How Katana Video is Revolutionizing Podcast Editing: AI-Powered Solutions with Sam Bhattacharyya

Welcome to Podcasting Tech, a podcast that equips busy entrepreneurs

engaged in podcasting with proven and cost effective solutions

for achieving a professional sound and appearance. I'm Matthew

Passe, your host and a 15 year veteran in the podcasting space.

We'll help you cut through the noise and offer guidance on software and hardware

that can elevate the quality of your show. Tune in weekly for

insightful interviews with tech creators, behind the scenes studio tours and

strategies for podcasting Success. Head to podcastingtech.com

to subscribe to this show on YouTube or your favorite podcast platform and

join us on this exciting journey to unlock the full potential of your

podcast. Going to take you down to Houston. Today

we are chatting with Sam Bhattacharya. He is the CEO of

Katana Video. That is Katana, like the

Blade video and currently the platform

is here to auto edit zoom recordings to turn them

into video podcasts. Something I'm sure that as people are hearing this,

are thinking, oh yes please Sam, thank you for

joining us here today. Thanks for inviting me. Before

we talk specifically about how Katana works,

just tell me a little bit like how did you get interested in

streaming media? Like, you know, why are you working on

content platforms? You formerly worked with Streamyard, but like

what drew you to this industry specifically?

Yeah, so let me start most recently and then get started with how I

got into video in the first place. So most recently I used to work

as the head of AI for Streamyard, which is a similar platform to Riverside that

we're using, a little bit more focused on the streaming aspect.

And there we got into kind of

AI and editing features which is how we got into some of

this world, how I got into Streamyard and how

I got into this world of video streaming. I have an interesting story

but I'll try to keep it short. So I originally started my

career, I've never actually worked for a real company or I mean

streaming is real. I didn't kind of like apply for real. Like I have a

bit of a non traditional career path. So right out of grad school

I started a company, my own company and the idea was to

make an E learning app for students in Sub

Saharan Africa, specifically in West Africa. And my

co founder was from Nigeria. He had studied there and then gone to

study in the US for grad school. And we both saw this problem of

the Internet not being very accessible for taking things like online classes. So our

idea was kind of like Khan Academy but for students in West Africa.

Despite my parents hesitations, I moved to Ghana And

Nigeria for a year. With my co founder, we built an app to help students

study for exams, like a Khan Academy. We did the thing.

We built an actual kind of exam preparation app with online

courses. And a big part of what we had done was we had built some

interesting video compression technology to make online

courses very accessible on really, really slow Internet

connection so that you could watch or use like an online video course on

a 2G connection. And in those countries, I

think still to this day, you pay per gigabyte. Right? Right. So thinking about,

you know, like, or being paying per megabyte, like, if it costs you

like $2 to watch an online course just in

bandwidth, like, you're. That's. That's a barrier. Right. Especially for people

in those countries and students, of all people.

So that was a big thing that didn't work out as a business. We had

users, we had actually about 50,000 students studying for our

exams using our app. But it was hard to monetize and get anywhere near

covering our costs. So we eventually shut that down and

made the content free. And then we moved back to the US and tried to

license our technology to other companies.

We actually created this, like, patented video compression technology.

Like, no joke, we got in front of the right

people at YouTube and Netflix. Like, not without

exaggeration. Like, we actually got in the doors with those people and quickly realized

that while we had an interesting idea, it wasn't practical to deploy at scale at

any real kind of video streaming

platforms. A couple more pivots.

We eventually ended up creating AI for

video streaming and video conferencing. So we

had specialized in making features like virtual backgrounds and background noise

removal. And then during the pandemic, we had gotten in touch with

Streamyard who ended up using our technology for their own

platform. And then they ended up acquiring us. And that's kind of how

we got into Streamyard and like video streaming, it was just kind of an

accident. But over the years, I'd built up experience around kind

of video, video processing, especially AI as it

relates to video. As someone who

created a platform and then kind of like realized that the video compression was

the big tool here and, you know, then was able to get in front of

YouTube and Netflix and then go work for Streamyard. Like, do you do a lot

of content creation yourself? Do you work with video or.

Really it's just you figured out this problem and the coding itself

is really where your passion lies. I think I'm much more on the

coding side than the content side, though. We had actually,

I Made some of our first online courses when we did that

app. But we quickly realized that it was better to actually hire real

teachers, especially from Ghana and Nigeria, to make those courses. So part of

it was that was objectively better to have actual teachers

making that content. And then more recently, you know,

I just kind of been stringing along. I had started making

content in when I was in Streamyard as a way to

generate empathy for the people who are making content on Streamyard. Like, so I was

like, you know, I was a product manager at the time, and I started

making content to empathize with people who were using Streamyard to create

content. But I actually kind of fell in. Like, I, I, I,

I, I grew to like it the same way. I actually taught

myself to code. So I figured, like, why can't I teach myself to make content?

It's very different to put your code out on the stage than it is to

put yourself out on the front stage and, you know, get that kind of feedback

that is required. Yeah, well, you know what,

that's true. And I am incorrigibly technical,

if that makes sense. But one of the things that I realized that makes me

kind of a bit different from most normal engineers is because I did a startup,

because I did a lot of this stuff, that you're inherently putting yourself out there

anyway. And so I had been no stranger to making pitches, to trying

to convince people, you know, it was a different kind of use case. I was

trying to convince people, hey, funder platform to help students and, you know,

Ghana and E. Sorry for the exams. But at some point I was still making

pitches, going on stages, trying to convince people, like, what we're doing is

interesting. And that didn't seem that different from

putting yourself out there online. It's just, you know, you have

to, maybe when you're building a show, for example,

treat it a bit more like an actual product. Like, who is your audience? Why

are they interested? It's not like, so transactional as, like, you know, you

donate to my company or invest in my company or whatever. But

there's a sense in which I had gotten used to

talking to people about what I was working on and trying to convince people to

be excited about what I'm working on. That already kind of came into the door

when I started making content. Gotcha. Okay, so

tell us more about the actual Katana video platform.

So how does it work? How do we sign up? You know, what can we

expect? What problem is it solving for us? Yeah,

so let's start with the problem that it's solving.

So when I started working with

Streamyard, I learned that many people make podcasts, obviously.

But then the number one alternative that people used

for recording podcasts on Streamyard

was not Riverside, it was Zoom. Okay? So most

people who start with podcasting start out recording

on Zoom. And there's a reason that platforms like Riverside and

Streamyard exist, because Zoom is not built for recording

podcasts. And one of the main things that

makes it hard to work with Zoom and turn that into a video podcast is

that Riverside and

similar platforms like Streamyard will give individual

recordings, high quality recordings for each person

doing the interview. And that's super important with editing to be able

to do, for example, camera angle switching and knowing who's speaking

when. And editors

generally dislike working with zoom recordings because zoom

doesn't give you that. Zoom just puts everything together. It's a nightmare to work with

those zoom recordings and people work around it.

So people will put like overlays on top of a zoom recording. But I

had a pretty deep background in computer vision

and kind of like old school AI

and I figured, well, I mean, there's no reason you couldn't actually figure

out like who's speaking when it just takes some upfront effort and work.

And so the simple idea was, okay, well, let's

actually take a zoom recording and then

extract the video and audio and separate them as if they were

local recordings. Okay? Then you could do things like multicam and

camera angle switching and adding like name tags and all the stuff you would normally

do. All the visuals you would normally put in a normal video

podcast. It would be much easier once you knew who was speaking

when, right? So that was very simple idea. Like, what if we just didn't

fight this idea of people are going to use Zoom, get them to use Riverside

or Streamyard or some other platform like that. What if we just met them where

they were? You're using Zoom. Okay? Whatever reason you have to resume, there's still valid

reasons for using Zoom. And so let's make

it easy to make a zoom recording look like it was recorded and edited in

a more professional platform like Riverside. And that was the high level.

So when you go to Katana Video,

the goal is to make it easy to turn

your zoom recording into a video podcast. And a big aspect of that is

automating those visuals of like the multi camera angle switching and whatnot

so that within like five minutes you have something that looks like a

professionally edited podcast with a lot of the

visuals that would normally be done with

tools. Like Riverside or Descript or whatnot. And then

I'm also trying to make sure you can have a lot of the actual edits

and cuts done. The basic ones, not like very, very artistic, but the basic

ones, like making sure you cut off the recording before the interview actually

starts and cut it off like after it actually ends. Because, you know,

there's, there's one of you make a recording. There's like the, the pre

interview stuff and the end of the interview stuff as well as like detecting the

obvious like outtakes that you would find. Like, you know, can you cut this part

out so that someone who's just getting started with podcasting can just get something

that's out of the box, like is better

than, better than what they started with for very little

effort. It's not to the level of a professionally edited podcast by any

means, but it's certainly better than what you started with.

And the idea was to get something

passable in five minutes. And so that's kind of like you just upload

a zoom recording and you get like a

good looking podcast in five minutes or 10 minutes. So what does it

do? Like, what is the AI doing? What is it looking for? Are

there things that we should be doing when we're recording

to make the AI's job easier?

Yeah, so. Well, one, everything's a work in progress, so I will be improving these

algorithms as we go. But one is in terms

of the core, who's talking when.

The only thing that seems to mess up is when two people are talking exactly

at the same time. And that kind of, I mean, can you really

blame, you know, even as an editor you would have a hard time.

So in that, in that, in those circumstances we just default to showing both people

at the same time. Just like you don't highlight one of them when two people

are talking at the same time. That's it.

I think the idea is very much like you don't have to do anything specific

to make the jobs easier. Like there's just like some edge cases that

I'm finding that I need to handle

for. So someone had an intro section that

they recorded at the end of their podcast. We're like, oh, we forgot to do

the intro, let's do it at the end and then we'll

fix it in editing and post production. And that's like such a normal

natural thing. But that kind of messed up my very simple algorithm that like assumes

that the start happens before the end, if that makes sense. And that just

kind of messed that, that whole thing up. And so, I mean those are edge

cases that you'd want to handle gracefully in the future. But

the idea is like you don't have to do anything special. Okay. And then

once it's done, is it

take it or leave it or can we take what you're

creating and then, you know, bring it over to one of our editors and you

know, make some finesse edits or you know, maybe tweak a few things here or

there to get it where we want it to be. Well, so it's,

it's got a built in Transcriptus editor

built in. I, I consciously didn't make

this a kind of. You can export this to Adobe and maybe I will in

the future. But the use case that I was looking for

was targeting a different segment. So there are plenty of

very nice editor softwares out there like Descript and you

know, Riverside, you can edit. But this was definitely designed for the

people who wouldn't otherwise have

or use those editor softwares or hire someone that does have those editor software.

So I made sure, I spent a lot of effort just making sure that it

works by itself. So I have like this own style of stack. So it's

basically like you can tweak it, especially on the adjustments, like the branding, the look,

the feel, the custom and you can, you can edit it based on transcript based

editing and then it renders, you know, it

has a full rendering stack and whatnot. But I haven't quite put in like you

can export like the RAW project file as an Adobe Premiere

profile or something like that. Maybe I will in the future, but I

don't feel like I have enough like of the auto edit stuff yet that it

makes sense to kind of do that. And I do want to improve the

auto edit capabilities in the future. All right,

so how does someone get started, right, like what are the pricing plans? Like what

does it look like to work with it? Do we have to upload? Is there

an integration with Zoom built into it? What does it look like for

someone who's hearing this and wants to try it out? Yeah, well,

so first it's Katana Video, that's the address

and it's free right now. You couldn't pay me if you wanted to because I

am still in beta and figuring things out. And the

idea behind it was to have a free option that's always

available. And the high level idea behind the free

option was you can make your zoom recording look good

and it'll have all of the camera angle switching and all of those like branding,

customization, Options just out of the box for free forever. And

there will be a play plan which has some additional auto edit capabilities.

So in terms of auto edits, I think the idea is like

to give you everything you need to get a

raw recording to something you would happily upload on YouTube. Like

there's a couple more things you'd need to do and one of them is like

generating a really nice catchy intro, for example. That's one of the big things

I'm focusing on right now. And so if you look at professionally edited

podcasts on a video show on YouTube, they'll often

have an intro section which is like a catchy

back to back compilation of sound bites. Sometimes with effects.

Like they'll zoom in on one of the speaker's faces, maybe they'll

highlight some words in the background. At the most extreme end

you'd see stuff like Diary of a CEO that's a bit extreme for what an

automated tool could do at this point, but you have like less

extreme versions of that where it's like, it's like a catching intro. So that's the

kind of thing that would be on the paid plan. And so you would kind

of generate one of these like intro teasers as part of the paid plan as

well as like social media clips. Like the clips. So there's that

functionality that would be on the paid plan primarily, but then the core

just making your zoom recording look good. That's free

forever. Forever. And yeah,

I mean it's, it's also. Well, as long as Katana Video

exists or whatever, like, you know, how am I supposed to know what, what

things are going to look like in 20 years? But where do you think AI

is going in terms of this stuff? Like, I love the idea that there

are aspects of content

creation and content editing that are

monotonous and repeatable and you know,

don't really require a lot of feel

to get them right. Right. Like switching between two speakers. It's a

fairly simplistic concept. But are you,

do you think AI will ever really replace human

editing and editors or ever be, you know, take it to the

level that it will be capable of making something that

is artistic or, you know, has emotion

to it, or is it really just, you know,

factual content editing and

sharing and you know, quality control more than,

you know, character control, let's say? Yeah, so I

have, I do have some opinions and that may be interesting to the audience.

So one, I think there's a lot of misunderstanding of like how AI

works and also even the people who, people who actually, like, do have an idea

of, like what? Like, you know, there's no one's clear on what the future is

going to be like. But I have a thesis of

how AI is going to impact video editing. I think one of the

things that you have to understand going off the boat is that there aren't really

AI models that are trained to edit video. And

I want to make sure and emphasize that point. There aren't really AI

models that are in any deep or

fundamental way trained to predict what edits you would make in

video content. And that comes from the fact that these

large language model labs like OpenAI and

Google haven't actually sat down and hired hundreds of video

editors to build the data sets. It goes back to

the fundamental, like, they weren't built for this stuff. And that doesn't stop people

like opus from using ChatGPT as a way of

editing. But that's why the results are so mixed. If you

use a tool like OPUS Clips or even like,

you know, Riverside's clips, like, I don't think anyone would mistake those results

for something that was created by a, you know, like a trained human editor.

Like if, if you got like 30%, if, if, like a human editor gave

you a clip that started in the middle of a sentence, you would say something's

wrong with you. Right? But there's so many obviously wrong things,

you know, with some of these clips. And I think that's why. So I think

that people will start to address it. That's what I'm doing. But I really see

as people kind of figure out, like, how to kind of actually

get AI, not just to understand what's going on in a video, but also to

decide what edits to make. I think you're going to see two

distinct directions, and that's coming from my experience with software.

So in software, you have tools that are being used to speed up

software tasks so people who are programmers can

now code faster because of these coding tools. And I think you're going to see

the same thing with AI, sorry, editing tools. So those editing tools that blaze

essentially have the equivalent of autocomplete or I think descript is probably

the best example that I've seen of this so far, where they have, like, smart

transitions and smart kind of layouts that'll predict

what it is you're looking for and just kind of speed up that process. I

think that's probably where most the most useful innovations

are going to be in terms of AI editing. And in that sense, of all

the companies that I've seen doing anything in this editing and creation space like

Descript is probably way ahead of other companies on doing that.

Like, I actually don't think that Opus is particularly interesting in

that respect. And then what I'm trying to do, which is

not build a tool for an editor, but rather kind of build something that's

similar to Squarespace, where like someone who is not an editor

and doesn't have the budget to hire an editor can still get something that's okay

very quickly. And so in the sense that Squarespace

lets you get a website without necessarily hiring a programmer or

learn to program yourself, the idea was can you build

AI that can get you something that is

maybe not as good as what you get from a professional editor, but passable.

Now, regarding the question of will you ever get something that'll

reach the creative levels of an editor,

I want to appeal to this meta sense of what is possible with AI.

So just the high level benchmark is if you

gave the same thing to 10 humans to do and

they would all give you 10 different answers, then that's not a good kind

of task to automate. And so if you're talking

about like really fancy edits and you gave the same editing,

you know, the same kind of like mandate to 10 different like high level

editors, and if you got like very different

responses back, that's probably not something you can edit. And that's why,

you know, I struggle to see how you could create like edits of the level

of like a Super bowl ad or like a Hollywood movie that's ever just

generically edited by an AI. But I

think the argument here is that a lot of more

mundane kinds of content that people are making are not Hollywood edits. And

the edits that you're making aren't like that creative. And so

there's a lot of this kind of like mid level content for which

there's an obvious like answer of like where does the recording start, where does it

end? And those are tasks that are very, very much automatable.

Because if it's like 10 people would all look at the same thing and say,

yeah, the right thing to do is start here, start there, do this, do that,

then you could imagine automating that.

And the goal with what I was looking for is finding this subset

of editing tasks that fit those cred that category of

kind of things. So it's like, you know,

I would never imagine an AI like just coming up with like a really great

super bowl ad, but most people aren't creating super

bowl ads, if that makes sense. That is very, very true.

So, all right. We are chatting with Sam Bhattacharya. He

is the CEO of Katana Video. You can learn more at

Katana Video. We've also got a LinkedIn

connection for Sam, so if you want to learn more about him and some of

the other places he's worked and the things that he's up to, you can follow

him there. Sam, before we let you go, we always like to ask folks a

few questions about the space in general. Now, our show usually

focuses more on podcasters. You're more of the content

space where this isn't just limited to podcasters. But I'm still curious.

Is there something else in the podcasting space where you would like

to see improvement made or have somebody

work on solving problems there?

I don't know. I'm fundamentally, instead of prescriptive, I'm

very descriptive in the sense that I don't like, imagine, like, this

is how things should be done for how people are making content. I just accept

that people are making content as how do you fix problems that

exist? I see a lot of debate on, like,

audio versus video, and at some point I kind

of get that there's this mix and merge of media, and

I see that there's a lot of people with opinions, and maybe this is just

me coming with, you know, very little experience in the, in this

industry up to date. So just let people make content like, you know,

people. If there's like this mix between, like, show and podcast, like, I

don't, I don't have strong opinions. Just, like, let people do what they want to

do. You know, call it a podcast if you want. Don't call it a podcast

if you don't want. Use the platforms you want to. I just, you know,

I, I see opinions from people who are more

experienced than I am in this space. And like, I, you know, it almost like

I empathize in that I have my own crotchety opinions in the space of, like,

programming and whatnot. But just, I don't, I don't get why people get so

fussed about, like, you know, the direction of content giving. It's all kind of

like mixing in this grab bag of, like, what does content even mean at this

point? Yeah, if it's useful, if it's valuable, if somebody else enjoys it.

Who cares what you call it? Just put it out there and let people access

it. What about, is there any tech on

your wish list, whether it's for content creation or for the,

you know, editing process, something that either is out there that you want to get

your hands on or something that has yet to be made that would be useful

for you. Well, I mean, I, I'm kind of building the

thing that I want to work, right? Like, so I think

the, the. The thing that frustrates me most about tools

like Opus is that it, it doesn't

have, like, built in quality control. Right? Like, you'll give it

a video and it'll give you back 30 clips, and you have

to still go through and curate which ones are obviously

good and obviously bad. And I kind of wish that you could have some kind

of quality control where at this point, AI is smart enough, it should be

smart enough. And that's what I'm working on to make sure that,

okay, not Every show has 30 clips that are worth surfacing. So surface the

ones that are actually with surfacing, even if it's not like 30, right. Like, if

it's. If I only have like eight moments that are worth sharing, give me those

eight moments, but make me make sure that those eight moments are actually like, you

know, good or at least passable. Right? Like, I think we

haven't even gotten past this basic filter of like, you know, like, there's

like, artistic creativity. We can all disagree on, like, what constitutes good, but

there's still a lot of things where people, the results are just obviously bad. And

it's like, let's focus on filtering those out first,

then we can have an argument on what's good. All right? And

then lastly, are there any podcasts or. I'm going to expand

this. Are there other content creators that you are following

religiously that you want to talk about?

I have come to realize that I have a very different information

diet from a lot of people. I was at a podcasting conference called

podfests earlier this year, and they mentioned four different shows and

podcasts. And two of them, everyone raised their hand except me. And

then the third, I was the only who raised my hand. Okay.

Just some random stuff that I like. So one is,

I follow a lot of AI stuff. So there's one podcaster called Dwarkesh

Patel. He's, you know, some very smart CS

guy that decided, I'm going to go into podcasting and interviews, like the CEO of

Microsoft, and they're talking about the future of AI. And you listen to that

stuff and it's just a very, very different kind of view of the world of

like, assuming that the whole world is going to be automated. They're talking about, like,

you. Are we going to have AI only companies, it's

like. And then like the actual conversations from like real world people

is very different. Or, you know, AI is just like a tool. Like,

AI just means chat, GPT. I don't know, it's just very different information diets.

And I'm sitting in the middle, I'm like, I don't know, like, just people have

different information diets. That, that kind of feeds into their

worldview, I guess, but just trying to make sense of that. But,

you know, those are some of the things I like. I also just have like

a hodgepodge of like, I like history. So I have some very random, like, history

podcasts that I listen to, but it's all very nerdy, if that

makes sense. That's okay. That's. I mean, I think that's part of what makes podcasts

great, is that it allows people to really get as nerdy as they want to

on a topic that interests them. And, you know, they're not just forced to consume

what is available. So the nerdier the better.

Yeah, exactly. You know, it's like some retired professor

who has some time and has decided, you know, I'm going to do a podcast

instead of doing lectures. Like, that's great and it's free and I love it.

So thank you for making those podcasts. Even if it has, like,

even if I'm one of the only, like 3K subscribers they have. I mean, 3K

is not a, you know, number to sneeze at, but still, it's not like when

I'm talking About like Darius E.O. kind of

popularity, but there's, there's people who listen to those things and I'm one of those

people. So. I'm sure

the creators are happy to hear that. And we'll try to put links to all

the ones that you did mention here in the show notes for anybody else who

wants to check them out. Sam Bhattacharya, the CEO

of Katana Video. Thank you for joining us.

Thank you for inviting me. Thanks for joining us. Today

on Podcasting Tech, there are links to all the hardware and

software that help power our guest customers. Content and podcasting

tech available in the show notes and on our website at

podcastingtech. Com. You can also subscribe to the show on your

favorite platform, connect with us on social media, and even leave a rating and review

while you're there. Thanks and we'll see you next time on

Podcasting Tech.