#294 Voice AI for Brands and Media with AudioStack CTO Peadar Coyle

0:00I think people sometimes underestimate the power of domain-specific knowledge and just how deep some of these problems are.

0:12Peadar Coyle:Hey everyone, and welcome to another episode of SlaterPod. To the end of the podcast, we welcome Peadar Coyle. He's the co-founder and CTO of language tech and voice CI platform AudioStack. I met Peadar for the first time earlier this year at a small voice CI meetup we hosted in London. and then he also joined a panel we hosted on Voice.AI at SlaterCon London together with Gradium and Poly.AI. So great to have you here, Petter, and thanks so much for joining. Thanks for having me, Florian. Where in the world are your coordinates from today, London or somewhere else? London. I'm calling from our shortage office near Spillafield in London.

0:49Peadar Coyle:That's where an AI company has to be based if it's in London. Yeah, of course, yeah. Cool. All right, so tell us a bit more about your background and kind of the journey to co-founding AudioStack? Sure. My background was in data science and data engineering. I worked for a number of startups. I worked for the likes of Amazon, Vodafone, largely on what we would refer to as big data projects. And then I gravitated towards machine learning. And, of course, I think we forget now in the modern era of artificial intelligence how transformative some of those techniques were. They look dated now because they've got large language models and foundational models, but they were very powerful at the time.

1:33They were the stating. I did a lot of stuff around natural language processing, which kind of led to my interest in things like text-to-speech. I met one of my co-founders via the Python community in London. So as a good piece of advice, go to meetups. You never know what could happen. and it could also help people get jobs and help people find co-founders. And we were very interested in the technological edge of what synthetic speech would do. But we were, I think, a little bit early in figuring out what our market was. So it took us a little bit of iteration. It also took quite a lot of research and development for the cutting edge to become good enough.

2:24and we've been going for nearly eight years now, seven and a half years. We have a number of large customers including iHeartMedia publicists and various other radio companies throughout the world.

2:40Peadar Coyle:So you started in 2019 and so walk us through some of the key milestones but also then follow up by telling our listeners the kind of expanded elevator pitch. So what's the key elevator pitch? Who are your clients? What are some of the use case? And we'll get to speak about those, of course, in more detail later on. But just kind of in a nutshell, a couple of minutes, explain what AudioStack does, because it took me a while. Would it be good to play a demo first? And then I can talk through the demo afterwards. That might be the best way to play this. So this is a demo from some work we did with the UK government.

3:20so let me just let me know if you can hear this and let me know it's only 30 seconds where's Ollie? he never misses a ciggy break sorry I'm late fellas list broken lift? we're on the first floor is smoking affecting your day to day? every cigarette takes up to 20 minutes off your life but fear not the power to quit is in your hands

3:593pm Sammy, you coming for a quick ciggy? Oh, OK. Erm, hmm, I really shouldn't.

4:14NHS Personal Quit Plan. With a quit date planner and advice on managing your cravings. You're more likely to quit with the right help. Search Smoke Free for your free plan and get the power today. So you should have heard there a couple of voices. So, you know, so we, that was all AI generated, right? So that is, you know, so our creative director wrote a script. He worked with the government as a client. and the government wanted to like evangelize and encourage people to quit smoking these are all very positive things right i think that's one of the things i want to get across you know this is not like a dystopian nightmare this is a very positive story um and we did that in individual accents throughout the uk because there's evidence that like the accent being localized has a higher ability to tell the story the soundtrack in the background was came through one of our sound partners we had our enrichment engine we have a whole dedicated r &d team who makes sure that sounds well we made sure i'd worked on various providers i believe it went live on the likes of spotify from podcasts and so that's what we did we enabled brands and we enable governments we enable like enterprises to tell their story um through the medium of audio and you know and beforehand that would have been cost prohibitive right so no nobody was going to like hire 30 like actors and go through all the challenges of doing it not to mention the retakes and stuff like that that would need to be done and that's kind of succinctly what we do.

5:52So that went out in how many how many accents? I think it went out between 20 to 30. It definitely went in different regional parts of the UK. There was definitely like Northern English, like Yorkshire accents, there was a liverpuddlean accent there was some kind of things in the south um i know also we're we've done similar things with like car companies you know so companies were trying to sell cars where we've done kind of localized stuff and even localized stuff based on different variants of of cars so you know honda in in edinburgh you know a different type etc in in different parts of the uk as well um and we've done that we've done that throughout germany we've done that throughout the US.

6:35And then in the US, another kind of thing that comes up is like multiple languages. I know you're based in Switzerland. There's always that like challenge of how many languages that you have to convey to convey a message to the largest possible audience. I think Switzerland has five official languages if I remember correctly.

6:53Peadar Coyle:We have four languages, but I was going to say like this would really work well in Switzerland because we have dialects, like kind of accents, but maybe even beyond accents. Like we have this, you know, Swiss German as kind of a half language of German, but then within Swiss German, there's like about 20 to 30 pretty distinctive dialects that people are using with like different words and things like that. So you could go really hyper local in Switzerland with something like that as well. Maybe not a ton of training data yet on them, but like that's something that could be solved, I guess. We can get more to like how we partner with voice providers and things like those kind of localization later on in this call.

7:33Peadar Coyle:So voice.ai is really in a boom right now. And we know we've really stepped up our coverage of it here at Slater, you know, not just on the content side, but also on the conferences. We, you know, get people onto speaking slots and panels, et cetera, really in a boom. So walk us through how you see the market right now, like the different levels from like kind of the infrastructure to models and then applications and use cases on top of it. I consider like voice models to be kind of like building blocks, you know, and so and we don't just partner with one because we've discovered that like we need the ability to cover a wide range of accents.

8:09Some providers are better at some accents than others. Some providers are better at, you know, more European centric. Some some, of course, also sometimes there's a bit of a race there in terms of like voice quality. You know, it's like, you know, like they're overtaking one another. and we're just constantly working with them and giving them feedback, the likes of Gradium, of course, who we met actually through the Slatter community, who recently came on as one of our partners, but we also recently partnered with likes of InWorld. And I believe we have between 10 and 12 that we have ongoing partnerships with and we're constantly evaluating.

8:48We even have a good kind of dedicated, dedicated kind of like operational function as such to evaluate these things, right? And we also have that on the music library side as well. There's a whole thing about whether we do generative music, whether we use that, or whether we partner with a more classical rights management company. Often our customers are leaning more towards the classical due to their experience, but I think there's also the chance that generative stuff becomes good enough. And we're also seeing some of the voice providers are moving into those spaces as well. So I don't think you're going to have a winner-takes-all type of thing here.

9:33I think you're going to have a sufficiently large market where you have multiple players. They may end up being split by geographical aspects because it becomes very difficult for a company to acquire the talent to be very, very good at European languages and Southeast Asian languages. you know for example there seems to be some element of like like kind of like local like pressures geographical pressures there um i think also there's like a number of like questions about how you know how these machine learning models evolve and what new architectures come i think we're you're going to see similar you know similar things to kind of like what we're seeing in like the large language model space which like there's a lot of questions in that space about whether we're hitting scaling limits and like that we need to go into new architecture so so i think like i mean the the the great thing for us and our customers is like um you know that when if you know if one vendor doesn't have the ability to do it we can go to another sometimes we can partner specifically with with vendors on particular uh voices and stuff i think we did a lot of work recently with gradient particularly around french which was a language that we were kind of struggling with.

10:49We also have some customers or prospects in. But the fact that they're based in Paris and have that staff on their payroll, you allow them to kind of iterate through that quickly. And also things like access to data and stuff like this. And there's a whole network of data providers. I know also Slatter, you wrote a report about this.

11:16Peadar Coyle:them recently as well. At what point was it clear to you that, like, okay, we're not going to build these models from scratch. We're going to basically become a kind of a model neutral platform, number one. And number two, then, like, where does the competitive advantage come from? Like, the long-term competitive advantage or USP, I guess, for AudioStack? It's a great question. originally we especially in the early days we were like doing dedicated work around building our own TTS models we had a dedicated team at some point we realized that that market was evolving very quickly and I think you almost have to like like pick your battles like I think I think one of the big the really important things in business I think this is very much for startups but in business it's very important to almost like pick your lane and decide where you are and we we find that our lane was around you know like high audio quality orchestrating and unifying things working very closely with our customers um you know collecting things like brand guidelines you know because we're predominantly in the advertising space but like that we do a lot of work in terms of like you know enabling brands to have their tone of voice and their own kind of like collateral and ingesting that into the platform and making sure that we you know and making sure that we build a great product by working with them.

12:43I think there's a bit of a naive impression out there that foundational models are going to do everything, right? I don't think that's true. I think we're just in this early innings and Tropics about to go IPO and so there's a lot of hype. Maybe. Maybe, maybe not. Who knows? But there's this kind of naive impression that they will absorb everything. So I would actually, Well, I think that in five years' time, the foundational models will look a lot like hyperscalers, you know, in the same way that we have in, like, AWS, et cetera. And there will be sufficiently large application layer companies on top of them.

13:24You know, we already have examples of that. We have the likes of Intercom and the kind of customer service space, likes of Sierra, Decagon. And we'll have more of those in the same way we had Snowflake, Datadog, et cetera, built on top of. and also in some sense semi-competitive with right so that's the other thing i think there's that's the one thing about the space i think the the foundational model players will always sort of come into your space but you can always like you know like be better than them and be more focused and i think people sometimes underestimate the power of domain specific knowledge and and just how deep some of these problems are they're not they're not super fit you know like um like we we were doing like some testing a couple of months ago around like um for a retailer and around like like talking about like kids uh like going back to school right there were so many things that could go wrong right even during testing you know so you could have like like a kind of like a very adult voice you could have like adult subject matter you don't want that for a kids back to school thing you could have the wrong jingle where like it's kind of like a bit scary there was just lots of different things that could go wrong there you could also pick the wrong kind of skew where it wouldn't make any sense as an example like you pull it from the website wrong there was like all sorts of like kind of like things and as we've been building more like a gentic kind of in particular application but building more robust testing, building more QA and making that like a much more robust experience.

15:03And I think that's one of the things. So like if you take that, you could have a very impressive demo that you could like show someone that would work in the demo. But to do that time and time again for like, you know, on day in, day out that people are using it, that they get value from it, I think it's extremely hard.

15:21Peadar Coyle:And people need help, it seems from what you're saying. Like you could potentially spin up, if you're a large enough company, you could like spin up an internal team that could go to these foundational models or whatever, build it themselves in some shape or form, but then what do these people do when the campaign's over or when they need to do something else? So it seems like you're very much kind of hand-holding the clients both on like, okay, this is the latest tech, you should do this, but also very much on kind of in an agency type of consultative approach, like how to solve these problems using the latest tech.

15:58Is that accurate? Yeah. Like, you know, we sell software, you know, and, you know, there are different parts of that. There's more front-end focus, there's more API focused, et cetera, depending on what your needs are. We often discover that there is a consultative approach needed. You know, people, you know, whether it's on the more creative side, you know, the ability to just think of what's possible, right?

16:26Peadar Coyle:Yeah. Because, you know, because people need examples and need handholding. They need to actually understand things in a very kind of like, and I think we're seeing that across the market as well. And, you know, we have a team, like kind of operational customer support team who spends a lot of time, you know, both, you know, doing workshops, doing training, working with their customers. to make sure that these things land. And I just heard yesterday one of our larger customers who is in the media space, about 20 % of their adverts that they're running use our technology. So about 20%. It's roughly 20%.

17:11It might be a little bit less, but things are going live. And that's been a journey. They've been working with us for nearly two years. But, you know, we've done the onboarding and we've started in a small part of that organization and we've expanded and we've worked with them and we've improved the product and we've taken their feedback. You know, we've added some better language features and we've improved sonic quality, et cetera, et cetera. All this kind of, you know, investment and, you know, in R &D to drive a better experience. But the fact that that is actually happening is a sign that we've gone past that kind of cool demo to actually driving business change out there in the real world.

17:53Peadar Coyle:Would you consider them kind of a direct client? It's a media company. They're using your tech for their own purposes. If I'm a marketing agency or I think somewhere else you mentioned you call them like a hold call. I guess that's like the Omnicoms, the WPPs of the world. like how do they partner with you like do they bring you in and then you're working directly with the end client or like you liaise with somebody at the agency or or that whole co it can be a blend it depends on how the so sometimes they'll want to white label our technology and integrate us into like a larger suite um so that they have more control and sometimes they will want us to like work directly with the client you know to kind of understand their creative needs and you know you know to get something off off the ground um uh i think like it was the kind of like with the how the current kind of advertising and media space is you know you know these kind of like large agencies exist for a reason you know they exist because of complexity and kind of the mark and the fragmentation of the market um but they exist also because it's quite difficult for even larger brands to really insource all that knowledge, right?

19:10You can insource parts of that stack. And also audio is just one part of a much wider advertising ecosystem, right? There's lots of other parts that people need to worry about and think about as well as part of their ongoing kind of marketing strategy.

19:32Peadar Coyle:How do you find those leads or clients in those spaces or do they find you mostly or like do you have like an active sales org? We have an active sales org. We do a lot of events. We do stuff like this. We have some referrals as well via people who've had success stories and via case studies that drives kind of other adoption as well. But I would say a lot of our activity is enterprise sales. Outbound is kind of a very classical go-to-market motion that you often see in software companies. Well, you call it classical, but actually there's probably a bunch of companies that are like one or two or three years old that think like, hey, I have a product now, please come.

20:16Peadar Coyle:So, yeah, it takes a while to understand that, you know, the outbound sales function is a real thing you need to worry about, even if you have a great mousetrap, I guess. Yeah. So just now, but also at Slatecon, you did mention that like, you know, successful deployment depends on consultancy integration. what are your thoughts around this kind of FDE forward deployed engineer discussion that's really popped up over the last at least on my radar like last four to six months like everyone's talking about it I think Palantir are very good at marketing that's all I can say a lot of this isn't new so I don't really like personally the forward deployed engineer term but I use it as a shorthand it's becoming in the zeitgeist I think it's interesting that like as you have like some of these foundational model companies saying we can replace your sales force with our AI, they're hiring large sales forces and FDE teams and consultancy teams because humans need to deal with humans and humans have specific needs and humans want, you know, and organizations are complicated, right?

21:26So I think one thing that I've learned from working with our team on that is just how it's not like there's the political aspect of the organization. You're going to have some people who are more proactive about technology than others. But also there's just the systems and the integrations and the different systems that the people don't always know. So you could have a sales call and ask people, what other systems do you have around? You send someone to do a workshop and they said, oh, someone mentioned Xsoft. It's like, oh, maybe we should look at Xsoft and maybe we should link into that. And I think that's like a big part of them, like, you know, growing as a product inside, you know, like what integrations you build and how you partner with them.

22:17You know, both on like the kind of commercial side, but also the technical side. but also like how you think about how they think about getting their job done because you're just you know your your your job is to make people's life easier like so many products don't do that but like but like I think that is like one of the fundamental things that you have to think about in terms of like like driving value right and you know and revenue is just a proxy for value at the end of the day.

22:47Peadar Coyle:Ideally, it is, right. So you mentioned a certain level of personalization, for example, with the 30 different accents, right? You want the message to land, you want people to really relate to the content. But at SlaterCon in London, you also said that personalization down to the individual can kind of get creepy even. So where do you, give me an example of how it could get creepy. And then like, where is there maybe a commercial opportunity for like super personalized audio that people may actually welcome i don't think we've ever done this but i've definitely heard some discussions around like like on the name level and i think like hearing your name in an ad would be really really creepy right you know like there's i think that that would feel invasive yeah and you know and some of that's a cultural thing your culture evolves um but in terms of like what we've what we've seen um we've seen a slow success around localization you know in terms of like state or county or whatever we've seen a lot of success around you know things on the school level particularly in like retail you know to drive campaigns for particular um like items in the store that people are trying to to drive sales in um we've seen some things around accents um and like i think in the very early days we were thinking of like everything being hyper personalized but there's like elements of personalization that are like almost socially acceptable and you know there's an element of of taste and brand identity and all these kind of things I think is quite important to kind of respect you kind of you can't be too far ahead of the market right you can only ideally you meet the market where the market is well you can you can try to be far ahead of the market but it often doesn't work commercially right yeah right that's I mean that's a very difficult kind of balance to strike today, right?

24:42Peadar Coyle:You can be too early or left behind within like a matter of weeks. Yeah. I mean, talking about personalization, I mean, like when I took a Waymo in San Francisco three weeks ago, which I must have told everybody I've met since, because it was the first time I took a self-driving car. Like it was interesting. Personally, you get into the Waymo, it greets you, hey Florian, you know, good to meet you, like where are we going? that wasn't creepy but I guess if it's in an advertisement context then it would get creepy very quickly if like I listened on on a podcast or something I'm getting these personalized ads on podcasts now which is kind of new for some of the very big podcasters sometimes they're in the wrong language though so the targeting isn't great so I hear like a French kind of ad because maybe they pick up that I'm Switzerland but in the wrong city or something but yeah if that ad would like mention my name it would get really creepy very fast now but yeah you're right maybe in five years from now it's like sure I need to get my personal ad read to me on on the podcast app yeah huh is that something by the way that you're seeing like podcast advertisement is that something you're coming across well yeah we're we're doing work with podcasts, radio, streaming, TV.

25:59I think those are connected TV as well. It's a big growth area because there's a lot of data in connected TV. Those are probably the main ones. Podcasts, sometimes people are apprehensive about the brand, because a lot of podcast ads are host-read. They're read by the host, so there's an element of... That can be quite lucrative. So sometimes people are a bit worried about that. There's kind of like some apprehension in the market. But yeah, we are seeing like, you know, we're seeing this explosion in inventory. And when you have an explosion in inventory, monetizing that inventory becomes like a challenge, right?

26:41So, you know, and or an opportunity. So people, you know, are often looking at things like this. And there's a range of kind of like marketplaces and partners in that as well, which I'll not bore the entire audience with.

26:56Peadar Coyle:It is interesting though. I mean, I'm a really kind of avid, for lack of a better term, podcast listener. And yeah, I mean, what's better? You get an ad read by the podcaster, but it's like maybe it's an American podcast and they talk about, I don't know, mortgage or something, like offerings, which are completely irrelevant to me being here in Switzerland, right? So do I want the podcaster to talk in his or her own voice about something that's irrelevant or do I want to get a localized ad read to me which kind of breaks my podcast flow? But at least it's something that I might actually end up buying.

27:33Then there's even things around the sonic identity as well. So if it's something like The Rest is History or such, my favorite podcast like which is very quite serious like about history you don't want like a very like like techno track or something in the in the podcast ad that you know doesn't work in terms of like the kind of like identity and you know we've done a lot of work around stuff like that as well both on the ai side but also on the the more creative side of like supporting people so that they they can they can land the land the right things but i think you're going to see i think you're going to see more and more of this as you know because basically you know as podcasts go into more uh inventory you know kind of country channels also they're spread on more like kind of different networks you know and you know you're going to see like increasing kind of consolidation and and also like new entrants coming in like we've seen the likes of podimo and kind of in the podcast space and there will be more like you know like as that you know you know, as this kind of like cycle of consolidation and then divestment happens, which seems to be happening all the time.

28:49Peadar Coyle:Last question about podcasts, but as a podcaster, you know, having you here is a great opportunity. Like, are the economics kind of similar to YouTube where like, okay, if you have millions of listeners, it makes a lot of sense and you actually make real money. And then if you're like a small niche podcast, don't even bother. Like we could, I think, theoretically monetize our YouTube channel now, but I mean, I'm not sure if the two clicks we need to do to do this kind of merits the you know few cents we'd get from it is that the same i think that's probably correct yeah this is like a lot of this is around um you know there's a lot of like there's a whole long tail of like podcasters who have like 30 to 100 listeners a month you know and then there's like the more dominant players like who you know who have often end up becoming quite lucrative media empires.

29:40You've seen this with like the Goal Hanger, which is associated with or partly owned by Gary Lineker, the former footballer. It's done a whole, it's done both, started in podcasts, moved into video, does actual in-person shows, has various sub-channels. It has really become kind of a mini kind of media empire in itself. And I think you're going to see more of that. We've seen this with Joe Rogan in the States. And I think we'll see that in other areas as well. My knowledge of the German or Swiss ecosystem is not very good off the top of my head. But I'm sure there are kind of local success stories like that as well.

30:26Peadar Coyle:Oh, they're totally low. I mean, there was one show they fill like basically kind of the equivalent of the Wembley. Like just after they've been podcasting for, you know, years, then they were like, well, let's do an in-person show. Boom. It was full. Right. And then you have the ability to sell merchandise. So, I mean, it's a classical, it's the same thing that bands did, right? You know, it's the same thing, like finding auxiliary, you know, build a brand, fine auxiliary like kind of business models on top of that that are more monetizable you know i think iron maiden famously makes more from merchandise than it does from selling like like tickets to see it in the stadiums and there's there are a bunch of examples like that in the kind of event i think you're seeing that in the kind of youtuber influencer kind of space which i think is a very interesting kind of like phenomenon and quite a new phenomenon and it's also really a function of the fact that a lot of the gatekeepers, the classical kind of media gatekeepers have become increasingly eroded by the likes of YouTube, TikTok, etc.