The VideoVerse

TVV EP 17 - Growing an OTT service via user-sourced subtitles

April 29, 2023 Visionular Season 1 Episode 17
TVV EP 17 - Growing an OTT service via user-sourced subtitles
The VideoVerse
More Info
The VideoVerse
TVV EP 17 - Growing an OTT service via user-sourced subtitles
Apr 29, 2023 Season 1 Episode 17

In this episode, we are joined by Omkiran (Omki) Sharma, Head of Video Engineering at Rakuten Viki.  Omki discusses Viki's origin story and how their user-sourced subtitles led to explosive growth in the early years of their service.  He also discusses their video "diff" process to share subtitles between multiple pieces of content that are slightly different from one market to another.  Omki also highlights the challenges and opportunities that Viki faced heading into and during the pandemic.

Watch the full video version.

Learn more about Visionular and get more information on AV1.

Show Notes Transcript Chapter Markers

In this episode, we are joined by Omkiran (Omki) Sharma, Head of Video Engineering at Rakuten Viki.  Omki discusses Viki's origin story and how their user-sourced subtitles led to explosive growth in the early years of their service.  He also discusses their video "diff" process to share subtitles between multiple pieces of content that are slightly different from one market to another.  Omki also highlights the challenges and opportunities that Viki faced heading into and during the pandemic.

Watch the full video version.

Learn more about Visionular and get more information on AV1.

Welcome to "The VideoVerse."

Zoe: Okay, so hi, everyone, this is Zoe again. Welcome, everyone, to join "The VideoVerse" podcast. Indeed it's also videocast. So today I have my colleague, David, as a cohost for this episode. All right, and then we also, it's a great pleasure again that to invite Omki of Viki and to join this episode. So Omki actually is the engineering director of the Viki team. And for more about himself, I want to let him, I'm being lazy on my part, like him to introduce himself. So Omki, you just want to say hi to audience of "The VideoVerse" and thank you to attend on our podcast. It's great to have you down here.

Omki: Well first, Zoe and David, thank you for having me. Absolute pleasure being here. And yes, as you rightly mentioned, I am currently the director of engineering at Rakuten Viki, which is an OTT service, a unique OTT service that kind of is a Asian drama hub site. And, you know, I've been here for about 10 years at Viki, grown from a role of, well, my first title was video engineer, which was one to kind of get to this space. So it's been--

Zoe: Yeah, we heard about this. Actually, the first time I heard that, I was a little surprised. I didn't know that you were previously just working as a video engineer with the video.

Omki: Well, my offer letter had a misprint, it actually said video codec instead of video engineer. So I was like, "Oh, I'm the codec, I'm going to compress all the video and give it to you." So and I said, "You've kind of mistakenly put video codec in," and yeah, that was actually the role that I originally joined for over here. And it's been the story of my life. I joined for something and I go on to do something else. And it's always been that way, whether it was my education, I started with computer science and moved to signal processing right, for my masters. And I was like, "Hey, what's the worst that could happen?" And then through the course I'm like, "Did I make the right decision?" Oh, yeah, and I think I did. But it's generally been the story of my life where I start with something, the intent is X, and then I'll end up always doing Y. Always been that way. So yes, it's a absolute pleasure being here. And looking forward to chatting more with you guys over here.

Zoe: Can you just talk a little bit, because you mentioned Rakuten Viki, right? So explain a little bit more about the relation between Viki and Rakuten, and then your team.

[00:03:02 Who is Viki? The relation between Viki and Rakuten]

Omki: Yeah, sure, yeah, so Viki originally was just a standalone organization, right? It was in back in 2010 that it was founded and it was VC-funded, trying to find its own space and Rakuten acquired it in 2013. So Rakuten acts like our umbrella organization that provides the primary funding, and all the things, all the benefits that you get by being part of a larger organization. Though we operate as pretty much as a separate entity of our own, we do get, I like to say we have the best of both worlds. We are a small startup nature that Viki gives and we get all the advantages of what something like Rakuten brings to the table. So we kind of get a mishmash of capabilities and I would say, dare even say cultures. A Silicon Valley culture from Rakuten Viki combined with the process and capabilities from a Japanese parent like Rakuten kind of gives us a very interesting mix of things to do.

David: And it might be good at this point as well to talk a little bit about, for our listeners that aren't familiar with Viki as a service, to talk a little bit about where, what Viki is, the type of content, type of products, and things like that.

Omki: Sure, yeah, so Viki's primarily a online streaming site with a focus, really, on Asian drama. We were originally known a lot for Korean dramas specifically. But we're more like a Asian drama hub with a very unique twist on top of all the other streaming sites out there, which is primarily, we do crowdsourced subtitling. And that allows for this Korean drama to be watched through subtitles by people from all over the world. Some of the unique aspects of this subtitling is the number of languages that it gets subtitled to, by just through the power of crowdsourcing for popular shows, it can go up to 50 languages.

Zoe: 50 languages.

Yes, English is still the dominant language. And then Spanish and there's Portuguese. These are the most, they're the most common languages out there in the world and therefore it makes sense for them to be the most popular ones. But really popular shows, there are people in Poland watching it with Polish subtitles. There are people in Germany watching it with German subtitles, and so on and so forth. So this crowdsourcing was indeed a surprising thing when I first landed up here.

The passion with which people subtitle, it is quite intense. It's both exhilarating and scary, in the sense that there are these amazing amount of work put into what defines a good subtitle. In fact, even before the subtitle shows, one of the unique things that we do with our video content is the community creates what is known as segments. So there's a little bit of audio processing that happens to create a waveform-like structure, and they use that as, used to create segments to decide where should be the timeline for subtitles be.

And then they have an entire training course, called the Ninja Academy where there's a 70-page document going through all possibilities around how one should subtitle a segment show. And they tend to be very, very passionate about following these guidelines and making sure the quality of the subtitles, to the millisecond, it's rightly set up. And that creates this really, really unique experience for the viewer that one of the most common feedbacks we've gotten from our viewers is that the quality of subtitles at Viki is par none. And that really doesn't come from anything else but the passion of these volunteers doing that on their side.

David: I so have a question about, sorry, I have a question about how that came about 'cause obviously, that's not a feature that just kind of pops up and it's like it's ready to go to market. Did that feature purposely get developed or was that at a request of your viewers or was there a business decision that said, we have so much content, but yet doing the subtitles ourself is just an astronomical amount of money? How did that come, how did the product feature kind of come about?

Omki: That's a good question, David. So that's actually the genesis of Viki itself, where the very first product that Viki built was the subtitling tool and the very first content around which it was built was actually freely available content on YouTube. So it actually started with the concept of providing fan-based subtitling of content that they couldn't themselves watch because they didn't understand the languages. So getting the volunteers to kind of subtitle existing content out there and Viki actually started with the tooling around it and the founders then had contacts with the Korean industry, content industry to then go license content and with the idea that hey, you're developing content for Korea, but what if we took it to rest of the world through this capability as well. And so it wasn't, it was actually the genesis of Viki, right? And always been at the heart of what we do, the community kind of taking content in one language and making it available to people all around the world in other languages.

Zoe: Yeah, I actually forgot to mention, so where are you located right now at this moment?

At this moment I'm in Singapore. So that's where our primary team is based. We are also based in US and Korea, different teams, but there's much smaller teams. We are primarily based in Singapore.

Zoe: Right, so that's where you mentioned that your old source originally is a mainly Korean drama. Right now it's mainly Asian drama. The hub, so Code.Hub is now from there, from Viki, you distribute the content to all of the world. So what are the major, for example, top regions of the viewers, the physical locations for those viewers of Viki?

[00:09:03 Major viewers of Viki]

: Yeah, so interestingly our top regions are more on the Americas, North America, Latin America, South America, you have Europe as well. Well, one of, even Asia is certain places like India, there's huge demand for this kind of content. So interestingly there's a lot of demand for this outside of Asia.

Yeah. So we're based, I would say we're global pretty much operate in most countries around the world and the audience does tend to concentrate in clusters around the Americas and Europe.

Zoe: I see, then based on that and that's why I think one reason that can well justify why subtitles are so important. That's become, I think heart and soul of this product that you provide this Asian content as a hub unto mainly for, because David and I, here we are located here in the California of America and then here the viewers can actually, they may not know the Asian language at all, but they can enjoy the drama and the great stories and then, yeah, and enjoy that.

Omki: Yeah. Absolutely, know that it shows up even in their feedback to us, it shows up there's an issue. I would say the way our fans of the content react to when they're not able to get their content could almost be a script by itself or a drama. But it shows you the power of content itself. And it's so universal that it's beyond a given language, it's just, it's written in another language, but the emotions or the things that it conveys is absolutely taken, it's something that connects to audiences in all countries, in all languages. It doesn't stop at just at a particular language. And you're right, Zoe, you don't know the source language, Korean, Japanese, Taiwanese, they still kind of are able to watch this in their own language through subtitles.

Zoe: Yeah, I heard that some of the audience mentioned that not only they enjoyed another country's, another region's culture, some kids or young people even adult, they want to learn the language. For example, they want to learn Japanese, Korean or Taiwanese like Mandarin and they can actually enjoy watching a drama while trying to learning the language and at the same time they can learn because it's language. And then there's another translation right there by the subtitles. So that's, yeah.

Omki: Absolutely. We've gotten feedback on this. We've tried experimenting with the dual language subtitles as well. Really, so to provide capability to kind of play two subtitles simultaneously so that a learner could kind of use that. It's not a very popular feature. Those who actually use it use it a lot and some do not, I mean not everybody's obviously watching drama just to learn. Some are just watching it for their own entertainment.

But you're absolutely right. people have used that feature and they're quite passionate about it as well and yep, they tend to write back telling us, I've seen tickets out there where they've just written appreciation for it and they're like, "Hey, I'm going to Korea and I'm going to Korea because Viki introduced me to Korean dramas," and I've seen viewers write in because of that. And it's nice to see that, right? That the influence that this kind of platform is creating that's beyond just watching the content but the spreading knowledge about cultures.

David: So I imagine with all the, originally when you were talking to the content producers, you had to educate them quite a bit 'cause you were probably one of the very first ones that said either probably give me clean content, don't give me burned in subtitles, right? So talk to me a little bit through about if you were there at the time, what that was like to work with the different content houses to get the content in the manner it's actually usable to then start the subtitling process.

Zoe: It is a wonder how come you have so many in-depth thoughts about subtitles? Those are actually now the questions I like to learn about the answer.

[00:13:29 History of subtitling process]

Omki: Absolutely. So to David's question at first. So, I did come in at 2013, so it's been 10 years here, so I feel like a slight old fart at Viki, but I've definitely seen the challenges the content team has faced. So I don't necessarily get to face these head on being more in the engineering role, but I've had chats over the years and they do need to educate the content providers in terms of what the goal over here is. And we've had things like, oh we don't want people to subtitle in X languages. And we're like, why? Why would you not want to subtitle in certain languages?

And I think traditional media took some time, but this was not just about Viki, this was the industry itself where they had to learn to trust the streaming model being a safe and viable model over the 2010s, I would say early 2010s. And just as Netflix was taking off and the other large streaming sites were taking off, traditional media did not necessarily see streaming as a partner. So I think our content team definitely had to go ahead and face those challenges and convincing a lot of, I would say non-western media partners in terms of, hey this is doable. You can take your content over to other places. Like why is there an Argentinian out there trying to watch Thai content for example? And, but there is, it's because it's human stories.

So, though I've not been privy to these conversations I've heard over the years talking to them on the challenges they definitely faced and trying to convince. I think now it's easier now that the power of the Internet, the whole streaming revolution that has taken place, I think everybody wants a piece of the pie now. In fact it's pretty obvious at this stage. I think during the early years, what helped I think the most was the founders and some of the early folks were tied in with the content network itself, the net, the people that they were familiar with from their previous roles or they had connections with, there was a lot of connecting out, reaching out and trying to get access to it.

And during the earlier years I would say good, a good number of challenges faced, but I would say very smart choices also made by our content team because they had to pick and choose. We were very small, so we had to really pick and choose what content will work for us. We couldn't go for the entire catalog or say, "Hey, we want everything." So they had to be really, really smart about, I only have so many dollars for so much content. And I think they did a brilliant job in trying to pick what would work and sometimes we got very lucky.

There were shows that didn't do well within the Korea primetime market and they're probably one of our biggest hits for the rest of the world. So you get lucky with it, sometimes you can, with content you can never tell what's gonna work, what's not gonna work. And there have been instances where on the prime market that the content was created for, it's been a dud, but on Viki it just blew up and I mean, it blew up, you know? So, it is one of those black arts of deciding what content to get.

Zoe: Yeah, but what are the metrics you collected in terms of, for example, how popular a drama is in terms of the feedback of the viewers?

Omki: That's a good question, Zoe. I think, of course there's just general watch time that people are watching, but there's also a lot of signals on, people do write to us what drama's upcoming. So the whole, the Korean drama, the Taiwanese drama, Japanese drama, they do publish their calendars in advance. They do their own promos in terms of there is this actor or actress now coming up on this content and that tends to create an excitement, a buzz and users start writing in, "We want Viki to license this show." So we kind of get signals like that upfront.

We have, we just don't host shows, we also create these basic fan pages for the celebrities themselves on Viki and people follow these celebrities, they follow them and they're constantly chatting, whether it's on the comment section or the chat section where they're constantly putting out that I'm looking forward to certain shows. So we do get metrics beyond just pure watch time, which is post factor, people start writing in and create a buzz on, "Hey we want this content," the discussion forums and around which all these discussions do happen.

Zoe: Is this mainly subscription based or with ads for free?

Omki: Good question, Zoe. I think it's both ways. So we try to cater to both markets. We do have a ad supported business and a subscription based business. So certain content is put behind the paywall to kind of, prime content that might have been really a good conversion content to get them to subscribe. But there is enough depth in the content that we have that someone not being able to afford the content because by being a global service, there is income parity that there is lack of income parity across the globe in terms of people who want to watch, someone who can pay $5 from say the United States. But if you're coming in from say the Philippines, $5 is still a huge amount. So there is enough depth built up with the ad supported model as well and you could come in and watch on the app today without subscribing and I feel there's enough library content out there to kind of start catching up on Korean drama without subscribing.

Zoe: Right, so even for this kind of like business model, you had to adapt to local culture and that to use the behaviors.

Right, yeah, yeah, no we're definitely facing challenges around just enabling that. So as a business we are trying to enable different payment models that are local. Some people are not, some countries don't have the culture of using a credit card for repeated subscriptions and local payment methods. And those standard challenges with any product that you're kind of trying to go global do exist with us as well. And is constantly trying to figure out ways to get that, or get our users to overcome those barriers. Whether it is one time payments, local integration payment methods or whatever that come with it.

Zoe: Yeah. So because this podcast is about video, so now you distribute videos across the global regions and back to that topic still and because you also were under the title as video engineer, so we want to talk about, I know that when talking about video of course is naturally 'cause for those audiences already attended our podcast and all, videos always face challenges because it demands a great amount of bandwidth to distribute. And then now you also mention you have users and users really distributed across different regions. They have different devices, different situation from their networks. So these are common things I believe your system had to address, but on top of it, of course you can talk about the common things regarding the video technology, but also unique things that Viki is facing and how you're going to get that resolved.

[00:21:16 Unique challenge that Viki is facing and going to resolve]

Omki: Yeah, absolutely right, Zoe. So, there is those standard things of CDN integrations. Try to minimize that number of bits you are trying to send to the users. So balance between the cost you're going to pay for streaming and the experience that you want to give to the user. And we've had our own unique challenges around it. By being global, being exposed to a variety of devices around the world with the Android fragmentation compared to say the iOS 1, which is much more consolidated. We've had all the standard issues.

What we've done I think specifically for the device side of things, we built our own in-house device database to kind of manage the way we stream video because as you know on video you have these various profiles and capabilities and while devices may advertise a certain capability, not all devices are true to their word. So they may say that I can play high profiles just to go bit technical there, right? But they may not and it turns out the decoder has an issue and the end user doesn't care. He's like, "I want my video now." He doesn't care that his device is not working. All he says is, "Hey Viki, you're not working."

So very early on we started building our capability to just capture the capabilities across devices. We started capturing statistical signals of what video is working on what device and create a fallback mode for the video to play. So this was with the intent of getting higher reliability on the playback side. What we've seen out there, and I think it's still quite common, is all these exceptions are usually done on the player side. We kind of built it on the server side. So the idea is if a certain device model is not failing to play a stream X number of times, then we do a fallback to a maybe lower capability stream.

So you go from a high profile to a main to a baseline and try and see, or there's also a fallback, like some devices can't play HD, they just might not have enough ram. A lot of users have, we have a lot of low end devices. So, the spread is actually quite huge in terms of devices with low amount of ram. So getting them to play 1080p or even 1080p with DRM is a challenge. So then discovering these needle in a haystack problems across all of them was definitely a problem for us. And building the server side capability where you can say, "Hey, this new device has come in and we tried playing a video five times over five different users and it's failing. So let's do a fallback of instead of serving X automatically fallback to serve Y," and that takes away the issue and then you can always build the capability to try X again in the future to kind of up the video profile that you're serving.

But just having that allowed us to address a lot of issues, which was quite, just because of the nature of the service, that it's free, it's available for low end devices, it's available globally. We couldn't target a very limited set of devices. We really had to target a lot of them. The second interesting challenge we faced around what was unique to Viki was because of the subtitling, the way we want to render the subtitles. So David originally mentioned like, I don't want subtitles hard subbed onto the video. So obviously we wanted content that came in without the hard subbing and then we had to render it, that kind of, with each device that we get that's more of a graphics overlay, but over time the player capabilities have improved right now, you can provide these subtitles in a manifest and it'll auto render it on that.

So getting the subtitles to render correctly was more, not so much of a video problem, but ultimately to the end user it's a video problem that the video and the subtitles are not in sync. So just a lot of tweaking around, again because the device diversity, getting it to work reliably was a second challenge. And the third interesting challenge that we faced that I don't know of too many OTTs that do that just by nature of the licensing and its global nature, we also sometimes get the same content from two different licenses for two different parts of the world. But the video files are not exactly the same, but we are only allowed to serve one in one region and another in another region. And scenes maybe out of order or some scenes maybe cut, but we want our volunteers only to subtitle it once.

So what we interestingly do behind the scenes is we do a video diff. So just like you do a file diff, we actually do a video diff.

Zoe: Really?

Omki: Yes, so it's a in-house developed algorithm where we do a video diff and a tool on top of it. And of course like any other statistical algorithm, it's not perfect. You could always gimme content where it doesn't work reliably. But we do a video diff and effectively the community subtitles one video and then with the diff and a little bit of human input to just validate that that diff looks good, maybe a minor adjustment. The diff is applied on top of the subtitles that you've done for one video to serve subtitles on the other video, which is where the scenes may be in a different order or certain scenes may be cut out.

So it's a in-house thing that we've kind of done it. It was an interesting challenge because of the way the content licensing works. We couldn't get the same video file to serve globally, like maybe music rights or just content rights by itself. But we don't want to sub volunteers to kind of subtitle it twice or try to figure it out. So we took the challenge in-house and it's an interesting problem to kind of say, "Hey, I have one video and I want to find all the scenes in the other video and I need to find them so that I can switch over if the scenes have moved over, I want to play the subtitles correctly from the subtitling of the first video." So these are kind of like interesting challenges just from, by the nature of the business that we faced over the years.

David: So speaking about challenges over the years, obviously we've gone through a very kind of pandemic now, we're come out the other side, I'm curious, for a lot of streamers when the pandemic started, business went crazy in terms of volume 'cause people were stuck at home. I'm curious, what were the effects of the pandemic on Viki kind of at the start of the pandemic heading into it and then now coming out of the other side of it?

Omki: Yeah, no absolutely. I think we got a preview of that when one of our competitors suddenly shut down a few months before the pandemic. And interestingly it was the day that AWS had an outage as well and YouTube had an outage as well. So our Twitter blew up in the form of, and we obviously got a huge spike of traffic, but we were kind of ready for a spike in the sense that we had just migrated our infrastructure over to be a more resilient, reliable infrastructure a couple of months ago, it was a year long project we had started to say, "Hey, when a growth of spike comes you need to be ready for it." And here was the moment that it came. Now of course it was the first time it had come that way and it was also the day, like I mentioned, suddenly YouTube had an outage and AWS had an outage.

So our Twitter just blew up with dramatic tweet in terms of, "Hey is this the end of my drama life?" kind of tweet, people were like, "No," like you could hear the "No." Through the tweet going out there. And so we had got a bit of a preview, one of our competitors kind of shut down and we were like, we are kind of ready for it. But you're right, the moment the pandemic hit, just like many other content sites, we saw a huge bump in traffic, which was great for business. It was also good to know that people are finding some kind of relief in just watching these Korean dramas and they're enjoying it as well.

We had to kind of make even certain key decisions because the volume just grew so much that there was also a cost imbalance in terms of serving these bids. So then as an engineering team, we had to find the right balance between what is the resolution we serve our users given this change in sudden traffic volume across the globe and it was not uniform across the place. So it was quite different and we actually turned a few knobs like on mobile devices. Android was one of our largest platforms. On mobile we went and lowered the resolution so that the buffering reduces, everybody gets a free flow of video but the resolution is lower.

What was very interesting, I was expecting some people complaining, we didn't get a single complaint and it just taught us a lesson that the smoothness of the video is so much more important than the absolute quality of it. And we turned that knob immediately into the pandemic and not a single complaint ever from, I don't know whether it's just the upscale of maybe on the Android devices are so nice that they make the video look good. As a video expert myself, I can tell that okay this is definitely a role resolution in that. But it was just interesting that people were just happy watching and as long as it kept streaming and I guess that was the case. People in the house watching three, four devices simultaneously, they just wanted to play smoothly, not necessarily, and especially if you're watching on the phone, it's not like a very large screen where you need the 4K or 1080p kind of quality. You're just happy to get consistent and acceptable quality. So that was definitely a big change that occurred during the pandemic. Post pandemic, I think we've seen some fall up for sure as people have moved on to the real world.

Zoe: Right. Everybody they just were talking about that. Do you also face cost saving or something like that along that line?

Omki: I think we were ready. So the good part is Viki has always been very cost conscious in its decision making through the years. We're a very small lean team. We're actually just for context, 30, 40, 30 to 40 engineers across the board, that's it. So we've always built things with cost as a variable because we never really had VC funding, we had fixed budget for the year and when we are building solutions, cost was always an equation to consider when building solutions, so luckily post pandemic, Viki's still in a healthy state and we are not looking at, we're still hiring just to give you that context compared to the world outside. That's a good place to be in.

But I think it comes from being frugal and conscious about what we are doing right from the get-go rather than reactive. So the world's kind of been a little reactive to it. I think we were optimizing through the pandemic, pre pandemic, we were always optimizing our streaming capabilities, our engineering capabilities to give us maximum ROI, not just throwing money at a problem and trying to solve it. So that has helped in today's environment as well.

Zoe: So back to again, back to the challenges because you said you are hiring, meaning the business are growing and then so what are the challenges, especially on the video set because we still want to get back to the videos. So regarding the, I think the content, the, also relate to subtitles, the different content serving different users geographically located in different places and different situations. So all of this as together.

Omki: Yeah, absolutely. I think it's a mix of business product and engineering challenges that need to be solved here to get to the next level. We still, I would say we are known a little bit within the Asian drama circles. Viki's still quite small compared to the giants out there in the streaming business and obviously that means there is room to grow. What we are trying to do, especially on the video side, is get to a place where our quality of video experience for our end users is really on par with competitors out there with what the users have gotten used to expect.

If you just think of it where videos come through the years. Right now you can serve a decent 1080p at one and a half or two Mbps, which several, a decade ago, which was unimaginable. If I just look at it from the journey of, the video experience itself has become so ingrained into everybody's experience in terms of expectation, the quality of the video experience, whether it is this video recording that we are doing here right now, it's just, it seems natural. It's no longer a challenge compared to where it was. So just I think the bar for what is acceptable for our end users is very high now compared to a decade ago when we started out. Right?

So that that maintaining that bar is a challenge. Especially with the way the ecosystem keeps evolving. You will have needle in a haystack problems, suddenly in Italy whoever devices will stop working because there's an over the air update given and that breaks something. So our goals are constantly to, at this point, just improve the reliability of our experience while trying to balance content provider needs in terms of they have security needs whether it's DRM, whether it's limiting the number of streams users can watch.

So you kind of need to do a balance of business needs from the content provider, from the studio guys versus what the end user wants, which is everything everywhere all the time. Just trying to find that right balance to then do it in a cost effective way that satisfies all parties. And that includes interesting new things to, well 4K is now quite common but then 4K really makes sense for a large TV experience, we've done visual experiments, we recently ran with our end users. We ran quality experience and we got them to kind of tell us are they able to see the difference between stream A and a stream B.

And what we realized is beyond a certain screen size, it's intuitive but we had to do it anyway. Users can't really tell the difference between a 720p and a 1080p and a 4K if the screen is small enough. So yeah, it's just taking those learnings and saying okay, now where's the good sweet spot of how do I sell this user good enough video that he feels he's happy? So it kind of optimizes what the business need is, which is give the best quality at the least cost, but the user shouldn't feel he's short charged, that hey, I'm not getting a good experience over here. So whether it's the reliability because of the OTT updates or the device updates, the OS updates or whether it's just the quality of the video experience. I think in terms of making it look, yeah, I'm gonna lean back and just watch on Viki because it's a good experience for me to do that. Y

eah, that's pretty much the overall thought and it's a continuous improvement cycle of finding marginal gains everywhere. And then once in a while it becomes and paradigm shift by changing something like bring in a brand new codec, we are waiting for AV1 to become popular enough on the devices so that we can shift to AV1 and that kind of is a, that's gonna be a big shift when it comes in terms of just the volume will fall by 20 30% from where it is in terms of the streaming volumes, but instead of just counting it as savings, we could also use it to enable capabilities for our end user. So that's what I'm looking forward to next.

Zoe: Right, you do, and you mentioned about AV1, just for our audience, as a new video codec center and so you're looking forward to leverage these new technologies. The video, codec technologies to further bring, I think boost up the user experiences. I do have one question indeed, beyond the video compression. So how do you think the current trend, GPT, that can potentially be leveraged in the Viki platform or validating some of the the user potential?

[00:38:34 How can GPT potentially be leveraged in Viki platform]

Omki: That's a very nice question, Zoe. It's one of the things we've been pondering internally too. There is so much buzz about the GPT out there and just, we've been behind the scenes trying to figure out can we do machine translated subtitles. And can we parse the video to build intelligence about what is being spoken about by looking at what's on screen to improve the quality of the subtitles. Not just audio, but can you use the visual cues as well to kind of improve the quality of the video itself. So there is, these are all ideas at this stage at the end of the day, it has to be mixed in with the right business knobs as well.

I think we are at early stages where we think there is potential. We've definitely tried machine translation of all our content in the past, it's not been satisfactory enough, though we've started approaching a point where we feel for maybe not so popular content. So there is content on Viki, which is not so popular that's happens to everything. There's some content is really popular and some isn't and that doesn't get the love to get subtitled. Everybody wants to subtitle the popular content. And what's interesting is the Asian languages, their processing has definitely gone a notch up.

In the past a lot of the AI around the world is to focus a lot on English. Even if it was voice recognition, it's around English, but if you're looking at Japanese drama, Korean drama, Chinese drama, Taiwanese drama, you're looking at Asian language detection. So you wanna kind of take the processing of Asian languages and then get a good translation model going and that's been challenging, so we do have internal teams, Rakuten has, the Rakuten Institute of Technology that's focused on these problems as well for us.

And I think, I'm sure we'll find ways to use GPT to kind of merge it with our passionate volunteer army out there. I would say is it a enhancement or is it a separate thing? Do we use it for the non-popular content and the popular is still human driven because there's so much passion around it or do we leverage it to kind of combine it with the human to accelerate their workflows. And I think it's gonna be interesting because some volunteers see AI as a threat. Some volunteers see AI as a tool, so there's, while there can be technology, ultimately the human using the technology, their feelings and their emotions will decide whether they're gonna adopt it or not.

And it is gonna be interesting for sure. On the pure video side, I'm looking forward to, like I said, how do I utilize some, I wanna build something like, I love this feature that Amazon has that's called X-Ray. I don't know if you've seen it, where while the video is playing it's giving you information about the video, the landmarks or the context or you mention a celebrity then there's a celebrity there and I'm like, oh wow, that's amazing to see that where technologies reached a point where a decade ago or two decades ago when we were watching content, it was so simplistic and you had to figure out all of this by ourself, but you've got it to a place where as you're watching it you can create all these live streams on the side of what that content is. Personally I'm very enthused by an idea like that.

Zoe: So basically I think that describe something, for example, I'm watching a video or I'm watching like I'm talking with you and then there could be another GPT too right at the side and then analyzing the conversation, then remind me, hey can you give this most challenging question to Omki? So that we can dig out something very interesting or more interesting then. So there's a lot of things that along that way and then I think we already like got quite into the time pass by. So as you also, you already mentioned that based on your experiences you are combining Argenova engineer and then product and business into this, and then just mention, because right now like a lot of young people, 'cause it's challenging times, I brought out like GPT also mentioned that there's a lot of potential down there and then we can't just, just to do something just for the sake of leveraging GPT, right?

There's a new request coming out from the market and then of course the technology driving the market needs on the other side and then there's some point brought out by the users that they request, they deserve, like as you mentioned previously, do you think 720p is good enough, then there's requests and resist to high resolution, higher quality to 1080p, 4K, HDRs. So then users really enjoy that. So I just wonder to close, I'm not sure whether David has some more questions. I just wonder what do you expect? Within say five, 10 years. What do you expect in your position for Viki and together with associated with technology, subtitles and AIs, video technologies, what do you think are going to be the next step. Right next step, this is something that we want to improve than from the technology point of view, this is something that we can do so that to further boost up the users' experiences.

[00:44:50 Next step to further boost up the users' experience]

Omki: Good question, Zoe. I think there's so many parts to choose from, but specifically if I had a choice where like you said, mix of business product engineering, I think one of the things we are finding interesting and it's an industry wide trend, there's a lot of viewing is shifting to the large screen. While mobile has taken over the world, the mobile is, there are pockets of the world where the mobile is still the dominant viewing device. But-

Zoe: So what's a percentage if you can disclose about Viki's viewers? Because you mentioned lot of variety of devices. So I was about to ask, I was already thinking because it's OTT, Korean drama and then especially you mentioned users mainly for example at least seems like the top regions is America. And then I would have expected most viewers watch movie, watch these dramas over their laptop or on big screens. But you mentioned a lot of varieties on the devices side.

Omki: Yes. But it's very spread, Zoe. So I would say it's evenly spread. It totally becomes which part of the world you are, the parts of the world where Android is 90% of the viewing. So on Android phone is just 90% of the viewing and there are parts of the world where it's evenly spread across the web platform. So that's your computer, desktop, TV, mobile. So it's very hard to kind of say it's only one, definitely TV has seen an uptick over the past couple of years. And it shows up in the industry-wide reports as well.

If you look at Conviva state of streaming or these other streaming reports, it's pretty much in line with that, that the TV experience is be becoming more and more common. Which yeah, it effectively becomes lay back and watch rather than necessarily interact. But what it's done is it's created a second screen experience opportunity as well where the mobile now is the interaction touchpoint to what you're watching on the TV and we are seeing trends of that obviously and there's a lot of thinking around that as well and how we could utilize that where your mobile is effectively your remote control in a way with the TV over there in front of you.

So second screen experiences, I would say seamless scanning over across devices, still a challenge, a little bit of a challenge. How do you carry over? It's still, there's, people expect it. I've watched on one device, I go to another one, of course it's there. You could always make it more and more seamless. People want to carry the context wherever they're going, which is now the default, everybody does that, we do that too. But the expectation of what default on that has risen the bar. When it was a novel feature early on, but now it's become something, yeah, it's become table stake.

So using technology, I would say the first thing is just make, keeping that bar, for a small organization like Viki, I would say it's an interesting challenge that we do. We do a combination of build by, we can't build all technologies ourselves. We need to buy some of them. We use certain in-house capabilities to where we can create differentiation. So personally for me it's always been, okay. at this point what's good enough for now, but what's needed for five years down the line, what's needed for two years down the line?

And I don't wanna go a decade because no one's seen what's happening in a decade. Two to five years is probably the best anybody is willing to, I would say would have the ability to kind of forecast a bit. So it's just, it's a challenge between what's an incremental improvement versus what's the revolutionary game-changing thing that's coming that might just shift the earth under us and we need to be aware of it and just being in touch with everything out there, avoid that from happening and being preemptive about it.

Zoe: Well, thank you, because you always mentioned you have a lean team but actually because the way that Viki is serving like more than 50 subtitles, that means there are more than 50 kinds of different languages serving different regions, different cultures, different users with the larger variety of end devices. I think through this like I will say small window of the Viki platform and then there's lot of novel technologies will just find their own spot to serve the end users, to boost up the whole experiences and then we can observe and follow up what Viki will come up with the next step.

Omki: Thank you, Zoe. Thank you, but, yeah, definitely, we should catch up sometime in the future.

Zoe: Yes, thank you for being in our interview and then thanks for my co-host, David. So then I think then we'll put closure in this episode and then yeah, we wish the best for Viki and we also wish the technologies will drive even more users to enjoy not only the Viki platform, but actually enjoy the new technologies, to enjoy what the new technologies has brought up to the world.

Omki: Thank you, Zoe.

David: Thanks, Omki.

Who is Viki? The relation between Viki and Rakuten
Major viewers of Viki
History of subtitling process
Unique challenge that Viki is facing and going to resolve
How can GPT potentially be leveraged in Viki platform