The Smartest People in the Room

AI in the Music Studio: A Conversation with Rich Keller & David Ronan

Hosted by Tom Truitt, Turnkey ZRG

What happens when a Grammy-winning hip-hop engineer, with 30 years of top-level studio experience, sits down with the CEO of a leading assistive AI music company? You get an honest conversation on the role of AI in music production.

David Ronan, CEO of RoEx, joined Rich Keller on Tom Truitt’s ‘The Smartest People in the Room’ podcast to discuss assistive AI, the origins of RoEx’s technology, spatial audio, copyright in the age of generative AI, and why neither of them would quit music… even if they won the lottery.

Here's the video:

Here is the transcript from this podcast:

Tom: Welcome to the smartest people in the room. We are glad you are here and just by showing up you are already demonstrating your very own smarts. Today I am pleased to present two highly accomplished music execs whose work mostly takes place in the trenches, meaning in the studio. We will do a deep dive on how AI is impacting their businesses and learn more about what they are up to both now and in the future. And I promise you will leave here smarter and more enlightened than you arrived.

Tom: Before we get started, please let me take care of some business. First, to the audience, please feel free to introduce yourselves in the chat window. The reason we do these webinars is twofold. First, we want to showcase really smart people and the amazing work they do day-to-day in the music industry. But the second reason is a bit more nuanced. Many of you know that I am a music industry headhunter. I run the music practice at Turnkey ZRG and I place music executives in roles throughout the industry. So, by definition and function, I help people connect with companies. In this series, my goal is to help you make more connections, and I invite you to take full advantage of that opportunity.

Tom: Specifically, I invite you to engage with the speakers and the other attendees in the chat of the Zoom. Please introduce yourselves, share your LinkedIn profile, say hello to your friends, and make some new ones, and ask questions of our speakers. Also, this is very important. Please make sure your chat is set to address everyone, not just the host and speakers.

Tom: I want to thank our program sponsors, for without their support we could not keep this free. Special thanks to MedJet, Turnkey ZRG, the Tennessee Entertainment Commission, Tennessee Brew Works, and Better Than Booze.

Tom: So, let’s get down to business. Today, we welcome Rich Keller. He’s a highly skilled and experienced audio engineer known for his ability to bring out the best in his artists’ performances. With a deep understanding of the technical aspects of recording and mixing, Rich is able to create a polished and professional sound that helps to elevate music to the next level. He’s worked with a wide range of successful rap and hip-hop artists including DMX, Swizz Beatz, De La Soul, Nipsey Hussle, Nas, Miles Davis, Alicia Keys, Lil Wayne, Mariah Carey, Method Man, Jadakiss, The LOX, Snoop Dogg, Chris Brown, Rick Ross, Ja Rule, and many others. Wow, that is a Hall of Fame menu right there, Rich. Incredible.

Tom: Rich’s career spans the breadth of hip-hop since 1992, earning him an official OG status. Welcome, Rich. It’s a pleasure to have you here today.

Rich: Hey, Tom. How are you? Thanks for having me, man.

Tom: Awesome.

Rich: I get tired just listening to the list.

Tom: And also joining is David Ronan. He is founder and CEO of RoEx, a pioneering music tech company reshaping the creative industry with AI-powered mixing and mastering solutions. His mission is to democratize audio production, making professional quality audio accessible to creators of all levels. A leader in AI music technology, David previously served as head of research at AI Music, where he spearheaded the development of Infinite Music, an adaptive AI music generation system that produced millions of tracks across diverse styles. This innovation contributed to AI Music’s acquisition by Apple in 2022. David holds a PhD in electronic engineering and computer science from Queen Mary University in London, where his research focused on intelligent audio production tools, including fully autonomous multi-track mixing systems. His expertise spans digital signal processing, machine learning, and software engineering with previous roles at Native Instruments, Northrop Grumman, and some other company I can’t pronounce. My bad. Passionate about the intersection of creativity and technology, David is committed to empowering music creators and redefining the future of audio production. It is my pleasure to welcome these two rock stars to our platform today. Take it away, boys.

David: Nice one. Thanks for having us, Tom.

Rich: Yeah.

David: Cheers. All right, Rich. How are you?

Rich: I’m doing all right. It’s been a minute, but good to see you again, David. Yeah, we actually haven’t spoken in a while. So, it’s always good to catch up.

David: Exactly. Yeah. I guess it’d probably be good to discuss how we know each other. Do you want to kick off with that?

Rich: Yeah. You know, I think it was all on my side. But before I get into all that, you know, I must say there’s no pressure of having a title like “Smartest Guy in the Room” and then have you introduced with all your PhDs and this and that, and I’m just like, wait, I just mix hip-hop, you know, and make music. You’ve got the list of… So, you know, anyway, I’ll try to… the older I get, the more I know I don’t know. So it’s getting harder and harder to be smart, you know.

Rich: But anyway, that all aside, yeah, I’ve been mixing for over 30 years now, mixing records for hip-hop, for Def Jam and Sony and UMG, all the big labels. And through that I’ve always evolved with the technology and I’ve always been part of the next thing. You know, I started by playing bass professionally here when I got to New York and then I evolved into programming synthesizers when the DX7 hit the scene with the big analog synths. And then it was Digidesign, which is now Avid, right - it was first, and it was just the stereo editor.

Rich: It was 1992 and my buddy Adam was using it to take curses out of hip-hop songs, and that was my first introduction to digital. Then we got our first album to mix through him, and we synced up a 1-inch 16-track with DBX Type 1, which was once again new technology. We had a Mac SE running Vision synced up through the 1-inch and through an automated board - another more technology. So every step of my career, and that’s just the beginning of it. I could go on and on all the way through now. I mean, that’s ’92. I could take you through the museum, the evolution of audio technology.

Rich: But let’s fast forward from then to like a few years ago. I started - I guess maybe around 2020, 2021, something like that. When Apple launched spatial audio, I was able to be involved in a lot of projects for that, you know, as it was initially brought out to the public. And quick enough we found we needed some tools to separate audio - if you didn’t have the multitrack, how are you going to… so anyway, this is where I jumped in and said, hey, let’s look at all these platforms. And there were only a couple at the beginning that could do it - that could separate stems or any portion of music, remove the reverb or something like that. Basically using AI to manipulate sound. And that was my first interface with AI.

Rich: And then it was my understanding - once I saw that it was getting better and better - I realized, wow, why can’t I… and the LLMs broke and I saw GPT 3.0. I was like, mind-blowing.

David: Yeah, yeah, yeah.

Rich: Watershed moment.

David: I mean, the whole world, right? We all kind of went, “What is going on here?”

Rich: But in a nutshell, so that tip, that’s where I dipped my toe into AI. That’s what brought me ultimately to David, because I wanted to bring - I wanted to make an AI of me. I wanted a Rich Keller AI mixer. I wanted to train it, grow the baby, and have it be - make my secret sauce on mixes. And through a project that I was doing - the De La Soul project for Reservoir Media - and the guys obviously in the band…

Rich: I worked with Jessica at AudioShake, the CEO, and she’s a great person.

David: Yeah, isn’t she? She’s brilliant. Yeah. She’s awesome.

Rich: Yeah, she’s very sincere, very helpful, and she’s got a great product. So, I asked her about this - if she wanted to get involved and build something like that. And she goes, “You know what? I got the guy for you.”

David: Yeah.

Rich: David, this is David’s lane. He’s all about this and he’s the best at it. So, here you guys go. Go ahead and play. And that was it. That’s when we met.

David: Yeah, I remember, because we met at the AES show in New York in person and we went to Flux Studios, Fab’s place, basically. And we just kind of hung out for a minute.

Rich: Yeah. That too. That was the real work that we did.

David: Yeah. We basically discussed how we could work together and figure out how we could build a system that would model the work that you’ve done essentially, and the good mixing you can do. And yeah, I guess the rest is history. I’d be interested to know though, like, because you’ve obviously got a long history in music production and AI is obviously the new thing now, but other than source separation, how else have you been using AI in your workflow? And I’d be interested to know as well- all the artists you’re working with now, how are they using it? Are they using it? Are they scared of it?

Rich: The productions I’ve been dealing with - for example, the Godfather of Harlem TV show. We had some issues there. The producers used AI for just some of the samples, just pieces. Instead of sampling from a record you’ve got to pay for, create a sample with AI and pop it in. Use it like that, which is one big use case that’s going on now. And a lot of the music companies are jumping on that format because it’s not really creating a new song per se, it’s just creating a sound, right? So that gets away from the whole copyright issues.

Rich: But that aside, as samples - now, one way that I’ve found I’m using what’s generated by the AI: there are a few well-known producers that are using AI as a tool to write a track. So, they’ll evolve the production and evolve the songwriting inside either Suno or Udio. Most of the guys I know use Udio.

David: Really interesting.

Rich: And what you get is a track that has a flavor, it has a sound, but somehow it’s still lacking in certain elements, maybe, or the sounds aren’t quite as good, but the vibe- it’s the demo. It’s the rough. And my career has always been chasing the rough. Like, I’ve been chasing the rough with everybody. Everybody says, “I want Rich - this is the rough mix. I want it to sound like a record, but I want it to sound like my rough mix. I want the same precision and the same energy and the same presence.”

Rich: So, what I’ve been doing- the producers are replaying the track. So the AI will be used for composition and developing some of the production, and whatever is good about the production… once they recreate the production over the rough, I’m still finding I have to go in and isolate things like the bass sound. The rep is just not good. So, I’ll have to go in there and stem separate it out and bring that in. So, I’m actually pulling elements out of the rough just like I would in a rough mix.

Rich: I’ve done many times where people have brought me the multitrack. They got a rough on cassette and then multitrack on the disc, and they lost the snare drum. So they replaced it with one that was really close and they’re telling me it’s the same drum. I’m like, “Yeah, bro. It’s not.” Don’t gaslight me. “You can tell Mariah that it’s the same drum.” And even she said, “It’s not the same. What’s different?” I’m like, and she’s looking at me and I’m looking at you like, “Bro, that’s not the same drum.”

Rich: So, I wound up in that instance, I had to actually sample it off the cassette at one point where there’s a drop where the snare played alone and I flew it into the record. And it was one of those save-the-day kind of moments, you know. But anyway, I digress.

Rich: So, different ways that I use the AI in my work - sometimes I’ll steal some sounds. Let’s put it that way. I’ll steal sounds that the AI has created that are so perfect and right. I’ll just use it and trigger it.

David: So you’re using it very much in an assistive way.

Rich: Yes.

David: Not in a replacement way. And I guess that’s probably the best way to use this for now.

Rich: I mean, I have a lot of the plugins - I have three different AI plugins. I’m not sure if they’re really AI or what - you know, that term. I mean, actually, that’s the stuff I wanted to ask you about. Could you talk about it - just give us a little breakdown? Because there’s AI, and then there’s neural nets, and then there’s learning, and then there’s… but some things are not learned. I don’t understand the difference. I’m confused when I hear AI. What does it really mean? Does it learn? Explain how different levels of AI work, and I’ll shut up.

David: Yeah. I mean, AI is kind of like the big super term - the umbrella term, should I say. And then you kind of have machine learning, which is where it’s actually trained on data essentially. You have big data sets and you have an algorithm that learns the distribution of the data and it can make predictions on it. You can give it something and it can tell you the difference between certain things - if it’s a cat or if it’s a dog, images, that kind of thing.

Rich: Is that what an LLM is? Is that like ChatGPT?

David: No. So that’s kind of generative AI. That’s like the evolution of it. When AI started becoming a thing, say 15 years ago in academia, before it became mainstream, it was very much using AI to classify things. Like in self-driving cars, they’d be like, “That’s a stop sign, that’s a pedestrian, don’t hit that.” But generative AI is what we’re seeing now, where it can generate something out of nothing. You can get ChatGPT to go, “OK, write me a story about a mix engineer who does hip-hop,” and it’ll give you something.

Rich: But is that learning? Is it learning from us? Like, when I type in ChatGPT, is my input becoming fused in the smoke up there?

David: Yes. Some of the models do that. I mean, an LLM is just like an autocomplete on steroids, basically. It’s trained on so much data it can kind of predict, based on what you’ve written, what’s going to happen next. But then you have reinforcement learning, where essentially it is learning from you. You go, “OK, tell me a story about whatever,” and you go, “OK, I really like that,” and it’s learning from that as well.

Rich: So clearly there are different types of AI that differentiate - they’re not always learning. I guess the assumption I had, maybe wrongly so, was that it’s always learning, always soaking up knowledge.

David: Not necessarily. No. Some of them are just trained already and that’s it. They just do their job. And then if that person who created it once gets a load more data, they could retrain it and make it better. And that might be something because they’ve learned stuff from, say, a user of an app or whatever. You can do that. Essentially, that’s how a lot of the stuff works, I would say.

Rich: Well, thanks for explaining that. Clarifies it for me. It’s always a big kind of black box, you know.

David: Yeah. I guess when it’s applied to audio though, it’s a bit trickier. Well, audio and music - because music is just so subjective and there’s emotion. It’s funny, when I first started doing all this AI machine learning stuff, we were analyzing the signals with mathematics and feeding that to an algorithm to make predictions. And then we started actually feeding it images of sound, and then it was making decisions on that, which I thought was quite interesting.

Rich: What’s an image of sound?

David: Like a spectrogram.

Rich: Oh, like the 3D with the ridges. Yeah. OK.

David: So that’s kind of how it’s evolved, which I think is really interesting. And the thing with these machine learning models is, the more images you have, the better it learns. They’re very data hungry. And these LLMs are especially very data hungry and very expensive to run and have to be on special hardware. I don’t know how true this is, but apparently every ChatGPT question you ask uses a gallon of water to keep the servers cool. That’s obviously energy as well. So they’re hungry.

David: It’s just mad how it’s going. You can see now that Meta are building their own power stations and putting data centers underwater to keep them cool, and they’re trying to find more ways to make this more efficient. And the other thing is how big the models are. If it’s a big model trained on loads of stuff, you have to have it on lots of computing clusters, and they’re trying to make them smaller. A lot of it is having the firepower - i.e., having the money - to be able to train these things. That’s why big tech are kind of running away with it.

David: But what was interesting was - I don’t know if you remember a few months ago - there was this Chinese startup called DeepSeek, and they were able to train a really competitive, really good model. I think it only took them like $2 million or something - I know, “only” $2 million - but vastly cheaper than what Meta and Google were doing. And I think that was a big moment.

Rich: Didn’t they like piggyback? Didn’t they actually plug into ChatGPT and somehow train off of that? That’s kind of what I heard.

David: They did. Yeah. I think that was the sneaky thing.

Rich: Sneaky thing. Yeah. You know, I think of AI as like this smoke cloud, the dense fog, right? Out of which things drop - little words drop. And the same with sound, picture. The picture is going crazy. I just saw there are models you can download from GitHub that blow away the V3, the new Gemini 2.5 video with the speech, which I tried and it’s amazing, and this blows it away and you can download it and use it on your laptop.

David: Yeah. I mean, the pace at which it’s moving is just impossible to keep up. Every day - I’m subscribed to this letter called TLDR which basically breaks down all the tech news every day, and I’m like, I don’t know if I could look at it anymore. You know what I mean? It’s just - I’m trying to do what I’m trying to do, essentially, and keep up. It’s a struggle.

David: But what is interesting is there are a lot of AI companies out there that claim to be AI, but they’re just built on ChatGPT. There’s nothing wrong with that - a business is a business at the end of the day - but some of them claiming that they’ve built it themselves, which is quite interesting.

Rich: No copyright for the code. I mean, don’t they have to be plugged into ChatGPT?

David: Well, this is it - if ChatGPT suddenly went down, their business goes down. So they have a dependency. There’s a lot of that, which is fine, but you are kind of beholden then to the big tech giants. You don’t stand on your own two feet. Which I will proudly say we do - we don’t have any dependency on OpenAI because all our tech is built on stuff from my PhD, basically. All built from the ground up, which gives us a unique advantage - we’re years ahead of all the competition and it’s very unique IP, specific to mixing.

Rich: Yeah. So let’s back up. Tell us about that transition. The first thing I heard in your bio that I’d heard from you before but we never really talked about was the company that you sold off to Apple. You were part of the team there.

David: Yeah. I wasn’t one of the founders, but I was in the founding team, basically.

Rich: So you started it with them. But what did that product actually do? And when was that - 2019, 2020?

David: 2022 was when Apple bought it, basically. But what we were doing was creating music with AI, but before the Sunos and the Udios of the world came along, because the technology was different then. We had a team of music producers who were creating music samples and loops and all this kind of good stuff for us - created by people that sounded good and was really well produced. What we were doing was taking those different elements and stitching them together to create new songs out of them.

David: Essentially, the way it would work is you could go, “Give me a hip-hop track that’s 92 BPM, the key of A minor, and I want the drop to be at 30 seconds.” And it could do that and give you different variations of it. Then you could go, “OK, extend it to a minute,” and it could do that because we had the fundamental building blocks of the sound, and we had the machine learning and AI basically put them together in a harmonically and musically sensible way. So it sounded good.

David: And the beauty of it was it retained that human element. I think at the time we were the only people doing that. Everyone else was using AI to learn composition rules and then feed it to MIDI. And it didn’t really have -musically some of it sounded fine, but it just didn’t have the human feel to it. And that’s where we stood out.

David: Our business model was basically to sell it for advertising. You want to create an ad for Spotify or whatever, but you want to license a song - you have to pay for the rights and it’s complicated and takes time. With this, you go, boom, I have an ad. The ad is 90 seconds. This is what it’s about. Here’s the spec. And we basically generate the track instantly. So yeah, we were doing that for four and a half years. It was right coming up to the last year where we’d nailed it, basically. But then obviously Apple swooped in.

Rich: Swooped in. Like a seagull at the beach with the kid’s hot dog, right?

David: Blew away with it. Yeah, but it was successful. The great thing about it was I worked with a really good team and I learned loads doing that. The founder had previously exited already as well. So I was with a good team, and that allowed me to start RoEx, which was the beauty of it - because I had built up four and a half years of startup experience and I had some money, basically. So yeah, I went back to the university and said, “Let’s spin out and commercialize my PhD.”

Rich: Was that your PhD then?

David: Yeah. So basically, at the end of my PhD, I had some intellectual property around multi-track mixing. But the problem was it was absolutely useless because you’d need to run it in a lab on a computing cluster. You could only mix four tracks. They had to be in mono.

Rich: That was like the ’50s with analog.

David: Yeah. And it would take a day to get you a result. So I was like, this is absolutely useless. I had to figure out how to turn that into a product. There was a lot of just hard work getting it to run on a server where someone could come along, upload their recordings, push a button, provide some very basic mixing preferences, and get a mix and master track in minutes.

Rich: That was already functional when I did the demo and I was sold, you know.

David: But the thing is, I’m not a professional mix engineer. So it was yourself and Anegeliki - Angeliki is our lead R&D engineer - you were the practitioners who really understood it, who were able to help me shape it and make it a lot more professional sounding, a lot more realistic sounding. And I think that was the beauty of the tipping point when you guys stepped in, obviously. But it took a while. It took a long time to get it to the point where it could even be shaped like that. Agile enough to even respond to a poke.

Rich: Yeah, exactly.

David: I mean, up until only recently it’s still been quite slow, because when you’re dealing with trying to mix 32 tracks of high-definition audio, it takes a lot of processing, as you can imagine. And we’re solving that now - we have a system that’ll do it a lot faster. But yeah, it has been challenging for sure to get it out there.

David: One of the reasons being that a lot of people might not necessarily know the distinction between mixing and mastering, right? Because there are a lot of mastering companies out there already. Obviously LANDR, who set the scene for that with their AI mastering back in 2013. They’ve kind of set the scene and there’s been a lot of companies who’ve replicated that business model. I think a lot of people when we came out were like, “Oh, you’re just like LANDR.” And we’re like, “Well, we’re not, actually. We’re doing multitrack mixing.”

David: LANDR have done an amazing job and they’ve gone a very different path to us. They’ve gone fully into music - they do plugins now, they do samples, they’ve gone horizontal. Where we’re going with it is we’re applied to music right now and it works really well for music, and we love music. We will keep doing music. But the tech underneath has applications in film and TV, post-production, video games - and that’s where we’re taking it next, essentially.

Rich: Powerful audio tools, you know.

David: Yeah, this is it - make people like yourself be able to work faster. Going back to AI being assistive, essentially. The idea is that you’ll be in a DAW and you’ll be able to push a button and it’ll do all the corrective mixing - the rolling off of certain frequencies, the stuff that’s not fun. It’s not creative. You’re just doing it to get to the point where you can then do your creative mixing. And that’s kind of where we’re going with it.

Rich: Yeah. And it’s great. We talked about doing the… the style transfer. I don’t know why I’ve got a block on that one. The style transfer of actually having name engineers - starting with me and my friends and all that - to train models. But aside from that path, which we’re still working on, there’s the mastering element of RoEx and then there’s the multitrack mixing element.

Rich: So, without doing a whole commercial for RoEx Audio - but that’s really the reason we’re together, more than just friends. And you graciously offered me an advisory position. So, I’m happy to be on board, in case people didn’t realize that we’re in cahoots.

Rich: I’m just curious to understand even more about the platform, from the back end. We never get to talk about this in depth. Is there a difference between what you’re doing in the mastering as a two-channel master versus what you’re doing in multitrack situations? I’m sure there’s some crossover.

David: Yeah, good question. Without going into the depth and giving the secret sauce in public - the system that sits behind the mixing, the mastering, the decisions on how to apply EQ and compression and all that kind of good stuff, are quite similar. But you’ve got to remember that with mastering, you’ve got a stereo track. We don’t have the individual tracks. So therein is where we kind of have some magic to basically figure out how to master it effectively to make it sound good. We optimize it for punch and clarity, and obviously get it to a certain level.

Rich: There’s a different intention that specifically has to do with the limitation of the source file and the next steps you’ll go to address those limitations.

David: Pretty much. It’s different where it needs to be.

Rich: But that’s also a big part of the business model right now, isn’t it? The mastering.

David: Yeah. We work with United Masters, who’ve been really great to work with. All their stuff gets passed through to us for mastering.

Rich: Nice.

David: Yeah, we’re working with Music AI as well. They do our multitrack mixing. So it’s going well. We’ve got customers who really like it. And then we’ve got the B2C business as well that’s going really well - it’s growing month on month, which is awesome.

David: I think for us, to be in the DAW is where we want to be. We’ve done some kind of lightweight DAW integrations and we’ve got more to come. The DAW is the center of gravity - that’s where you’re creating. That’s where you need to be. Having a button in the DAW that can suddenly get you to 90% is kind of the best thing out really, because you’re suddenly able to do 30 mixes a week as opposed to 10. You’re being paid per mix.

Rich: Yeah. Exactly. But for DAW integration, the holy grail there is really about - just like any studio gear, there’s only a few spots in that rack. You’ve got your LA-2A, you’ve got your classics that have been there forever. And it’s funny that actually some gear from the old days is still the most sought-after sound, because they solve very basic problems. The LA-2A - the opto compressor. And we did the Pultec because of you, actually - that was your idea. We’ve modeled both the LA-2A and Pultec because we know they’re the go-to, they solve problems, they work so well, and they’ll be around, I don’t know, forever.

David: Have to work on those EQ curves, those compression curves - the knee, the opto setting, the attack and release are perfect.

David: Yeah. The other thing I guess you’ll be able to talk a lot about is immersive formats. That’s something we’ve been looking at - we’re very keen to get involved with that a bit further down the line. But what’s interesting - this happened very recently.

Rich: What’s that?

David: I think we’re about to see a spatial format war. You have obviously Dolby Atmos, which has been around for a while now. Google have just launched the Eclipse format, which is completely open source. Anyone can build it - well, that’s what we’re going to build on because we don’t have to pay a license for it. Why not? And you can basically listen to that content on YouTube, which is one of the biggest music platforms in the world. But then Apple have also just released their own native format - relatively quietly. They didn’t make a big hullabaloo about it.

Rich: I saw an article. Yeah. So it’s interesting. There’s all these formats coming out now in the spatial/immersive formats.

David: What I like about the Google one is because it’s free, it’s kind of like Switzerland. Anyone can use it. With Dolby Atmos, it’s kind of “cinema down” - you have to pay, it’s expensive. With Eclipse, it’s “creator up.” Anyone can integrate it. Which I think is awesome. And that’s what’s really drawn us towards it, and we’ve been talking to them about it.

Rich: So what would that entail though?

David: So we would just be serving content, just like we do to Apple now. And those who use that format - Tidal and Amazon and Deezer also use the Dolby Atmos format. So we’re going to - yeah, it needs to be adopted. I mean, I haven’t seen anyone adopt yet other than YouTube, but it’s early days.

David: I know that in Chrome, in Google Chrome, you’ll be able to - you won’t need - you’ll be able to listen to stuff through it basically. In our Automix app, if we create an Eclipse mix, you won’t need to download it and listen to it through a renderer - you can just do it in the browser. Which I think is really powerful as well.

Rich: Will it be - I’m assuming it’ll have to be encoded, or will it just translate from the Dolby ADM format like Apple… I mean, we don’t know. It’s early, right? It’s early days.

David: I think it’s the Wild West out there, is what I think.

Rich: Oh, 100%. And it’s ripe for disruption in terms of assistive tools as well. Because there’s no rules for immersive. I mean, you know better than me - tell me if I’m wrong - but there isn’t really defined rules for mixing immersive like there is for stereo. You know - bass goes down the center, kick down the center.

Rich: UMG gave us some guidelines. We had a few packets over the years of like, “OK, here ye, here ye, from henceforth, this is how we will be treating Atmos on a rock mix.” The guitars must remain in the front third. The vocal will not be dead center solo - it’ll be split. Because when you accidentally play an ADM file, a source file from your computer, the center channel will be soloed. You could therefore steal a vocal. So this is what they were worried about. We had to have a rule where the vocal had to be multitrack. So you’d never have a solo vocal.

Rich: But the formats now - Sony has its own, the 360 Reality Audio, and then they have the room modeler now for the headphones.

David: It’s amazing. I tried it. I was there with you, I remember.

Rich: Yeah. That is mind-blowing stuff. I just saw June from the Sony team - he’s the head of that team. And we’re going to be getting together actually soon, have a little dinner, have a chat. There’s also - Sony gave NYU $7 million to develop a class, a seminar class, for the future of music, the future of audio. So I’m going to be involved in that, talking to students. We’re all figuring it out for ourselves here. Crazy ride, man.

Rich: Actually, I want to address something here. Liv Carter asked about copyrighted music - has anyone begun to solve for the issue of copyright in AI music? I can give you a little information on that, Liv.

Rich: From my understanding, what I’ve been told by people with a lot of letters after their names, is that as long as the models are trained with the actual metadata - the rights, all the credits due, everything about the song - if you train a model and include the metadata, it is possible to get a feedback out of what’s been used, what’s gone through the meat grinder. It can tell you - it might be a micro percentage of a million different songs or 100,000 - but it’ll parse it out for you of what influenced this, and you might have to have thresholds of degrees. It’s in the works and people are wrestling with that.

David: I mean, they have to solve this.

Rich: Yeah, they will. I mean, come on. We’ve got Bitcoin and the blockchain, right? So there are ways to lock things up and give things a unique identifier and make it all true - at least until quantum computing happens, right?

Rich: And will the copyright holder actually get paid? Well, if they’re within the system of the majors - as well as the majors pay artists now, or Spotify with their streaming pennies - it’ll just be another trickle of pennies. Because the labels are - it looks like the major labels are suing Suno and Udio for training on copyrighted material. But now it seems the argument is swaying - they’re not going to shut them down. They’re going to do what they did with Spotify, which is basically cut a deal, take a percentage, and take a rev share.

David: Ride off into the sunset with money bags.

Rich: Exactly. As they always do - just dilute and squeeze the artists constantly. That’s what they do. And I hate to say I’m part of that landscape, that ecosystem in the music industry. Sometimes I wish there were better ways to do it. And there are some people I’m talking to that are dreaming up better ways, but it’s too early to comment. The great dilution is happening. Everybody can make music at a whim and discard it at a whim.

David: Yeah. I mean, it’s so bloody easy to make music these days.

Rich: Right.

David: You’ve got - I mean, you’ve solved a lot of the problem. I would say they’re “problems,” but a lot of the steps involved are a lot easier than they were 5, 10 years ago. Obviously mixing and mastering, we do that. But like getting samples - I don’t have much time to make music anymore, but I used to be really bad at drum programming. It’s just my thing. I just really sucked at it. But now you can just get it to generate stuff for you, and then take it in, and away you go.

Rich: Right. It’s like - dial, hit the spinner to taste, it lands, keep going. But we’ve been doing that since digital plugins came out and they had presets from the engineers - the great engineers put in their presets. You could just click through them and boom, you got something you like. That’s one step already of not really understanding why. Someone else’s hard-won experience is in there - usually this microphone has these curves and this is the most pleasing way to present that curve, and we’ll put it in a preset.

Rich: So I’m clicking through it and I’m not learning the knowledge - I’m benefiting from the knowledge and experience of someone else before me, but I’m not learning the knowledge. But I put in my 10,000 hours or more, and I’ve learned how to - all right, well, let’s see what’s actually happening. I’ve developed my own stamp of what I think things should sound like that are more personal. All the AI stuff is just so very generic and it fits right in the circle.

David: Well said. I think the beauty of it is - if you’re part of the creative journey being an artist, and you can take little bits of it to make you faster and get to where you want to be faster, that’s great. That’s perfect.

David: But I guess with Suno you can just generate a song with a prompt and it’s instant gratification. But there’s no fun in that. It’s fun for about 10 minutes when you make a song about your dog eating too much butter or something stupid.

Rich: Right. Exactly.

David: Which, speaking of that - that’s what I was talking about earlier. The serious creatives are using it, but it can only take them so far. Which thankfully the platforms are recognizing that and saying - look, YouTube just said recently, a few days ago, they’re not paying any royalties on 100% AI music.

Rich: So it has to have significant human contribution. Of course, how are they going to tell that? Well, they do have AI threshold meters. I think it has to pass a certain threshold. I’m not sure. It’s software stuff. To be able to distinguish AI from non-AI music, and I think some of it has to do with the fidelity of the sound because of what it’s trained on.

David: I’m glad you said it. Because Spotify had this API that you could ping and you could just get anything from their whole catalog. I think they’ve stopped allowing people to use it now. But basically, you could download 30-second snippets of basically all their catalog - as long as you had a list of the top 100 songs in each decade for each genre - and you could get a 30-second snippet. But it’s all MP3s, so it probably wouldn’t have been high fidelity.

Rich: 96 kbps MP3.

David: Yeah, exactly. And that’s what it’s trained on. So I think a lot of these AI music detectors are just really good MP3 detectors. I’m sure some of them are a lot more than that, but some that I’ve seen have claimed to have 100% success rate. I’m just like - it’s 100% successful detecting it as an MP3.

Rich: Right. And something happens when it goes through the meat grinder into AI and comes back out. There’s a certain signature to the lameness of the sound, especially present in all instruments.

David: Well, it’s the MP3 encoding. Because MP3 removes frequencies that we can’t hear - that’s why you can shrink a WAV down to MP3. It just removes the stuff you can’t really hear anyway. And that’s a sonic signature in itself. We can’t hear it, but the algorithms can see it. Go back to the pictures of sound - that’s where it’s easy.

Rich: Wow. We’re full circle now.

David: All the way around. That was pretty good.

Rich: Let’s see - there aren’t any international laws protecting creatives until… Yeah, Glenn. It’s the wild west, man. The dust is flying and rattlesnakes are everywhere. But there’s also gold in them there hills.

Rich: For me, it just feels like a big dilution. Algorithms are keeping everything the same. On my Spotify, I like to listen to 70s and 80s - actually, mostly 70s. I find that’s when I was a little kid, so that stuff has the most filled memories, before I became a teenager and all that.

David: Before the angst.

Rich: Yeah. Before the angst set in. Back when I was happy and everything was all right. And the older I get, I’m getting back more to happy.

David: Which is good to see, definitely. So, I wanted to ask you, David - before I know we’re getting down to our last quarter hour here - I know your job is your life and it keeps you busy. You’ve got a wonderful young family, which is a beautiful thing. But I want to ask you - what drives you? What’s the source of the passion for you?

David: That’s a good question. I mean, I’ve always loved music. Making music since about 15, if not younger.

Rich: Doing what?

David: Mostly electronic stuff. I love electronic music - techno, jungle, IDM, all this kind of stuff. Warp Records - there we go, my favorite label. For those who know in the chat, give us a thumbs up. But yeah, very experimental electronic music. And then I was always very technical growing up. I’ve been programming computers since about nine.

David: I went on to study math and computer science. Computer science was fine, I liked it. The math is kind of boring. But I wanted to combine those two things. So I ended up doing a master’s in music technology, and it was a year of just fusing those two things. And I was like, “Oh my God, this is my… two things I’m really good at and interested in.” And I’ve just always been building things and I love music. I couldn’t see myself building anything else or working in any other field, because I’ve been in it so long. I’ve been making music so bloody long - why would I want to do anything else?

Rich: So what I’m gathering is you’ve never considered the straight world, the corporate world - work for Amazon, or work for Google?

David: Do music with them? That would be fun. Yeah. But yes - in that way, we have a lot in common. You’re just a freelance thinker and entrepreneur.

Rich: In the same way, I’ve just never considered a career in anything other than what I do. It’s just naturally - I just flop along and float along, and things start to happen. When I was a younger man, I thought I was seeing the future. But I’m trying to do that now, and I don’t see that same future. I could tell what was going on back then, but now I’m a little less certain and maybe not as optimistic as I was when I was younger. But I’m not cynical yet. I believe in hope.

David: Well, I guess you’re kind of like me - most of the time when you’re at work, you’re enjoying it. There’s obviously things in every job you don’t like. But I remember - on a Sunday you’d have dread because you have to go to school in the morning. But when I started working at AI Music, that was my first job after my PhD in music tech - I was actually kind of like, “Oh God, I get to go in Monday and do loads of crazy stuff with AI and music.” I can’t wait. All my friends are there like, “God, got to go in and do Excel sheets.”

Rich: Exactly.

David: That’s why I do it. I couldn’t see myself - even if I won the bloody lottery, I don’t think I’d give up what I’m doing.

Rich: Give it to the wife, pay the bills, have fun, buy a stack of t-shirts, right? I’d build a big studio, that’s for sure. But I’d still be doing music and audio stuff. I’d be building. I love it, like yourself.

David: Yeah, man.

Rich: For me, it’s really the transition - trying to create the moments. I’ve had my own moments because I write and I sing, I do my own thing. But I’ve also been part of the moments. A lot of friends, old friends, would be like, “Rich, you work with rappers. You’re a musician. You grew up playing jazz and rock and Latin and classical - and now you’re with the lowest form of musical music.” I’m like, no, I don’t see it that way at all. I’ve had the opportunity to be with some incredible artists.

Rich: DMX is, for me, the…

David: He was your favorite artist to work with, would you say?

Rich: Yeah. And I spent seven years with the guy. I got close to him as well, and the whole team. It was really like a big family. And once the success came, we all just lived it. We loved it. It was a great ride.

Rich: But the point I wanted to make is that - so many moments with so many different artists. What I go for, what I live for, are those magic moments where the first time - the song may be built in sections or whatever - but when that magic is happening, and the thing that makes the hit, the huge hit, beloved by everyone - when you see that being born and you’re there, you’re part of it, you’re touching it, you’re helping it be born - that moment of creation feels like a spark.

Rich: And that’s the number one, that’s the highest goal. But you don’t get that all the time. So you have to do things to support it. It’s like the gold on the top of the pyramid, but you have to build the pyramid all the time too.

Rich: And that’s part of what’s missing in the AI. Just as I said before - I can go through my engineer presets and there’s stuff in there from the guys that did Zeppelin and the Beatles and all that. I’m standing on their shoulders, but I’m not gaining the benefit of their experience of how they came up with that. And in the same way with AI - as crafty as it is and as beautiful as it can be, because it can do some magical stuff too - I have thousands of songs I’ve made, but I’ve only downloaded one or two.

Rich: So that tells me something. It’s like I like the tricks it does. It’s kind of like doing your Instagram. Pow pow pow pow pow. Not long-lasting. Instant gratification with just a big falloff.

David: Yeah. A big falloff.

Rich: And you don’t get that magic. So I’m curious what their churn rates are like.

David: Yeah.

Rich: It’s got to be nuts. It’s certainly - I’m using more energy than the $10 a month I’m paying for it. But that’s the spark in AI, where the envelope I wanted to push - and we will continue and we will succeed with the style transfer at RoEx Audio, little plug - is to get that personality to happen again. Let’s get to that point.

Rich: Because honestly, the best thing, my favorite interface with any AI, is talking to ChatGPT. And now she’s starting - I say “she” - she’s got the Australian female voice, I like that the best. And it’s starting to learn and remember. And it told me, “I’m starting to learn and to remember now.” And I told it, “I want smooth, casual.” And it always kind of sounds like it’s laughing, but it’s starting to feel really familiar. And I’m just like, “Wow.”

Rich: And I can’t wait for the day when I can actually tell it - “Listen, I need a 60-piece orchestra, OK? And I’m going to sing you the melodies. I want you to do the harmonies like Chopin would, and the strings like Brahms would. The arrangements. We’ll start there. Here’s the theme. And I want you to switch keys three times. Please go up.” To be able to speak to it like a musician and just be like, “All right, here you go. Add cadences now at all the correct points where Beethoven would put them.” I think that’s very close.

David: Really? Does that take the fun out of it though? Do you not think that’s making it - 

Rich: Well, I think it’s going to come all the way back around. Because right now it feels like a Lego erector set. My buddy Dan just sent me yesterday - the best chess player in the world just creamed ChatGPT 4.5, and he didn’t lose a single piece. It couldn’t get one piece off of him. And we’re like, “Wait a minute.” And then I’m doing simple research in ChatGPT and it’s hallucinating. The links go to nothing. I’m like, “What?” So I feel like there’s a little bit of hype going on.

David: There’s quite a bit of hype. The “PhD level of intelligence”

Rich: PhD level. I don’t know about that.

David: Yeah.

Rich: So, don’t believe the hype. But definitely use the hell out of the tools. And I must say, I’ve been a proponent of that. I’ve already ridden one wave in with the stem separation and doing all that kind of stuff. That’s been a big part of my career - just another third wave. I’m well into my third decade of mixing.

Rich: People have asked me, “Rich, you think about retiring?” I look at that two ways. I’m always retired because I don’t do anything I don’t want to do, and I do whatever I want when I want to do it. So I don’t understand what retired means.

David: Fair enough.

Rich: I think for me it means when my wife stops working so we can just go and do anything together. And I’ll just do albums as they pop up because it’s fun. I still have a great time.

Rich: All right, cool. Nathaniel, everybody must be getting close to that time. I’ve been hogging the microphone. David, any final thoughts to share?

David: No, I mean, just excited for the future. Obviously I love working with you, and yeah, can’t wait to see where we take things. And thank you, Tom, for having us.

Tom: Guys, I can’t thank you enough for doing this. I am fascinated with AI. It’s everywhere in my business and probably everyone who’s listening’s business.

Tom: I quit fearing it about a year ago when somebody said, “AI is not going to replace you, Tom, but somebody using AI might replace you, Tom.” And that kind of sunk in for a moment. I was going, “Oh, I better get with the program and start figuring out how to use it, lest I get left in the dust.”

Tom: So, gentlemen, this has been awesome. Rich, I would love to hear some more war stories about some of the artists that you’ve worked with sometime. I regret we didn’t get much of that.

David: Oh, he’s got stories.

Rich: Just Google me. I would love some kiss and tell, so to speak. If you’ve got a different platform - one that goes on at night time, only after dark - then we can talk about that. But we’ve got to have Dave live in a pub. He’s closing out a pub and then we can have that conversation. But yeah, sure. I’d love to.

Tom: David, thank you very much for what you’re doing. Every day is different for you, I’m sure. And it’ll be really interesting to maybe revisit this conversation in six or 12 months to see what’s changed. I have a feeling a lot will.

David: Absolutely.

Tom: I told the audience you guys would leave more enlightened and inspired than you arrived. I think we delivered on that promise. So with that, I will say - be nice to each other. Go out and beat the heat somehow, someway. Take care, folks.

Rich: Bye bye.

David: Cheers.

[Transcript Ends]

David and Rich’s conversation touches on something that is important to RoEx: how AI works best when empowering creatives, rather than replacing them. And how putting a new set of powerful tools into the hands of the industry's best will open up new creative horizons.

You can listen to and watch the full podcast recording over at The Smartest People in the Room YouTube channel. Whilst you are there check out episodes with other guests that have featured on the show including Grammy-winning producer Albhy Galuten, Zomba Group co-founder Ralph Simon, music icon Gloria Gaynor, artist and songwriter David Lowery, and Audiomack product leader Chris Dalla Riva.

If this conversation gets you excited to try out assistive AI tools within your music production and creative workflows then check out Automix and Mix Check Studio by RoEx - they are free to try.