The Future of Live Multilingual Captioning Ai-Media CEO Tony Abrahams

SlatorPod #163 - Live Multilingual Captioning Ai-Media CEO Tony Abrahams

Tony Abrahams, CEO and Co-founder of Ai-Media, joins SlatorPod to talk about the journey to building a market leader in multilingual live captioning.

Tony discusses his transition from working in finance to co-founding Ai-Media with Alex Jones and introducing large-scale captioning to Australian Pay TV. He gives an overview of Ai-Media’s technology stack, which delivers high-quality automatic captioning through three key elements: encoding, the iCap network, and LEXI.

The CEO talks about the use of respeaking versus LEXI in settings where captioning accuracy is critical, and where there are multiple speakers, mixed-quality audio, or background noise. He discusses how Ai-Media measures live-captioning quality using the NER model, which weights the types of errors as editing errors or recognition errors.

Touching on the multilingual component of Ai-Media, Tony explores the possibility of using AI instead of respeakers and having a fully-automated translation product in the near future. He believes that large language models are an opportunity as the technology has enabled them to interpret sentences more accurately, resulting in a better outcome with LEXI 3.0. 

Subscribe on YoutubeApple PodcastsSpotifyGoogle Podcasts, and elsewhere

Tony gives his thoughts on growing through M&A and the strategy behind acquiring EEG to gain a competitive advantage in terms of its technology and product suite. He shares his rationale for taking AI-Media public.

The CEO reveals Ai-Media’s roadmap for 2023, such as improving the iCap network and launching the LEXI Library, which allows customers to search their media library by captions.

Transcript

Florian: Tony, first give us a little bit of info about your professional background and how you got involved in language access services. 

Tony: It was a bit of an accident for me, actually. I’d just come back from Oxford University on a Rhodes Scholarship where I’d done a Masters in Economics and an MBA, and I went into finance for the first year and actually ended up trading a hedge fund in terms of merger arbitrage and decided that I didn’t want my life to be worried about what the gold price was going to be doing between one market closing in Sydney and another opening in Toronto. And so I quit that job and, actually, my partner at the time, Alex Jones, is profoundly deaf. And just two days after quitting that job, we were at a cocktail party and we met the Director of Television for Foxtel, which is a major subscription TV service in Australia, and Brian Walsh, who was the Director of Television there, asked Alex if he had Foxtel, and he said, “well, why would I pay 100 bucks a month “for something I can’t understand?” And he said, “What do you mean?” He said, “Well, you know, I’m deaf. “I can’t hear what’s going on and I need captions. “And if you had captions then I would pay 100 bucks a month for that service.” And then you could see Brian start to think how many people like this are there that would pay $100 a month if we did provide this service and asked us if we could come in and give a presentation and effectively, you know, being two days recently unemployed, I thought absolutely. And my background before that was in management consulting. I worked at BCG for a bit and so this was kind of right up my alley. This is the cost of doing something. This is going to be the benefit of doing it. You know, captions are going to be beneficial, not just for people who have a hearing impairment, but, you know, people who speak a language other than English. And also it helps improve comprehension for everybody. And I think that, you know, fast forward sort of 20 years, what we’re now seeing is an entirely new generation only watching content with captions because they’re used to it through social media feeds being silent and perhaps they don’t want to engage quite as much with that content or they’re multitasking. But we are finding now that well over 50% of millennials and Gen Z are actually watching everything with captions. And that certainly has helped to improve, I think, that business case for our customers and that was something that, say Foxtel saw pretty early on. So after doing a six month consulting engagement, Foxtel decided that they were going to introduce captions to the platform and in fact invited us to set up a company and tender for that work. And that’s actually how Ai-Media began. So it began actually by providing recorded media services for the pay-TV industry in Australia. But very quickly we realized that actually the opportunity was much bigger than that and effectively the start that we had and the research that we’ve done into that industry made us really clear that we needed to focus not just on recorded media, but we needed to focus on live. And that live content was actually where the real greatest degree of difficulty was and that we needed to be doing this not just in entertainment and media, but we needed to be doing this in education, first and foremost. Because, you know, as great as it is to provide access to the marvels of television, actually providing access to grade school and high school is far more important. And 20 years ago, 50% of people who were deaf or hard of hearing in Australia ended up never, ever getting a job and ended up on disability support for, effectively, ever because they never actually got access to education. Thankfully that’s changed because of the absolute breakthrough in access that we were able to provide initially with re-speakers. So, you know, we used the AI in the very early days to effectively have a respeaker as an intermediary in that live process where the automatic speech recognition couldn’t give that level of accuracy that you needed. The hack, if you like, the AI hack, was to have a respeaker over many hundreds of hours train and tune the AI to their unique and individual voiceprint. And then they would literally listen and repeat what was being said into this speech recognition engine attuned to their voiceprint. And that would give us the level of accuracy of sort of 99.5% that at the time was required. Of course, the huge changes in AI in recent years and our recent acquisition of EEG, which is why I’m now based here in New York, has massively accelerated that transition away from a respeaker dependent model towards one now where over 90% of the content that we deliver is through a fully automated Lexi solution. And it’s amazing to see really how that transition over the last 20 years has evolved. Getting, you know, kind of more and more scalable, more and more accurate, and more and more languages and more and more applications and more and more industries for live captioning.

SlatorPod – News, Analysis, Guests

The weekly language industry podcast. On Youtube, Apple Podcasts, Spotify, Google Podcasts, and all other major platforms.

SlatorPod – News, Analysis, Guests

Florian: Respeaking is, like you say, 10%, so just in highly critical scenarios or where would a respeaker still be required? 

Tony: If you’d have said to me a year ago, and I think this is really important technology context, when would you put a respeaker, like a really good respeaker on a piece of content over Lexi I would have said, “Well, if it’s really important content, right?” You would put a race speaker on it. Actually today, because the technology has evolved so much just in that time, I would actually say there’s some pieces of content now that I would actually recommend Lexi for over a respeaker and those situations are where you have really good quality audio encoded. And in fact that encoding of the audio is one of the key assets that we acquired with the EEG business. So they’ve got 43 years of intellectual property built up in their encoding products, and these encoding products are all connected via the iCap Network and all connected to the iCap Cloud. And effectively what that system does is it ensures that we get the cleanest possible audio into the system. We then match the most appropriate speech recognition tool. We then dig in behind our customer’s firewalls to understand the data and the metadata associated with the particular content that’s being captured. So for example, if it’s a weather bulletin in Chicago, we’ve got all of the Chicago suburbs and nearby areas automatically featured kind of in that dictionary set so that the AI understands the context that it’s providing that for and where all of those elements come together, where you’ve got good quality audio and you know the subject matter that you’re talking about and you’re plugging into the iCap Network, Lexi actually performs better than humans at the moment. Why better? Because it’s quicker. And even if you’re getting that same level of accuracy, but you’re doing it with a latency that’s two seconds faster, that’s a much better viewer experience. So where would I say today that you would put a respeaker? You would put a respeaker on something that, A, is really important content, but, B, where you do not have that situation where the AI can deliver those results. So, for example, where you have multiple speakers, mixed quality audio, background noise, singing, multiple languages. That kind of stuff is still going to require humans for that sort of 10%. And I think that percentage will shrink. But the reality of it is that while that percentage of content might only be 10%, it tends to be the most important content for our customers. So it tends to be things like, you know, the halftime show at the Grammys where you’ve got someone singing bilingual in English and Spanish, and it’s really important that the lyrics come up. Well, there’s no automatic speech recognition tool that’s going to do that. You actually have to understand that this is a really important 10 minutes of programming. You know, that there’s tens of millions of dollars of advertising that’s being sold off the back of it. It’s worth getting this right. And as Ai-Media, given that we have hundreds of highly skilled respeakers and stenographers, some of whom are bilingual, we make sure that we have the resources for those customers so that our proposition to them is, you know, don’t worry about it. We will put the most appropriate form of captioning on air. Where it’s possible, we will do that with Lexi, but we will also make the judgment where that’s not possible to do it with one of our premium services. And that’s really, I think, the value proposition for at least say the next five years. This need for the high quality customers to have this mixed mode of delivery and this mixed mode of delivery is enabled by the iCap Network. That iCap Network was designed so that it could be delivered with either a premium quality stenographer or respeaker, or being delivered through the automatic services with the Lexi family.

Florian: Now, you mentioned iCap and Lexi. Now just give us the overview of kind of the tech platform that you’re operating. So what is iCap? Is that the broader kind of umbrella tech, and then Lexi is what exactly? Just give the listeners a bit of a breakdown here.

Tony: There’s three key elements really to our technology stack, which is delivering high quality automatic captioning. Effectively, it does start with the encoding, right? So the encoding as I said before, right, it gets the audio into the system and it’s capable of being configured with a customer so that it optimizes for the automatic captioning output. So the automatic captioning is Lexi. Now the network that connects all of these devices to each other and to the iCap Cloud, that is the iCap network on which all of the traffic is carried. So effectively three elements of it. You’ve got, first, the encoding, second, you’ve got the iCap Network that allows that audio to be matched with the ASR, the automatic speech recognition engines, many multiple of which, right, sit on the iCap cloud. Then you run that audio through the engine on the cloud. Using iCap it then sends those captions back to the inserter and then makes it available to the viewer. So effectively, you know, we often get the question, “well, you know what, if Google or someone comes up with this, “you know, with a brand new ASR engine? “You know, does that threaten your business?” Or someone comes along and says, “well, ChatGPT, it’s really good now, “does that actually threaten your business?” And it’s like, no, no, no, that actually helps our business because the better the automatic speech recognition engines perform, we just put another one up on the iCap Cloud. And then we don’t need to pick a winner, a long term winner. We just need to know what’s best on any particular day. And Lexi 3.0 is actually the result of working with our partners to ensure that the ASR that, you know, was not developed for live captioning, it can be applied to live captioning but you know there are some tweaks that need to be made to kind of how that’s delivered. So we need to know speaker changes, we need to automatically place the captions and move them if there’s text on screen or if it’s over someone’s face or if it’s over the play of the ball or something like that. And that’s what the latest Lexi 3.0 does, as well as delivering 30% fewer errors than Lexi 2.0. So we really are seeing a massive increase in the capability of the AI to generate really accurate speech to text, and we’re seeing literally breakthrough results month after month as this technology is evolving and maturing. So they’re the kind of three, does that make sense?

Florian: Absolutely, and I also want to talk about kind of the enterprise versus consumer SaaS because I think what you’re alluding to is often kind of the outside perception that what you’re getting, as consumer SaaS, is like instantly enterprise grade, ready and solves all those niche devil in the detail cases, what you’re describing is very much enterprise.

Tony: We are fully Enterprise. We are not direct to consumer. And we’re also are not, you know, we’re not looking to add our Lexi captions inside somebody else’s walled garden, for example, whether it’s Zoom or Teams. They have solved that problem themselves because they’ve worked out the encoding because you’re all within Teams or Zoom. They’ve got their own cloud where they’ve got the speech recognition engines and they’ve already got the data that you need to improve the speech recognition because they know what meeting you’ve got. You know, in the case of Microsoft Teams, they know your emails and your colleagues names and that helps improve that. So effectively we’re doing the same thing, but in a professional broadcast environment and broadcast adjacent environments. And what I mean by broadcast adjacent is any kind of video streaming. And so we’ve owned EEG for two years next week. And what we’re going to be shortly doing is rebranding and relaunching an integrated product range that really makes it absolutely clear that, you know, everything is designed to support the delivery of Lexi. That’s being done over iCap. And we do that with the expertise of the encoding.

Florian: I was going to ask you about, for example, something like the Whisper launch, right? Which generated a lot of waves. But what you’re saying is “this is good for us, “because it’s just yet another, better ASR engine “that we’re putting on top of our enterprise grade, “or we may choose to put on top of our enterprise grade delivery platform,” if I get that right.

SlatorCon London 2024 | £ 980

SlatorCon London 2024 | £ 980

A rich 1-day conference which brings together 140+ industry leaders views and thriving language technologies.

Buy Tickets

Register Now

Tony: Yeah, that’s right. I will say that, Whisper, we did look at it, it is available on the iCap Cloud. And is available through EEG devices, but it doesn’t actually perform as well in the context of live captioning as some of the others. And again, I think this is really important, is that there can be debates about AI and, you know, will it be a kind of winner takes all or will there be, you know, 5 or 6 big winners, or are there actually going to be winners in certain niches? And therefore, are you going to actually be able to compete as a relatively small business if you can find a particular niche and apply it there? I’m probably more of the latter view. I think that there will be some obviously big players dominating and the big tech companies will. But I also think that actually getting the best use of the data and getting the smartest application of it for particular enterprise and consumer needs is going to give some space for smaller, innovative AI start-ups to actually succeed as well.

Florian: Now you mentioned fewer errors. So what were some of the KPIs that you’re using to kind of measure against Ai-Media?

Tony: That’s a great question, Florian, and I think there’s really kind of two ways that people measure captioning quality. The first is with a simple word error rate. So you look at, you know, what percentage of words were wrong. We use a slightly improved version, if I can call it on that, which is one that recognizes that, yes, we want to check how many errors there are, but not every error is equal. And so there are some errors that are actually more important than others. So if I say, you know, “it is not cold outside,” but the captions came out and said “it is cold outside,” that’s a really important error. Whereas if I missed out, the word “is” that’s probably not so important because people will still read it. And so that system is called, rather than a sort of word error rate, this is called NER, which stands for Number Edition and Recognition. That effectively weights the type of errors and the type of errors could be either edition, so it’s editing errors, or they could be recognition errors, i.e. the speech recognition engine has come up with the wrong word in that particular case. And then things either get weighted as a 0.25 of an error, a 0.5 of an error or a 1 depending on the severity of it. And then that’s the measure that we use for a live captioning quality. This is a system that was developed in 2010/2011 by Professor Pablo Romero-Fresco, at the time in Roehampton University. It’s been adopted by just about every regulator that sort of regulates captions. And there’s an internationally accepted benchmark of 98 on that score, right? 98 and above is OK, right? Like that’s good enough to meet sort of legislative standards. Our respeakers, our really good re-speakers, our really good stenographers, they typically get 99.5 or above, right? Now Lexi 2.0., so the version of Lexi that we had out prior to the release last week of the new 3.0., it was getting on average about 98.2. So it was above 98, but you couldn’t put it on every piece of content because the thing about the 98% threshold is you need to hit that 98% threshold all the time. So what we were doing is we were assessing, OK, an average of 98.2 means if you’ve got the bell curve, then although most of it’s going to be above 98, not all of it will. And so we still needed a third to a half of the content to be done in a premium sense. That NER has moved to about 98.7 to 98.8 with Lexi 3.0. So as you can imagine, that bell curve having shifted so significantly and that is a third less errors. So you are, you know, a third of the way closer to 100. And in fact, you’ve closed the gap with the respeaker by half with one new product launch. And it’s only going to continue to get better. Keep in mind, this is in the situations where we’ve got studio quality audio and access to context. But in that environment it’s just much simpler. It’s all automated now. You don’t need to make a booking. There’s, you know, there’s a failover. You don’t have to worry if someone isn’t turning up for your session. And because of that nature of 98 being the minimum threshold. And because we’ve now shifted the bell curve a full half a point, it means that vastly more content than before can be put through Lexi and deliver an NER of greater than 98. And that’s now about 90% of content, which is obviously massively significant.

Florian: Let’s talk about the kind of translation slash multilingual component of your business. Because, hypothetically, maybe even practically, if you were to convert and translate, you’re almost becoming like an RSI provider, like a remote simultaneous interpreting provider in certain scenarios, right? So what is the translation multilingual component and how hard it is? And, you know, in what languages does it present particular problems, maybe?

Tony: We’ve been doing this, we’ve had this product in market since, actually, 2018 and our first customer was the World Economic Forum at Davos. And what we did in terms of the workflow was we effectively got the captioning done in English. At the time, again this is 2018. So it’s, you know, five years ago we were talking, you know, we still needed re-speakers for everything and what we would do is we would have a re-speaker respeak what was being said, but they would paraphrase. So they would paraphrase what’s being said into really short, sharp sentences. And if it was a really long sentence with multiple clauses, they would really, they would cut it up into shorter sentences. That really optimized for the machine translation algorithm. And so what you would do is you would get that kind of simple text, as we called it in English. You would then send that simple text up to the iCap cloud. You would then interrogate one of the different translation engines. It would then send the multiple engines in text, back through to like a different page or in the World Economic Forum on their app. And then what we would also do is we would plug that into synthetic voice engines so that you could actually listen to the automatically generated transcription with a synthetic voice. And so you actually only needed that one human and then you would power, in the case of the World Economic Forum, I think we were talking about 12 languages. And now some of those language pairs obviously did better with the automatic, with the translation tools than some of the others. But I think what’s interesting now and I think where we’re really exploring the possibilities, is can you cut that respeaker out of the mix, right? Like, can you actually just have the Ai kind of generate that and do those translations? And I think what we’re certainly seeing in the industry is very rapid development on that front. I think the large language models and the Ai clusters in some of the really big tech companies, particularly when you’re talking about language combinations between the top 150 languages, I think there’s a lot of work being done now into training the AI clusters and the bigger ones I think are delivering breakthrough results on that. So yeah, I think I could see even perhaps within the next 12 months that kind of fully automated translation product being a reality. And again, that kind of talks to that sort of mythical Babel Fish opportunity. From The Hitchhiker’s Guide to the Galaxy, where you put a little fish in your ear and you could sort of magically translate everything in any kind of language that anyone said. Well, if the AI can transcribe it, and the AI can translate it and it can do it instantaneously, then the future is here.

Florian: You did mention large language models. I mean, you must have gotten some questions from clients, investors, etc. over the past six months about ChatGPT and these various other large language models. What have you told them in terms of like threat versus opportunity?

Tony: It is all an opportunity and in our case literally the same technology that has made ChatGPT such a breakthrough success, and the particular application of it in our context is that, half a point, it can write whole paragraphs, it can interpret from one sentence to the next. It’s like the nature of the speech recognition has just become so much more sophisticated. When we were talking about, you know, the AI with the respeaker, that was using a much less sophisticated speech to text engine that was effectively doing it word for word. Whereas now we’re effectively doing it sentence by sentence. And of course, if you’re doing it sentence by sentence, there’s a much greater likelihood that your sentences are going to make sense. And if you’ve actually got that constraint, that you want your sentences to actually make sense and for it to have some kind of logical flow, then you’re going to get a much better result from the ASR. And that’s fundamentally what has seen that shift from sort of, you know, 98.2 to 98.7, 98.8 in terms of that shift to Lexi 3.0 over Lexi 2.0. So we can point directly to these underlying improvements in AI as really being very beneficial for the Ai-Media business because it helps improve the outcome that we can deliver with our encoding solutions, with the iCap Network and with Lexi.

LocJobs.com I Recruit Talent. Find Jobs

LocJobs is the new language industry talent hub, where candidates connect to new opportunities and employers find the most qualified professionals in the translation and localization industry.

LocJobs.com I Recruit Talent. Find Jobs

Florian: You did mention that you’re at the office of EEG before, which I’m not sure if you’ve rebranded already, but it’s part of a kind of acquisition, a series of acquisitions you’ve been on. You know, you bought Caption IT, Caption Access, EEG. Tell us a bit more about your thoughts around growth, like organic growth versus M&A and like what’s your strategy there?

Tony: Yeah, so we’ve done five acquisitions. We did one in the UK about ten years ago. And then in 2020 we did three acquisitions. All of services businesses very similar to Ai-Media, but with a footprint in the US which helped us get scale in North America. EEG, though was a completely different acquisition. EEG was a vertical integration play. We were already a very, very large customer of EEG and we were their value added reseller outside of the United States. And what we understood was that the technology that EEG had developed is going to be a huge competitive advantage and provide us with a defensible moat while the world transitions to automatic speech recognition and having that hardware and the infrastructure installed in 90% plus of US broadcast allows us with really zero intervention to turn on Lexi for any of those customers. And what we also knew was that this encoding technology could be deployed right outside the US and outside of broadcast. And so what we’ve done over the last two years with the EEG equipment, and we’re retaining that brand as a product line. So the EEG product suite has a 43 year history and some really, really important brand equity, particularly in the US. But also within the engineering community there’s a lot of love for this, for this product line. So taking that product line and making it work outside the US has meant that we’ve had to obviously add additional standards, we’ve had to add additional functionality. You know, we broadcast the Indian Premier League on Star TV, which was the first time live captions have ever been delivered on a sporting match in India. And we’re doing that right now. And really effectively taking the success of a technology that dominated the most difficult live market in the world, being US live broadcast, and saying, “well, we can apply that technology around the world. “And we can apply it to enterprises, as well as broadcasters.” That’s really the strategy behind it. And the great thing about the EEG acquisition was, you know, in terms of scale, it was by far the biggest because, you know, we paid a few hundred thousand dollars for the UK acquisition. We paid 6 million for ACS, and 1 million for each of Caption IT and Caption Access. We paid $35 million for EEG for its technology. And the easiest integration of any of those companies was with EEG. Why? Because we already worked really well together and there was no replication of any of these functions because it was a vertical integration play. With all of the other businesses, everyone had their own way of doing the same thing. And so you had to consolidate, you had to find, you know, kind of new ways of doing it. And, you know, people are always resistant to change, With EEG there hasn’t been any change. There’s just been a real success in the sense that we’ve doubled the EEG business line in the first 18 months of owning what’s been a 43 year old business. So I think everyone’s really excited by that. And the cultures were very similar to begin with, which we knew about. And also, I think, you know, it was very clear that it was an exciting new career opportunity for people, not just to be involved in a business headquartered in Long Island, but to be involved in a global business that’s about taking this great technology and using the sales and marketing teams that Ai-Media had built up since 2017 globally to really help distribute this product. And that’s been where most of the growth is coming from. And I think people are pretty excited by that.

Florian: One of the reasons why you are able to do these types of acquisitions and finance them is because Ai-Media is a publicly listed company. So one of the very few LSP’s that’s actually listed still, which always baffles me why there’s not more. What are some of the pros and cons in your eyes, probably fewer cons, of being listed? And then why do you think there’s not more of this really large language services industry that that decides to go public?

Tony: I think we obviously picked our moment well. So we listed, you know, in 2020 when it was a much easier sort of IPO window than it is today. I don’t think that we would necessarily choose to list today if it were the environment. That being said, being listed did allow us to make a very clear bid for EEG and raise the funding within 36 hours which, you know is a very, very efficient thing to do. Frankly, the discipline of being a public company, I love because it means that we can disclose the performance of the company and be completely transparent with all of our employees. We’ve got a five year plan, we’ve got a three year plan, we’ve got a one year plan, and then we break the year into four month intervals or trimesters where people are really working on, you know, four month projects to really help move the entire business forwards. And I think that clarity of communication that starts with our public disclosures and then runs all the way through the organization has really helped people feel part of a team that’s motivated in a really kind of focused direction, which is on delivering high quality live captioning to professional customers and being an indispensable partner to them. 


Florian: You mentioned one, two, three year plans. So give us the highlights on the roadmap for 2023 and beyond.

Tony: It’s all about Lexi Encoding and iCap. So starting with iCap is probably, you know, because that is the foundation on which all of the traffic is carried. This network needs a major refresh. It needs more security. You know, it needs to be at four nines uptime rather than three nines uptime. This is an area where, historically, EEG did not invest that much because they weren’t getting any revenue from that. What we’re doing now with our customers focus on increased security, increased reliability and increased uptime as the amount of captioning that is provided by Lexi is sort of doubling every year, we are investing very heavily now in an upgrade of that iCap network that will be released in the next few weeks, which will have updated security, a much simpler and cleaner pricing model. We inherited a system where there were kind of six different charges for various elements of iCap, and so we’re just simplifying that to an hourly charge to help support the network. So getting into a kind of sustainable commercial model for iCap, which historically hasn’t had a proper focus or development is top priority for 2023. In terms of encoding, we continue to develop our encoding products which are the physical encoders, and of the physical encoders we’ve got four different encoders available from sort of 4K to a sort of basic SD option. And then making sure that those encoders and the firmware and the software that’s sitting on those encoders can deal with every form of video content that’s being delivered in every form of caption insertion. And so that continues to be developed on the encoding front as well as Alta, which is an absolute breakthrough product because that is an IP encoding product. And so effectively that ALTA product can take any form of IP video, encode it and then encode the captions back. It’s the only product that can do that that I’m aware of in the world, and it’s available as a virtual install. So we can just email customers a license key and they can install it on their kit and be up and running this afternoon. And, and then the final form is obviously our cloud based encoder, which is Falcon. And what we’re looking to do is to provide further sort of product releases in Falcon to make it easier and easier for people to stream. There have been some pretty… We can improve the pricing structure of Falcon as well. At the moment it’s only available at a sort of monthly subscription fee. So we’re looking again to make it easier to sort of bundle that in with Lexi and have it as an hourly fee so that we can we can attract more users to that and more people can get access to Lexi. So, you know, iCap first, encoders we continue to invest in. And finally, which is probably the area where we’re spending most of our investment dollars is in Lexi itself. And Lexi itself is around making sure that you’ve got more and more applications, and more and more value from the data. So we’ve had live Lexi going for a very, very long time now. 2017, when EEG started it. When we acquired EEG 16% of our total traffic was Lexi. That was 35% by the end of December. And we’re seeing that continue to increase sort of month on month. And then a couple of weeks ago, we launched a Lexi recorded product where people can get the same quality captioning back, but where they upload a video to us rather than when they’re sending us a live stream. And then the other thing that’s really exciting, I can give you a really early scoop on this, is that we’re also launching a new product called Lexi Library. And that is an evolution of our sub silo product. And effectively what this does is it stitches in the captions to our customers media library. So that they can search that library by caption and go to exactly the point where a particular word was said. And that is proving to be a real killer app for a lot of our customers. And then, you know, as you mentioned earlier, sort of further evolutions of the Lexi translate options as we continue to get more confidence with more language pairs. Offering more and more of those commercially as well. And so they’re really the sort of three areas where we’re focusing the investment while continuing to ensure that we can deliver that premium quality service with that infrastructure that we’ve already got, and had invested over $50 million in over a 15 year period, is really continuing to make that available and to make it all fully compatible with this upgraded and more secure iCap network.

Florian: Final question, where do you see the market, like the biggest demand over the next 2 to 3 to 4 years? Like where maybe there’s kind of accelerated demand, where it’s already there, maybe some new pockets of demand emerging?

Tony: It’s exactly how we think about things Florian. And so there’s kind of two sales opportunities for us. One is acceleration mode. And so where is it where we can just go and sell something and there’s no need for an additional product market fit. There’s already product market fit. We’ve got all the evidence. Like it should be a really simple proposition. And that’s the case with the broadcast industry, certainly the broadcast industry in North America. Also, I would say the broadcast industry in Australia. Where there’s market development to be done is in those enterprise sectors. And we’ve had a few really, really, early green shoots in terms of cities of. So the city of Baltimore, the city of Austin and the city of San Francisco have all acquired our kit to put in there their local streaming services that they’ve got to their constituents. And so that’s kind of what I call broadcast adjacent in the sense that they’re not a broadcaster, but they actually are a broadcaster because they’re streaming this stuff in the same way that, you know, Peacock would stream the Olympics. And it’s really that mix between, how do we develop those new markets? You know, I mentioned India. You know, that’s obviously a market that’s never done captioning before. So it’s going to be a more of a slow burn. And then obviously other industries where in the past people have relied on, you know, the only available option, which is the kind of human curated captioning services done with an army of tens of thousands of people online that could help with that. You know, we believe there’s also going to be a niche for people just to get back a very fast, rough. transcript. If I call it rough, I mean not 100% accurate. But if they can get it back fast enough and they have the tool to edit it themselves, that might actually be a better option than them having to wait 24 hours to get it sent back to one of these other people. Now we’re not saying it’s going to capture all of the market, but we think that Lexi Live is kind of also very similar to a kind of Lexi fast, which is I really want my captions for my recorded content really fast and I can’t wait 24 hours. And that’s a place where obviously technology can play. Our focus is going to be on the scalable technology solutions. Our focus is going to be on live. Our focus is going to be on where the AI network can bring value to the customer, because that’s what’s going to be a sustainable competitive advantage to us. And then we’re very open and we’ve been very public about this, about partnering with other players in the industry to provide services that are adjacent to that. And we launched our Lexi Preferred Partner Program a couple of months ago with an iCap Preferred Partner Program and finding our space in that ecosystem with industry partners whereby we can focus on what we do really well, which is, you know, live, accurate, automatic captioning, transcription and translation, and then we can look to others to provide those wraparound services. So really focus the business in on where we see the biggest opportunity in the next 3 to 5 years. 

Florian: Just on those partners, give me the profile of these partners.

Tony: This is publicly available as well. So it’s in our first half results. We’ve got a mix of partners really. So we’ve got event partners, for example. We’ve got technology partners like Grass Valley. And we’ve got third party captioning providers like Dynamic Captioning who are reselling Lexi for us. We also work with companies like Rev to provide some wraparound services where quality needs to be improved, you know, overnight kind of thing. And so there are suppliers to us of some of those services as well. And we’re very open to working with, and we work with other vendors as well, particularly where we’re talking about localization services. We did have an in-house localization service that we shut down once we did the EEG acquisition, because had that acquisition not happened we probably would have gone further down that down that language localization route. And it’s worth saying that in 2020, 45% of our business was recorded media and this year it’ll be less than 15%. So we really are doubling down on Live. And therefore, you know, given that most of the industry deals with recorded media, we’re very happy to partner with anyone as long as they can deliver the level of quality and service that our customers ultimately require.