#277 LTP Growth, Voice AI Valuations, RWS, Appen, Lionbridge

0:00Anna Wyndham:A lot of language technology platforms have a great product, they have great models, but they reach a point where the product is strong, but momentum is not really scaling as quickly or predictably as expected.

0:17Florian:Hey everyone and welcome to another great episode of SlaterPod. Hello Anna.

0:22Anna Wyndham:Hi Florian. Hi everyone.

0:24Florian:So Esther's traveling today. So we have our very own head of research, Anna Wyndham, joining us from Madrid on the podcast today. And great timing because there is a major piece of new research out. Maybe we'll start with that, Anna. Just give us the kind of highlight points of that latest piece of research. I think we published it last week.

0:45Anna Wyndham:Yeah, just last week. It's called Slater Pro Guide Growth Hacks for Language Technology Platforms. and it's a 45-page growth playbook and it's for companies who are building and scaling language AI platforms. So basically it's a playbook for growing revenue. It spans things like product design, sales execution, the realities of procurement and so on. And it's really targeted at early to mid-stage companies and scale-ups. And just some background about why we decided to create this piece of research is that a lot of language technology platforms have a great product, they have great models, but they reach a point where the product is strong.

1:35Anna Wyndham:But momentum is not really scaling as quickly or predictably as expected. So we looked carefully at the language tech platforms that have done this well and distilled this down into 10 actionable strategies that can be applied without kind of major reorganizations or process change, things that have been shown to be linked to growth that language tech platforms can implement.

2:04Florian:Yeah, because it's very hard, right? I mean, again, you have a good maybe model, you have a good initial idea, a good application on a specific vertical or what have you, right? But then there's like a gazillion other things you need to do to actually go to market and make it work. So this is basically marrying our expertise and our research on the technology side, but also with the background that many of us in the team have from like the realities of actual procurement in the back. I mean, some of us have like 10, 15 years experience of working with big enterprise buyers. And so we kind of put that all into, what is it, 45 page, 10 different ways to grow.

2:45Florian:So yeah, great stuff. So if you're early to mid-stage, go and get that because you're competing and that's what we want to talk about next with very powerful organizations. At least some of them are scaling super quickly, like 11 Labs, which raised, and we did allude to this before, but now we have a lot more context there. So they raised$500 million US in their Series D and the investors valued the company at$11 billion. That's crazy. And that is arguably the fastest growing and likely most highly valued kind of LTP that we've seen so far in our 10 years it's later, I guess.

3:33Anna Wyndham:an acceleration of this pattern that we saw emerging last year with voice and speech, attracting the highest levels of funding out of any of the language categories.

3:45Florian:I mean, these guys started in what, 2022? I mean, there's this origin story, you know, like they didn't like the Polish dubs where like one person just speaks the entire, just all of the dubs, one person in Poland in the past. And so they wanted to fix that. And lo and behold, four years later, they're raising at an$11 billion valuation. I want to talk a bit more about that in a second, but also Synthesia, our friends said they raised$200 million in a Series E. Let me check how much they were valued in this. Probably$4 billion. So Synthesia, the multilingual avatar company, raised$200 million at a$4 billion valuation, which is also incredible.

4:28Florian:Now, in terms of the investors, Synthesia lined up, you know, the big ones from the hyperscalers, Google Ventures, NVIDIA's Nventures were there, a couple of other like XL. You know, they obviously like the Google Ventures and NVIDIA's, they want to make sure that you have the apps that are using all those TPUs from Google and GPUs from NVIDIA. And then 11 Labs, the round was apparently led by Sequoia and then a bunch of others like Andreessen Horowitz. And Sequoia is interesting because Sequoia is also an investor in Lilt. Yeah, but it's been a while since LILT, so it looks like they have found a new language AI darling with 11labs.

5:09Florian:Also, the marketing machine, I want to talk about this, that kind of VC marketing machine is working overtime. So Andreessen Horowitz put out like an 11-minute documentary on 11labs. You got to watch it. We can link it in the show notes where like the founder of 11labs walks into the Andreessen Horowitz office and, you know, hugs the partner and sits down for an interview. Yeah, quite the viral marketing they're trying to do there. All right. So in terms of implications for this, for the kind of the broader LTP ecosystem, I think we can just leave it for now at, you know, it's a major outlier.

5:44Florian:It's kind of a 10x in terms of the funding of what we've seen in the kind of core localization space over the past five to six years. So, yeah, I guess we hope to see 11 labs at our May 2026 London conference. One more though, I want to talk about 11 lamps also just launched something called expressive mode because the expressive thing is obviously a very, very important component here, right? That getting away from the robotic, you know, crossing the uncanny valley and all of that stuff that they're saying and that in their press release is saying that expressive mode gives teams unprecedented control over tone.

6:24Florian:so agents can de-escalate reassuring guy conversations to clear resolution. So clearly in this particular press release, they're kind of referring to contact center use cases, et cetera. All right, Expressive, also very important for YouTube with all of the AI dobs, all of their efforts around the AI dobs. And I want to quote you an important stat that the CEO, the YouTube CEO Neil Mohan, released in some type of CEO letter in late January. So they were saying that now in December 2025, they had more than 6 million daily viewers watched at least 10 minutes of AI-dupped content. Seems okay. I mean, not like billions.

7:16Florian:So we're talking about 6 million daily viewers watched at least 10 minutes of auto-dupped content. And I got to say, there are certain things that I'm starting to use it for also with the kids, like some kind of like YouTuber documentaries where, you know, somebody walks around with an iPhone and like, you know, speaks English and, you know, I can just click and the kids can watch it maybe five or six or seven minutes in German. Although I think after five minutes, I think they're starting to tire a little bit now because it's not perfect yet. But going back to the expressive component, they also launched multilingual expressive speech, they being YouTube.

7:55Florian:So I don't know, like if expressive on YouTube is becoming vanilla, like a vanilla feature on YouTube, that would arguably take a bit of kind of tam away from Eleven Labs. So back to you, Anna, so you've never watched anything on YouTube in Spanish dubbed or anything like that? Or you haven't been presented with these translator captions or YouTube headlines?

8:20Anna Wyndham:Presented with auto-dubbed content, Spanish content auto-dubbed for me into English. And I had to go back into settings and undo that because this is not how I would choose to consume the content.

8:37Florian:So what are the settings? Because I keep getting presented German videos with English headlines. So what would I need to do? Yeah, because that was also one of the things that people commented in a video where they launched the expressive mode where one user said, and this was the most liked comment, it's incredibly frustrating. That's a quote now from a YouTube comment. It's incredibly frustrating. the number of times I'll see a video in a language I understand fluently with a poorly translated title and description and click on it only to hear an awful autodub to English. Same for me. So I'm on YouTube.

9:17Florian:I see a German video from even a creator I know in German, and then I see the headline in English. The YouTube title is in English, and then I click and it's autodub. So I don't know. What would I need to change in the settings? Or maybe YouTube just gives you the option of having like more than one language because there's many people that are multilingual. And, you know, I want to watch videos natively in German and in English and maybe even in Spanish. Anyway, so why don't we do a test? Let's do a test now because they did a launch video. And I think I can share screen here for those of us who, you know, watch this on YouTube.

9:54Florian:That's a little meta there. Let me try to pull this up. And then we watch this. So I share the sound. And we check this here. And then I'll briefly do the German and you can do the Spanish analysis for a second. So let's get cracking. So this is looking at a video from the product lead talking to Creative Liaison at YouTube and they're explaining the AI dubbing. So I'll just click in here and then we go to English, the original. In the audience.

10:34Anna Wyndham:So what you want to go and do is look at this at the language level and YouTube Studio makes it easy.

10:40Florian:Okay, that's his original voice in English. Now let's go and change the sound to German. You can filter the performance values according to language, by using the audio track filter in YouTube Studio.

10:56Anna Wyndham:That's great. I always like to rate creators to see these values. Math is hard, statistics are terrible and average values are always stupid.

11:04Florian:Not bad, not bad. So two voices, different voices, totally different from the robo voice. So why don't we go to another one and get your live reaction here, Anna? Because I think you haven't looked at that before. So let's go to Spanish. They've got to have Spanish US. All right.

11:51Florian:Wow. and I remember a lot of videos of YouTube, it was only a single miniature, a single miniature. Thoughts, reactions?

12:00Anna Wyndham:To an intermediate speaker, like a second language speaker, it comes across as, I mean, it's perfectly passable, but even I can hear that the intonation and the rhythm is very much copy pasted from English, no?

12:13Florian:Absolutely. And I did watch probably half of that video before the podcast in German. It was, I think, incredibly impressive the first 30 seconds got a little harder on my kind of trying to focus maybe for the two minutes and then I started dropping off a little bit there were a lot of kind of weird glitches here and there but like it's just I mean it's a it's a quantum leap compared to the robo voice we had two months ago right so uh yeah uh yeah I mean I find it uh find it super interesting because that I mean they're launching this at massive scale and I think it's just a matter of time until it It gets pretty good in terms of the voice.

12:51Florian:I find the translation sometimes still, it's just, I mean, not literal, but just the usual kind of NMT level, quite okay, but like, you know, a few misses per minute, I guess.

13:03Anna Wyndham:I mean, the naturalist and the intelligibility are both, yeah, impressive.

13:09Florian:They're very impressive, right? Cool. Well, so I guess in the next 12 months, we're going to have a lot more natural voices in all kinds of languages. Frankly, I think the Spanish was just a little bit better than the German, actually. I thought it was pretty good. The German was quite good, but not quite the Spanish level. I wonder how it is in maybe slightly lower resource languages. Cool. Now that we spoke about a pretty at-scale application of expressive voice, you looked at a research paper by, I think it was somebody from, was it Tiautong University? Shanghai and then Microsoft also. They looked at a similar theme topic, right?

13:52Anna Wyndham:Yeah, this was a piece that we published today, written by our researcher Maria. And it's looking at research from, as you say, leading universities in China plus Microsoft. And they're looking at how speech synthesis models have improved rapidly, but the methods that we use to evaluate their quality actually under test what matters in real world production and deployment. So exactly what we've seen with YouTube dubbing there, the kind of aspects that are most important. So most or traditionally text-to-speech evaluation focused on naturalness, intelligibility, of course, latency, and then if voice cloning is involved, how similar the voice is to the original.

14:39Anna Wyndham:But what these researchers are saying is that it really misses, that this approach really misses what matters when you need to deploy a system at scale reliably. So whether or not a voice remains consistent over long form content, if it follows punctuation and structure correctly, expressiveness, as we've talked about, and how it behaves across varied and messy inputs. So I guess it would be interesting to see if the performance, say in the YouTube example and other examples, if the performance remains consistent over a long period of time. So this really has implications for everywhere that voice is applied.

15:25Anna Wyndham:So audiobooks, dubbing, enterprise voice agents. So imagine bringing your bank and you're speaking with an automated customer service voice. if that voice changes tone mid-sentence or the accent drifts or there's some kind of unpredictable behaviour that then becomes a trust issue. So this is why production, reliability, consistency, control are becoming more of a focus now that those earlier thresholds of intelligibility and naturalness have been reached to an extent anyway. So, yeah, the researchers have put forward a new method to measure those aspects.

16:06Florian:I like the messy inputs. Remember we had the AIOLA on the podcast. I mean, they're doing a lot of like very messy input type of use cases, right? At airports and like real life conversations out there in real situations.

16:21Anna Wyndham:And the researchers, they do acknowledge that if we expand how we evaluate models in this way, it will increase complexity and cost. But what they're saying is that this is a necessary trade-off in order to keep pace with real-world deployment, because these are the conditions that will be relevant anyway. So we need to keep pace with what the expectations are around how well the models are going to perform in those scenarios.

16:54Florian:All right, let's move to Appen from Australia. Yeah.

17:00Anna Wyndham:Appen reported double-digit revenue growth and an EBITDA turnaround for Q4 for the financial year 2025. Though just to caveat that this was a quarterly activity report. So it was based on unaudited management accounts. So it's not a formal filing.

17:18Florian:So you're not trusting the auditor there. So you're not trusting the internal CFO. Let's just unpack. So they said EBITDA turnaround. I mean, this is a big turnaround, right? I mean, even the stock was down like 99 % and now it comes roaring back. So basically, we're talking about the turnaround story here in the age of LLM.

17:40Anna Wyndham:Exactly. And so revenue is up 10 % year on year. But the more important point is that a larger proportion of revenue is coming from higher value projects. So this has implications for margins, and it seems to have also significantly affected how investors are viewing the company.

18:01Florian:It would be incredible. I mean, they went through a near-death experience. And if they're coming back, if people start having confidence again that their data work is actually sustainable and growing in an LLM world, that would be a massive turnaround. I mean, we just spoke about, you know, 11 labs raising at an$11 billion valuation. I mean, that's literally like 25 times higher than, you know, the publicly traded Appen valuation. So, you know, imagine if investors, public investors are starting to have confidence again in Appen. So where's this growth coming from? Where are we seeing? Yeah. What's the sources?