Welcome to NETINT's Voices and video where we explore critical streaming related topics with experts in the field. If you're watching and have questions, please post them as a comment on whichever platform you're watching. We'll answer live if time permits, otherwise we'll respond after the show. Today we chat with Krishna Rao Vijayanagar, who is founder of OTTVerse, which covers OTT technologies and business topics. By way of education Krishna has an MS and PhD in electrical and computer engineering from the Illinois Institute of Technology. I met Krishna years ago at one of the several positions he's held in both US and Indian companies. Today our conversation will focus on streaming in India,
which now has the largest population in the world. We'll cover the market forces that impact OTT strategies in India and then focus on the technical decisions that these market forces entail, like codecs selection, ladder formation, player support, CDN selection, and other issues. Then we'll spend some time discussing OTTVerse and particularly the issues that OTTVerse readers are finding most compelling today. Krishna, thanks for joining us. Thank you so much, Jan. Thank you Anita as well for having me on this podcast. Okay, well we're glad to have you and I know it's late for you and we appreciate your working with our time constraints. But you and I spoke about this a couple days ago. It seems like the Indian market is a very challenging market to get into in a number of ways. Could you
cover that? I mean, what's the Indian market like to get into as a content provider? India is a very beautiful, complex and ancient multicultural country. That creates an atmosphere of challenges and opportunities as well. To give you an example, we have approximately 26 different languages each with their own script, each with their own vocabulary, each with their own dictionaries. So when you want to communicate with somebody, and all the states in India are divided based on linguistic, based on the languages. So I come from Bangalore, the tech capital of India. If I go to my neighboring state, there's a very small likelihood I'd understand their language. So if you think about this in terms of OTT, if I produce a movie in my language, distributing it to other parts of the country becomes a challenge, primarily because people wouldn't understand it.
So this, though it looks like a challenge, it's very interesting from a technical perspective, brings in interesting concepts of dubbing. Can you create one language and translate it or dub it in five or six different languages? Can you do close captioning? Can you do subtitles in different languages and different scripts? How do you deliver all of this? I was just thinking about this with a couple of friends after our discussion, and we realized that several OTTs have to struggle with the UI. So in the US you're probably accustomed to having the UI in English. In India, most of the OTT players have a English UI, but how do you change the entire UI to a different language, and how do you do it for five different languages in India if you want to be a national player? So it's a beautiful complex country, a lot of challenges, and it's a good place to be if you want to solve interesting problems at scale.
So I leave it at that and we can pick up these topics one by one, I guess. Okay. What about pricing? Because it felt like pricing was also pretty pressured over there. I would say most vendors who have probably come to India have also faced this. So India is a price sensitive market. It's primarily, because the population is very young,
they're coming out of the towns, villages coming to the metros. Recent stats show that 65% of the population is below the age of 35, so it's a very young population earning their first dollars, getting their first paychecks. So people want to pay less for content. People appreciate free stuff in India, so it makes it quite complex to run a business over here. How do you produce content? How do you set up the technology, how do you deliver it? Do the marketing, while charging something which is sensible. So for example, Disney plus Hot Star,
I think their top tier comes roughly to a few dollars per year, probably 10, $20 a year. So that's how competitive it is. On the other hand, you have forces like Geo Cinema who's running this year's IPL and running their entire app for free on a pure reward model. So this is a conglomerate with very, very deep pockets. So they can do a market exercise like this to capture the audience. So if you are an OTT, if you want to enter the market, how do you actually counteract such a force? So that's something interesting to think about as well. Who's doing a good job coming in from the outside? Here in the States we have the big names, Netflix Prime, Hulu, Paramount, who's doing a good job penetrating the Indian market and what are they doing to accomplish that? So when you look at the Indian context, it is interesting. You have to look at it as the national play versus the regional play,
just because of the language differences that I spoke about. So if you look at the national players, there are the likes of Zee, SonyLIV, you have ALT Balaji, you have Amazon Prime video, which does a very good job aggregating content, JioCinema, so these are interesting players. Then you have the regional players like SunNXT, which is for the Tamil market, AHA, which is Telugu, you have Marathi, you have Planet Marathi, then you have Gujarati, ShemarooMe. So you have a bunch of regional players who are doing fantastic work capturing their audiences. And then you have a very interesting phenomenon in India where the telcos are aggregating all these OPTs. So you have these super aggregators, so JioCinema is one, Airtel Xstream is another one,
Tata Play Binge, Hindustan Times, OTT Play. What they do is they tie up with all these OTTs, they provide all the content on a single app, through a single searchable interface. So they charge a single price. I remember Airtel Xstream would charge roughly one and a half
dollars a month to access 15 OTTs, more than 3-, 4,000 titles. So it's a price war on one side, market forces. How do you compensate the content creators? How do you still play your bills and keep the lights on? It's a market trying to figure itself out and it's in a very interesting phase.
Let's start to look at some of the technical issues that producers face. What's the typical bandwidth that people are streaming their videos over? Again, if you look at it geographically, we typically divide India into the metro cities, which are the largest ones like Delhi, Mumbai, Calcutta, Bangalore, Chennai. These regions typically have pretty high bandwidth, good mobile penetration, all of this. Then you have the Tier one, Tier two, Tier three cities where the bandwidth drops. But from a business perspective, many people would agree that the metros are saturated.
People have two, three subscriptions and that's not where they're going to get new subscriptions. They have to penetrate into the cities and the towns which do not have very good connectivity. So this goes into the mind when you actually decide bitrate ladder, if that's what you are alluding to, the bitrate ladder and the bandwidths.
So typically you would see people streaming at 720p around the 1-1.25 Mbps range and a lot of lower bitrates like the 640p, 480p, 360 goes down. Your 1080s are generally between 2 and 3 Mbps and with the caveat that I'm only talking about H.264 right now. For mobile consumption being around 2.5 about until around 3 Mbps. And then when it goes to large screen, it comes to our 5, 6, 7 Mbps.
Looking at the encoding ladder, are companies doing a different encoding ladder for metro areas and also for the outlying regions, or do they have one encoding ladder that's going to serve everywhere in India? From what I know, and there might be a few exceptions to this, it is typically on the device profile, rather than the geography. This is what I've seen. So you have these per device manifest sort of configurations. Where when you're using H.264, you cap it at close to 3 Mbps for 1080p, and you can go higher for large screen. And I
should add that India is a mobile first country. There was a report by Hotstar two three years ago, Disney + Hotstar where they analyzed an entire year streaming and they said that 94% of the subscribers accessed it through an android form, through a phone, and most of it was Android. iOS has a very low penetration just because of the price point, but if you look at it that way, it doesn't really make sense going up, having very high bitrates, HDR, all these capabilities, when 90% of your subscriber base is going to access it on a handheld phone. Typical ladder might go up, what's the top bitrate you would see? I think you said five megabits per second. Is that where you're capping out? Yeah, five, especially if you're on the
large screen. But with the OTTs I've worked with and I've inspected just out of curiosity over the last few years, it's typically been 2 to 3 Mbps the top bitrate. The quality on your mobile phones is fantastic on the laptop, it's great. It does the job. What's the player situation look like? You mentioned that most of the playback is on Android phones. Apple is very premium, very costly, here in the States. What about smart TVs and OTT devices, even in the big cities, are those making penetration? They are. So I was in a talk a couple of weeks ago where a person from Samsung Ads actually said that there are 14, 15 million TVs being sold every year. So it's not well penetrated in the
Indian market, primarily because it requires data, you have to end up using your bandwidth or your internet connection. So DTH and normal linear TV, are very, very prevalent in the towns and villages of India. Smart TVs are catching up as data is getting cheaper and as bandwidths are improving. So we aren't there yet. Cord-cutting isn't isn't such a big phenomenon as it is probably in the US. We've talked about the streaming side, what about the broadcast? I mean, is there a broadcast infrastructure over the air that makes up a substantial percentage of what people are actually watching? So Cable TV, IBTV is very, very common over here. It isn't as expensive. I remember when I was in the US I had a Comcast Xfinity connection a hundred dollars a month and I'd get internet, probably a phone and a lot of channels. It's
really not that expensive in India, it's probably 300 of 400 Rupees, which is roughly around $5. And you get a host of channels, probably a 100, 150 channels. And again, it was government regulation that you actually get a sheet of paper to your house where you can tick and you can mark what channels you want and only those will be provided to you. So I still remember doing that. So in our house we speak couple of languages, just because how my parents came together. So we just choose channels from those two languages and we're done. We don't need all 100, 150 channels.
And that's all MPEG2 or H.264? I believe it's MPEG2. What about Codecs over there? I mean I heard talk a few years ago about VP9 making good penetration because A, it's supported on Android devices and B, it's more efficient than H.264. Are you seeing other codecs being used over there? I was speaking to a friend a few hours back and he said yes, there's a difference between penetration and adoption. People are trying to adopt new codecs like HEVC, but generally when you deploy, when you're deploying on smartphones, AVC is good enough, you don't have any royalty issues and for the bitrates we're streaming, I think it does a perfectly good job. Plus hardware support is there, it's a legacy codec by now. So a large number of these mobile devices aren't your high-end Android devices. It's not your Samsung or your OnePlus devices. A lot of them are Chinese phones, which you get for roughly $50.
So massive consumption on those devices as well. So anyone trying to deploy a codec has to keep this in mind. They might not have the best DRM support on board as well, so that restricts you on what you can actually do. You might not be able to serve 1080p on certain devices,
just because of the hardware capabilities. We talked about the high end of the encoding ladder, what's the low end of the encoding ladder, what's the resolution and bitrate? I have seen it go down lower than 360p at some cases, but 360 is roughly where I see it cap off, 360 at roughly 200, 300 kbps. I don't see what people could make out of that resolution, but yeah, that's lower end to keep the streaming going I guess. Yeah, I remember when 640 x 360 was pretty high quality. When you're streaming at 200 or 300 kbps,
it feels like you want a pretty consistent stream. Are you seeing a lot of CBR over there or is it 2 pass VBR or even exotic things like constrained CRF? So speaking from experience and having been a vendor in the compression space, people do pick up, I would say platforms like Elemental, you have few people running theirs off of Brightcove, certain other vendors in the space. So whatever they've provided is typically what they use, but people who do-it-yourself or choose certain vendors in India, I've seen that the common use case is either CBR, not 2 Pass, but plain CBR or capped CRF. I personally have dabbled with capped
CRF in my previous role, we found it to be a very good approximate to CBR in terms of quality, it was very good. It didn't overshoot the constraints that we had put too badly, and was perfectly good enough for streaming in India, without any buffering issues or any of that stuff. Can you just briefly describe what CRF is and what capped CRF is for people who may not know? CRF, so we are talking about constrain rate factor. Constant rate factor. Constant rate factor, right? So this is a mode in FFmpeg, which is probably the most popular open source implementation of video codes out there, where essentially they try and hit a video quality by adjusting the number of bits allocated throughout the video sequence. This is in contrast to CBR where CBR tries to maintain certain bitrate
by adjusting the video quality. So you would start off, let's say you start off with an I-frame that would get the highest quality, then your P-frames would get the most allocation of bits and then it'd be distributed amongst the B-frames trying to hit your bitrate caps. CRF obviously has to do similar stuff, but it places a greater emphasis on the quality, rather than maintaining a very constant bit rate. So you might see it overshoot, you will see it overshoot actually. So modification of this is the capped CRF where you can put a cap and tell CRF, maintain your video quality while not exceeding a particular bitrate by a certain percentage. So this makes it very suitable for ABR streaming. And if somebody asks why to that question, I would say bitrate is in my understanding, is a contract between the player and the server.
So when the players is told that this stream is 5 Mbps, it expects and believes that the server will send five megabits per second. Anything more or less by a large amount will cause problems at the player. So it's almost like a contract. CRF will end up violating that contract capped CRF not much, and it provides very good quality. So for CBR you would set a bitrate and then you would set the maximum bitrate and for CBR, those two would be the same. What do you do in a CRF situation? What's the setting that you use in the command string? I believe it's the max that you set the -bv and then you also mentioned the CRF setting, so it typically has the CRF setting with zero is almost lossless pristine quality, which no human can make out and 51 is really poor. Typically in practice, we have gone between 20 and 25
and this is through just golden eye testing and VMAF scores where we realize certain genres don't require 18,17. We didn't see any difference on a mobile phone. So this is why I go back to the business side of things, where you understand the geography you're streaming at, the devices you're streaming to, and if you realize that 95% is going to be consuming on an Android phone, there just isn't a lot they will make out if you change your CRF setting from 24 to 25. You might save a lot on your CDN card bills and your storage bills, but you're not harming the user experience at all. Yeah, you and I talked about this issue just, I guess, a couple weeks ago I gave a presentation at Mile High Video earlier this week and I sent you some 2 Pass VBR end codes and you said, well, that it looks great, but you're doubling your encoding cost, and then you relayed your experience where you used capped CRF, with a CRF value of 25, I believe, and then the maximum setting. Thank you for that, because it was a really interesting portion of the presentation, but we did the overall VMF scores were about the same. The encoding time was cut in half. There were some variability issues. I sent you that data,
but overall it seemed like a pretty good strategy. We talked about VMAF, or you mentioned that really briefly, are you using the phone version or the phone profile for VMAF? Have you experimented with that or are you using the default? So again, we do this sort of comparisons where we run it on the phone model and then on the desktop model, I know there is a 4K model, or am I wrong on that? There is. One, right? I've never tried that, to be honest. Was never in the situation of test 4K. So the phone model and this, and we actually see a very big difference, you could change your
bitrates quite a bit and the phone model, the scores wouldn't budge. So we realized that hey, here is something that could actually save us a lot on CDN bills and storage costs. And that naturally leads to per device manifests doing something, figuring out who's asking for a manifest and serve the right manifest, doing edge manipulation of the manifest. So I think it naturally tends itself to a lot of innovation, just having those scores. And how do you see that working? I mean, what different profiles are the typical publisher supporting? You've got a mobile profile, is there a smart TV profile at this point, or it is computer playback? I guess, it's not such a big deal for premium content here in the States, but what about in India? I honestly have worked to only a couple of them who have seriously thought about per device manifests. It has been tried, but if I look at... The premium publishers, yes, they are looking into this, they are doing experiments, they are delivering it this way, but then you have a bunch of OTTs which are very sure about where their audience is, and they know that quality is good at certain bitrates, so some of them don't have these considerations also.
So HDR you said, there's not a lot, I mean, unless you're distributing to the living room and unless you're using HEVC, HDR's not a factor. What about DRM? Is that typically used over there? DRM is a big thing. So either people roll out their own using this AES-128 or they go with Multi DRM vendors who provide them with the common Widevine, Fairplay streaming, and Microsoft's Play Ready. There isn't, I would say, major innovation going on in the DRM space, it's just pick up a vendor and go with it. And what about the CDN side, are there multiple CDNs that are covering the different regions, or is there one big CDM that everybody uses? How does that work? So Akamai and CloudFront are pretty popular in India. I've seen several companies do multi CDN switching between these two. You also have AirTEL who's entered the race with a tie-up with Qwilt to role out
their own CDN and Jio's also doing the same. So you have these local, the homegrown vendors, and plus you have Akamai and CloudFront. These are the big guys in the market. Are people doing multiple CDN support with switching and- It is becoming a thing over here, but to be very frank, and I have been looking at data also for the past few years, CDN's have become very reliable and very good at streaming. You only hear of an occasional hiccup and crash. They are pretty good right now.
You might do multi CDN for certain issues probably in a Tier two, Tier three city where certain transmission isn't very good or to save costs, I might have a better rate from A versus B and then I switch them during the prime time. It sounds like overall the market is a challenging market to get into, because of the languages and the regional, but the encoding picture is actually pretty simple. I guess it's not... It's relatively... It's not that hard. Multiple codex are not a thing. HDR isn't a thing, exotic encoding ladders aren't a thing. So it seems straightforward, the content play and getting people to buy your stuff, I guess, is the big thing. Yeah, so it's the business side. Sorry to interrupt. It's actually the business side which is being on all these technical decisions.
So I'm sure a lot of the teams are dying to deploy a HEVC VP9 and try the data stuff, but does it actually pay off? That's the question. What percentage of revenue, if I'm an Indian OTT provider, what percentage of my revenue and percentage of revenue, not number of subscribers, what percentage comes from inside India and outside India? Oh, John, probably I don't have the right answer to that question, but- We were talking and you were mentioning that there's a huge market outside of India for all the people who have left it, and they're paying US prices, they're not paying- They are. Yeah, so that point is true. I don't know the exact split, but definitely Indians are there across the world. I think the biggest hotspots are the US, UK, Canada, Middle East,
and a sizable population in Africa as well, South Africa. So people streaming to these regions, they end up paying in the local currencies. I think the example that I was talking to you the other day was a particular streaming service which charges like $60 Canadian over there and probably $2 or $3 US in India. So you can make a lot of money out there, and it's also profitable to stream outside India, just because of the add CPMs. You can earn several dollars, in double digits probably in the US, whereas in India it's probably $1, $2 CPMs. That's where it caps off at.
And just, I should have asked this back when we started talking about the content, but the subscription rates are low. Is it advertising supported, or is it subscription only? Most of OTTs today do a hybrid, or are going towards hybrid having started off only subscription. So in India it's a thing to have a mobile only plan, where you can play back only on a mobile device. You can't play back on anything else. So that's your lowest price, which might come with ads, might not come with ads. And then you say, okay, here's another tier
where it's on large screen, it's on mobile, but you will see the occasional pre-roll ad, mid-roll ad and if you want to turn it off, go to the premium plan, add free streaming for one year. That's typically what happens over here. I mean, the IPL is an oddity in this entire screen where they're streaming everything for free with an add supported play, but there have been talks that they'll be rolling out a subscription plan in the next couple of months, after the IPL.
Let's switch gears and talk about OTTVerse. I had my site streaming earning center and I've contributed a lot to streaming media over the years. You kind of came out of nowhere and really grabbed a big share and in a big way. What was the idea behind OTTVerse and what has it grown into over the years? So the story is very simple. I was tired of working 12 years on the trot. So when the pandemic hit, I just took a break and
I started writing. I had always blogged and written about obscure stuff, recipes, cricket and stuff like that. So I had a lot of notes lying around. So I cleaned them up, put them online, and it suddenly became popular. People approached me to sponsor articles and put ads on the website and then one thing led to the other end, we decided to make it a business. So that's how it started off. I think our foundation and our goal is always spreading knowledge.
I found it very tough to switch, I was in Harmonic career, so I found it very tough to switch between transporting and OTT. I just couldn't find proper material which would go in depth. So I tried to just put my notes online and help others as well. So what statistics do you share about viewership? How many eyeballs? Geographically we are 55% US and EU put together, 20% India and the rest is spread across Asia and Latham and the UK. That's our geographical spread. It's mostly from the US and the European Union. Then in terms of viewership, we are flirting with a hundred thousand views per month number. Sometimes it's up, down, we'll probably be consistently above that in the next couple of months.
What's the distribution? I mean who's looking? Is it big company, small company, developer, OTT programmer? I mean, who do you see coming to your site? That's interesting actually. So a lot of statistics actually we get out of LinkedIn, because we have a pretty active channel over there. So what we've seen is when it's deeply technical, we have the folks who love tech who come in. So it might be a programmer, it might be an entry level engineer, it could be as high as a CTO, but mostly on the tech side of things. And we have a lot of explainer articles which kind of simplify stuff, like what does client side add insertion. So I've actually met a lot of marketing teams who read that and find it easy to understand and then explain it to others. And then we have also started a few opinion pieces
interviewing OTTs business owners. So that obviously attracts a slightly different crowd, the C-levels and directors. So we have a healthy mix and I think that's how we'd like it to be. What other, you and I talked about some of the educational initiatives that you're producing from OTTVerse, what are you doing there? Inspired by a lot of your workshops, I think, truly there is a need for concentrated workshops on probably using FFmpeg for compression for also pre-processing, post-processing in terms of packaging, actually putting DRM together. A lot of this knowledge is hidden and it doesn't have to be, because FFmpeg is also open source. Why can the knowledge be also open out there? So these are where we want to do a bit of educational initiatives, and specifically targeting colleges in India, universities, because video isn't really a big topic which is taught over here. So they go on the more mathematical concepts,
like image processing, signal processing, they skim over the video, the entire topic of video, despite it being a super complex place, you can probably enter and exit, you can retire working on this OTT pipeline. There's enough work for the next 20, 30 years. Students aren't really aware of that when I interact with them. So that's something that we want to do as well. So what does this translate to? You and I talked about potentially doing workshops in- Correct, so this is probably going to be a couple of days workshops, if it's in person, go down to an office. Oh sorry, that's the other thing, right? We see a lot of engineers coming into the workforce not really understanding what is OTT. Specifically, I would even go around saying that teams when they test, the tests are pretty simple. I press play, if it doesn't play, hey, it has a bug. But can you go a little deeper? Is there a problem in the manifest,
or can you create test cases by deleting certain files, by having them on the manifest. So when the player tries to play it, something happens. Is the player supposed to crash? Is it supposed to fail [inaudible 00:32:14]? We don't know, right? So just the general education, either in the corporate region or the educational region. It could be a couple of hours, or a half a day online course, or they actually go in person and walk them through an entire week, talking to them about end-to-end, right from content upload to playback, recommendation, search, and then get their hands dirty. It's not very difficult if you have like five hours on your hand, you could learn compression, how to use FFmpeg command line, do the HLS, set up a server and actually stream it to video.js. And I think when somebody
presses play and it actually works, that much is enough to encourage them to take the next step. Does light you up when it actually works. What are the topics that your readers are finding interesting today? Looking at your page views, what topics are most compelling? We find a healthy mix between transcoding and in transcoding we see lot of questions still on CBR, CRF, 2 pass. There are very good articles online. I mean I don't say mine is the best. So there are very good explainers. There are good guides from the FFmpeg website itself,
but people still want to understand more. They ask about the fundamentals of compression, like what's an IDR frame? What's a CRA? What's an IPB? People want to understand that. Those articles are pretty popular. How to package that seems to be trending. Then it goes on to DRM and ad insertion, client side, also server side ad insertion, because these are the jargon which are thrown out there in meetings. So people want to understand what's the difference,
how do they work, and stuff like that. Getting some questions in. One question, I guess, more specifics on the publishers who are succeeding in India, wanted to know how is Netflix doing, how is Prime doing, how is Hulu doing? Do you have any information to share about how successful those companies have been? I know you mentioned Prime seem to be doing pretty well. Prime is doing well. They have a lot of original series that they've produced and released.
So Prime's good. Netflix hasn't had that sort of penetration, because they're primarily, I would say, a premium platform, price point is also high. They were forced to reduce their price, because they didn't find any traction in India. So they have a mobile only plan, which is 149 Rupees. Prime is still restricted, I would say, from my guess is to the metro and the English speaking pockets of the country, but it won't have many takers in the large part of India where 80% of the population lives. But Prime is big. Jio has recently tied up with HBO
and Warner Brothers, so all of that can now be accessed on Jio in a couple of months, I suppose. They were early on Disney Hotstar, but that relationship broke and they moved over to Jio. Hollywood content is also available through Lionsgate, Lionsgate Play. I don't know if they have an app of their own, but they are available on aggregators. They are available on Airtel Xstream. A question came through about what's being used in terms of packaging. Is it more HLS or Dash? I guess it's more Dash, right?
It's actually both. Funnily enough, most of the news channels in India also livestream onto their websites and all of these are unprotected HLS. So HLS is big and Dash is also big. What about CMF, is that making any penetration at all? Or just Because of the fact that a single DRM cannot be used, I don't see lot of, a single file format across all DRM. I've spoken to somebody a month ago and they said that it's
still not on the radar to try a single file. Just kind of an off the wall question, what has been the impact of the success of Slumdog Millionaire? I guess that's a few years back, but I mean that... How did that change? Is it the perception of content in India, or what changes did that deliver? I don't know from the Indian context, it was just great that an Indian movie won an Oscar. But a lot of things, for example, the music producer, A.R. Rahman who won I think the best music as well for Slumdog, he composed 11 songs. This is a guy who from 1992 has probably been producing like 50 to 100 songs every year, every song different, across different languages. So when Rahman won, everybody was, high time he won something. So I would say it had a massive impact, because the Bollywood industry has been a thriving industry for 70 odd years. Several superstars who have been popular across the world as well.
How do they do it? I How do you produce a movie, Slumdog Millionaire was one language in the US. How do you do it for dozens of languages in India? What does that look like in the theater? Honestly, it's a voiceover. For most movies, it's where you have multiple dubbing artists. So I've seen two, the most common phenomenon is you shoot a movie in one language and you dub. So
you have dubbing artists for every actor. So it's the right language, the right accent, all of that. Or the movie is completely re-shot. I've seen that as well. So you have a very popular series of a super cop, Dabangg. In one language the storyline is the same, but it's one actor who's popular in that region and then it gets shot in another language, same storyline, but with a different actor. So it's either that or very expensive ways to actually have the same cast,
the same background, everything. Have different people coming shoot their parts and move on. Organizational nightmare. We have a question about NAB. You were at the show. What did you see that was kind of impressive to you? Before NAB I was talking to a few friends and we were kind of guessing what is the next big thing from the India context at least. We realized that
OTT content, the same content is going to be on multiple platforms very soon. It's either Sony on Jio, or SonyLIV on Airtel Xstream on Tata, now it's the same content, same time of the day. You get it on the same day. What makes you decide where to go? Is it price or is it content? Is it quality of the app, or is it the quality of recommendations? So that's when we realized that probably the next thing which people are trying to crack for many years is churn reduction, personalization, and reducing the amount of time you have to spend searching for something.
I think that's a persistent topic in every trade show. How do you reduce the amount of time? I spent nine minutes trying to find a movie. How do you reduce that? So that kind of stood out this year at NAB, at least for me. I saw a couple of companies Think Analytics personalizing the EPG page, which to me struck a call. Because in my house we subscribe to probably 20 channels on our cable, but we still have to scroll through like 600, 700 channels or memorize all the numbers. Now, why can't you just personalize the EPG for me?
I watch those 10, why can't you just pull them up to the top, or create my own channel? Virtual channels are now becoming common, what to live. Have you covered the personalization side? Because that's not something I've looked into. I know it's a thing, but I've not ever written about it. Pretty hard to test, I guess. A couple of articles. We actually have a
contributed article which will come out next week. I won't mention who, but it'll be coming out in the next two weeks. It's by a popular OVB. They're talking about personalization and data collection. Where's the boundary, how much data do you collect to personalize. So it's a very interesting take. One last question. Did you see anything AI related in the codec space that you thought would be impactful in the next two to five years? So I've always loved the use of AI in codecs and it doesn't have to be super complex. I've done tests myself where you can actually use
a genre of codecs and you'll know what are the popular settings which work very well for them. That's itself a use of machine learning. Thousands of parameters which are generated every frame, if you can run them through AI is a buzzword to me. So I typically say machine learning, which is the more technical and the right way, but if you run certain algorithms on it, it makes your next compression easier. So that's where I would love to see everyone go towards. Scene understanding, understanding the different scenes of a movie, which I believe couple of them are working towards. A movie can have multiple scenes where you have people sitting in a coffee shop. Can it dip into a repository of settings to come and compress that particular scene,
versus then somebody's running behind a football and use a different set of settings over there. I would love for some machine to be able to look at every fame and say, this is the ideal bitrate distribution. Go for it. Do it. Interestingly, at Mile High there were a bunch of codec vendors, well not really even codec vendors as much as encoding, and Netflix talked about what they were doing. Sky talked about what they were doing and it feels like there's going to be a lot of machine learning based innovations coming out in the next two to five.
I don't know how it's going to hit something like FFmpeg. It's going to be interesting to see, I mean, you and I specialize in making technology usable to the average Joe, I don't know, Netflix showed some AI based scaling technologies that it's great, but it's not something that you can access from Ffmpeg, unless they open source it. I would actually suggest a different route, perhaps something like the VMAF library, where you don't have to understand what happens inside, because it's pretty complex, but you can actually use it. So if somebody converts a database or let's say a hundred thousand clips, and then you are able to somehow connect FFmpeg to that database, that's good enough. It understands, sends some statistics there, that server sends back saying, here are your optimal bitrates, go for it. That itself is good enough for 99% of the population, I would say. I'll keep an eye hour for that. Listen, we're
out of time and we're out of questions. I really appreciate you spending time with us. I know it's late at night, but it's always great to chat with you and I appreciate your taking the time today. Oh, thank you so much, Jan. This has been a lot of fun.
2023-05-20