it, if you want to show up, and a very common situation, for example, is, I show up, and I've got a vocal effects pedal that I want to insert directly into my microphone.
It's the same thing. Yeah, right.
Look, I mean, we already negotiate that, and every once in a while, I guess, you're going to find a sound engineer who just won't put up with that. Yeah, I hate them.
If I'm doing sound for people, I hate it. But the reason I hate them is that people, singers, usually aren't sound engineers, and so they don't understand gain, and specifically gain before feedback, well enough to tune their pedal that's the only problem i have with them they can be great if they are set not to cause feedback which is which is usually my issue with those things you know they've got some heavy gait and heavy compression on them
and as soon as they talk it's like it's like okay well
And so here's the thing is that you know already in those situations um it's it's very few and far between that singers will lose that battle yeah normally They get to insert whatever they want, and the sound engineer just puts up with it. Now, the whole point here is that the sound engineer's main worry is that it's going to sound bad, and it's going to cause feedback.
This is the opposite of that.
This is the opposite of that. It's like if you plug this in, what sound engineer is going to hear the feedback and the background noise and reverb get reduced in the vocal mic and go, Oh, no, give me the feedback.
I want my raw signal back. No, that's not going to happen. No, it's not going to happen.
They're not going to say no to this. And at this point, they're going to know what it is anyway and probably even be using it, you know? Yeah. So, but, but that's, yeah. And of course, this goes far beyond musicians.
I mean, this is, you know, anything from, you know, if you're, if you run like a restaurant or a hotel and you have like a, You know, or you've got a club or something or a theater and you just want to, you know, have microphones that, you know, a lot of times they feedback because maybe you don't have a regular sound engineer. Oh, yeah. The system kind of has to just run itself. You know, there's like a wall panel and people can go turn on a microphone and do their conference or whatever.
This is baked into mixers that people are buying so they don't even have to think about buying an extra thing, you know. Yeah. I mean, it's these are early days. I mean, it works like success. This is amazing. Thank you. Really. I mean, for telling your story. Yeah. Yeah.
The conversations with console manufacturers, it's a little bit difficult because it takes a long time, years and years to even plan ahead for a potential console release that's going to incorporate a radical new technology that's… When.
I was meeting with Alan and Heath at NAMM, it was right after I spoke with you, right? So I just had this demo from you, and it just works, folks, like I said. And I went over to their booth, and they're showing me the new SQ6, I guess it was, SQ6, I forget.
I should have it in front of me, but I don't. And they were showing me, they're like, oh, it has this really advanced feedback detection thing, and it actually works, and it's doing fine, tight notching, so that it's not just killing all your high mids or whatever, which is what a lot of the kind of automated EQ based on feedback things do. They're like, no, this really works better. And I'm like, that's awesome. Do you know that guy over there? And they're like, no.
And I'm like, have you heard of D feedback? And they said, no. And I'm like, are you allowed to leave your booth space? Because if you are walk over there, that's what you're going to want to replace this with in the next model. But that next model is going to be three years down the road, right? Like that's, that's just how that, to your point, that's how that works.
Well, and, and, you know, look, I mean, it is pretty amazing what some other manufacturers have been able to accomplish using legacy methods and they're just getting better and better and better at it.
And that Allen and Heath, it does a great job at it.
It can do probably pretty good. But the thing is, let's say you got to the point where it could use EQ notches to try to recognize and fight feedback. Well, one of the issues is still with a lot of those systems. In principle, they have to wait for feedback to start to occur and then squash it. We want to prevent it from even beginning to happen. And then also, that's not all our program does. It does de-noise. It does de-reverb. Soon it will do de-echo. It already does some de-echo.
It's being used in a lot of NFL stadiums now on the ref mics. It does a pretty decent job already of getting rid of echo, but that's not what we designed it for. We can do a way better job. This is what happens. You put a plugin out in the world and everyone's like, great job, guys. It does this really well, but what else can it do? Can it do this? Can it do this? We're like, happen.
For you yeah so
You know but yeah so at this point in time we've had no outside investment money we're just running on our literally we're running on bootstrap our own financial steam yeah that's great so.
It was the allen and heath qu5 i went and looked it up that and it really does like the feedback elimination in there uh but you're right it's it's it's essentially happening and then they're notching it out of of it when when the mixer hears it and and it's not it's not nearly as responsive as what D feedback does.
And, you know, I mean, people should have realistic expectations about the performance of D-feedback at this point in time. Like version one is essentially the result of me running out of development money. And I was like, we have to start selling what we have or I'm just going to go bankrupt. Yeah. And, you know, I'm going to have to move back in with my parents or something.
Well, I mean, you know, we always say perfect is the enemy of good, right? You know, and version one is very, very good.
Yeah, it's incredible. We tell people you should expect about 6 dB usable gain before feedback out of the gate. And then just situationally dependent. Some people get as much as 10 or 12. Every situation is different, but you should get about 6 dB automatically. That's just a given. And then you get all the other added benefits of it too. It just makes it sound tighter and cleaner because of the reverb and the denoise and stuff like this.
But, you know, we, yeah, set expectations accordingly. If you think, well, I need to get 20 dB of added gain. How come it didn't do it? Well, yeah. We'll get there.
Yeah, yeah, right. Yeah. No, thank you for doing this. What a pleasure to chat
¶ Gig Gab 524 Outtro
and get to know you and learn about the nerdy stuff behind it. Everybody, thank you folks for indulging my nerdiness today, digging into this with Devin. You had said it before. It's dfeedback.ai is the URL. Is there anywhere else? Where are you most active on social media? I want to make sure people can follow you there, too.
Um, I am most active on Facebook and Instagram. Okay. Um, so I still kind of, um, I haven't created like a, like a, we have a website that's just for alpha labs is alpha labs, audio.com. Yep. Um, but on social media, we're still kind of operating under the banner of the legacy family business, alpha sound. Okay. And so, um, um, that's kind of, if you look for alpha sound on Instagram or alpha sound on Facebook. And then of course on YouTube, I've got some videos.
I'll put those, I'll put those links in the show notes. This is great. Yeah.
Yeah. YouTube. I've been, I had a historical presence on YouTube, especially for making videos about just explaining basic audio concepts and doing fun things. And a couple of them got a lot of views, millions and millions of views, which is super cool. I've since sacrificed the algorithmic love on YouTube to putting out, like tutorial videos and announcements for dFeedback, which get like no views, but they're very useful for people using dFeedback, so unfortunately.
But yes, that is also where I post things.
I'll make sure to link people to that. No, that's great. Really, truly, thank you. This has been wonderful. Before we go, as we send people on their way for the week, there are three words that we always say here on the show, and I know I've said them a few times this episode, but if you wouldn't mind, Devin, I would appreciate you saying those three words for us as we say goodbye to everybody.
Always be performing.
There it is. Thanks for hanging out, folks. Thank you, Devin.
Yeah, thanks so much.
What a pleasure. Thanks for taking your time. It's nice chatting. This was great. This was great. Feedback at giggappodcast.com, or you can just go to the website, and the show notes will be there. You can sign up to get the show notes automatically delivered, so you get all the links for all the things. Thanks for hanging out, folks. Always be performing. We'll see you next week.