The beauty is the decentralized nature. I don't need a sponsor to go create a show unlike I would need to for TV versus a podcast. Anybody, you have a passion, you want to talk about it, post the files on the internet. That's also the problem with podcasting is the decentralized nature. As a listener, like if you want to know, should I trust this show? Where would you go to to find that information? Like most things, the biggest superpowers are also often the biggest weaknesses.
And in podcasting, that's very true. I love it. Arnab, you just made our little intro snippet right there, Alison. He said, on both sides, it's like it's a gift and a curse. It's a responsibility and it's a superpower. I know. I was thinking anyone can do it. Anyone can do it. Exactly. Anyone can do it.
Why do we as podcasters willingly subject ourselves to hours of editing, awkward ad reads, and the existential dread of shouting into the void? If you've ever asked yourself that, or if you're just here to revel in the weird, wonderful world of podcast... Welcome to Sound Strategy with Lucas Dickey. Without a doubt, this is the most meta podcast about podcasts ever.
Hello there. I'm your host, Lucas Dickey, founder of DeepCast. I'm your podcast producer and occasional co-host, Allison Melody. And today's guests are co-founders of Metacast. It's Arnab Deca and Ilya. Ilya, forgive me if I can't get your last name right. Let's see if I can do it. Bezdilev. Yeah, you got it right the second time. So the Metacast co-founders are with us today. Welcome to Soundstrap.
oh nice to be here yeah thanks for having us you're clearly both mega podcast And Medicast was an idea. So what was Arnab since I'm assuming you went to Ilya and said, or maybe you guys were debating multiple ideas, but how did Medicast come to be as the idea you decided to invest your bootstrapped time and effort into?
there was no venture capitalist throwing a bunch of money at you so you this is a passion and like it's a meaningful part of your life so i'm curious how you decided you know what you know what medicast was going to be why do you Pretty quickly, we both converged on this that first of all, we both love podcasts. Second, we don't like the apps we are using right now to listen to podcasts.
And third, there is a pretty big marketplace for this, which is evident because there are already some established third-party podcasting apps that people really are passionate about. so we're like okay this is something i think we could build we could do it better than anybody even though it might take us like years and years to do it
But let's try it out. You noted the other apps did not do these things well. So I'm not going to, you know, you don't have to name names unless you prefer to. But like, what was it that they were not doing sort of generally speaking? whether it's the big dogs or it was other sort of you referred to them as third party but I think independents that are not Spotify, Apple, and YouTube, right?
I'm curious as to know, what was the job to be done that they were failing at? Or what was the sort of, you're an ideal customer persona for Metacast? What was it that was not being done for you as a customer profile that you were looking to create? So this was year 2022, right? So now there has been a lot more development. in the podcasting space when Spotify started to add more things to the app.
Apple added transcripts, YouTube shut down Google Podcasts and started to invest in sort of video podcasting with YouTube Music. But in 2022, all of the apps, like Apple Podcasts, looked no different from what it looked like. 10 years ago 10 years before right it was still in a sort of a glorified audio player that has access to a catalog of mp3s right
And then same thing with other apps, like Overcast and Podcast, Podcast. They had more functionality, but still overall, they were still like audio players. we were looking for right what our problem was that we wanted to solve the job to be done right how do we actually remember information i heard Because as you heard, we listen to information rich podcast.
with lots of insights. Like if you listen to Alex Hormozy on Diary of a CEO, it's like a two and a half hours podcast. And every five minutes, there's something you want to remember later on. But audio is linear. so you listen to listen to audio like especially our audio like you must listen to it when you're doing something else so you can't just like stop and take notes
And there's no way to go back to those moments that you wanted to remember, right? So I was taking screenshots of my screen just to see the timestamps, but I ended up never going back to those times. i even wrote the telegram bot once uh where i would listen to a podcast and i would like quickly type a note and then send it to to the bot and it would store it for me i had all these workarounds to try to retain information I've heard and go back.
And so when I and I were talking, you know, in Seattle, we were like, wait, there must be a better way of doing this. Right. And then we realized, okay, so if you overlay the transcript without. and then we make the audio and text interchangeable as in like they're always in sync one always mimics the other you can use the text almost like a roadmap for your podcast where you can take all of those notes, you know, do the bookmark.
copy paste things like take screenshots share those share deep links So basically the transcript, and I guess we should say that Metacast is a podcast app whose main feature is the... transcript transcript with the audio when if you if you open the metacast and you start playing an episode there will always be i would say 85 of the screen will be taken by the
So we wanted this experience to be like seamless switching between text and audio. So you can, like the moment you heard something interesting, you just pull up your phone, swipe and you get that bookmark. And we also add more things like doing it in CarPlay and through headphones and all that stuff, right? But the key premise for the app is that we want to give people access to the information that they've heard and would have otherwise forgotten.
such a convenient app as Metacast to take bookmarks and stuff. My canonical example, I think exactly what you said, Ilya, is when we were talking, I remember this moment. What I told you is I had listened to Neil deGrasse Tyson. very eloquently describe the Gregorian calendar and why we landed on this system of like leap years and stuff like that in a podcast. And for the life of me, I could not find, I wanted to share that with you. I could not find out where.
Okay, so that's one of the things that we want to basically make it very easy to find that sort of info. so we are still not there exactly there yet like you couldn't go to metacast today and just type in neil degrasse tyson gregorian calendar it won't find that snippet immediately We will get there. We are building up towards that. But what you can do right now in Metacast is if you know the podcast. then you could search for like calendars inside the podcast it'll pull up that
You play that episode. Inside the episode, you can actually search for things that have been said, right? So unlike just audio, and I think you do a lot of this in Deepcast, so you're like familiar with that. Right. So, but we're trying to find that information retrieval because like Ilya said, we listen to a lot of information heavy podcasts and we want to like kind of come back to those moments or share it with.
The other big difference that we saw at that time, and that's still true even like three years down the line. Podcast started as broadcasting, but just on an iPod. And that's basically like news kind of shows, hourly shows or weekly shows. that are essentially recorded and stored on the internet, unlike on a radio, right?
so you can listen to whenever you want but i think the focus of most of the platforms and apps till today are still within these are the new episodes listen to these and then the old back catalog is like god within a week or two and you can never find those really good episodes and go back go back i have a perfect story actually about this right like a perfect example that we were discussing uh so there's a person called josh wadeskin who is the chess prodigy like martial arts champion etc
He was on Tim Ferriss' podcast. It was the second episode. josh was on it so it was like 2014 2013 i don't know like many many years ago and josh is not a public person so he had not appeared on any other podcasts ever until just recently he went on rogan and modern wisdom maybe he has a new book i don't know but he like there was nothing about josh wade king for like 10 years except for his books on amazon
And well, good luck finding that episode in other apps. Like they would never recommend you the episode from 2013. I think actually many apps still don't have the episode search within the podcast. And good luck scrolling 2000 episodes.
to find that one. So yeah, and that was one of the things that we wanted to, we want to eventually address. It's like, how do we also recommend people the shows based on their interests, regardless of when they were like news sports i agree they're relevant at the time but an interview with the chess prodigy from 10 years ago still relevant today But no other app actually prioritizes the old content. They never really looked at back catalog.
it almost like expires right but there are 100 million episodes right now according to podcast index and there are probably at least a few thousand there that are worth uh listening going back into yeah yeah it almost reminds me of how we have like
streaming television today you know my daughter turns 12 soon and she's watching like a gossip girl from whenever that was and she got into friends and how i met your mother and like these shows are they're not in they're not running anymore right so the back catalog has
staying power and that format enables it whereas ilia you're getting at that like this format doesn't do a good job of making rich info non-highly contemporaneous stuff still has a lot of value and so how do we make that available to users to continue to like enrich their by the way slight off topic um those are the three exact shows that my daughter is watching right now that was uncanny like are you talking about me are you talking about me because i just watched There you go.
If I listen reflectively back to you guys, a lot of it was knowledge retention and you wanted to be able to digest it at some later. And then there were elements of wanting to disseminate what you learned to others. And then there was the discovery angle R&B you were getting at earlier with the, like you can search transcript at an episode level or all episodes within a podcast, but not necessarily globally yet across the entire catalog of all podcasts.
Not yet. Yeah, it's a lot of data. You were your own ideal customer. Has your sort of customer feedback and user engagement validated or maybe pushed you in different directions that you didn't expect? Like, are people doing more of X than Y? And you were like, I don't even think they were going to be doing this thing. man this is a great feature alice and i haven't decided if we can curse on this show yet but this is a great freaking show you know feature
And they're not discovering it and they're not using it in the way. So I'm curious if there are some subset of what you're doing that you were surprised by. We can call it an aha moment or maybe multiple aha.
but where users are behaving either in a way you didn't expect or maybe you got the exact validation you're looking for. They clearly want to do this thing because 85% of our users are doing this thing. Yeah, I can take a stab at that. So I think we had three sort of groups of... feedback of three groups of customers.
They want exactly what we want. And they just keep sending us like... huge emails with like how about this how about that some ideas some even some technical ideas maybe because of who we are we attracted a bunch of software engineers as our early early adopters people we don't know but they're software engineers So they do have opinions. And I think we even borrowed some of those ideas from some of the emails. So those are sort of people like us.
So the second category is people who have challenges with other podcast apps. And I guess I don't want to name the names, but there is one app that's very loved by users, but it's been abandoned by its creator, according to the Reddit. So there is a lot of frustration going on. And so we get a lot of people coming to our app from that subreddit.
And they're like, oh, I'm the user of that app. Here's an opml file. And actually, that's how we know about them, right? We look at the opml file, opml file. And like, oh, but can you also have X, Y, and Z? That X, Y, and Z is actually features from that other app that they're used to.
so we sort of we have been catching up on some of those features too so that uh we can provide the replacement for those users what were those features like you're not naming the other company is fine but like were there particular features that they were getting yeah like playlists okay uh like custom playlists uh some sort of audio manipulation
in real time. It's like sound boosting or audio or clipping out the silence and things like that. Sound boosting. Yeah. Yeah. Clipping silences. Yeah. So you probably know which app I'm talking about. right uh but what also these people say is that they actually love the simplicity of our app compared to the other Like if they come from other apps, they are very feature rich, but they actually want something simpler. And then the third group, which was the most surprising to us.
So at some point there was one user. who listened to more episodes, according to our metrics, then both Arnab and I combined the thing. We were like, wait, what's going on? Like, we are the power users of our app. Somebody uses our app more than we do, right? And not at some point, till date, that person is the most...
Like basically biggest user of our app. Yeah. So, so we reached out to her and we were like, uh, can we talk? Right. And yeah. And then we jump on the call and yeah, the person moved to our app after the Google podcast. and she's like i just need a simple replacement of what they had and you guys seem to be like very simple and intuitive and it just has what i need and i think she started using that like over a year ago so it was still very rudimentary she never uses the transcript
Like she doesn't care. So, but at the same time, she gets a lot of value from the app just because it's simple. And that's, so I think talking to that user and also hearing feedback from others about the simplicity.
think about features in a way that like how do we add features without creating sort of confusing ux like like like some of the other apps created yeah like like one of the problems that i had many users have with uh spotify and youtube music it's like if you put everything in the same app like it's impossible to find anything and you just get confused like if you search for i don't know iron maiden like
Actually, I want to listen to an interview with somebody from the band, not the music or not watch like their concert on YouTube. But if it's all mixed together, you just start to get like you have to do more work to find the things that you need. And so we. Our goal is to build a specialized audio podcast app that helps you get the knowledge from podcasts. But also, if you just like podcasting, we also have a great app, even if you don't care about knowledge.
What happened is our very first release, we focused pretty much on the transcript and bookmarking and sharing and searching. Those were the four things we had in the app. Everything else was like the most basic thing you would look. But on that, the player itself, it was turned out. Nobody else.
We had this in beta and we're testing with maybe, I don't know, 20, 30 people at that time. A few months after we've built it, right? And then Apple makes an announcement that they're bringing transcripts into the podcast. And at this point, Apple had not touched their app for over maybe a decade. right so it was a big surprise for us and it was like okay we gotta do something about this because this is what we've been building for the last six months and maybe apple will just come and take over
And that happens all the time with startups, right? So that was one of the key things that made us realize that it can't be just about the transcript. We need all these other things that people expect from the app too. Yeah. we also started realizing through our customers that it's not like people don't like the UX of Spotify or Apple or all these other apps. There are like quirks in them. And there is a way to like build a much simpler podcasting app.
right so that that's the other thing we kind of had a this is about a year back i feel like ilia maybe last year march or feb or something we decided okay we we need to build like a more complete app and we started progressing towards that we launched i want to say about six months ago publicly launched uh till then it was beta so you you asked that question lucas like did are are those early things that we were going towards are those the right
uh are we seeing validation on those yeah and on that i would marry that with what ilia said the users who are most similar to ilia and me who seek that knowledge dance podcast those are the ones who use the bookmarking and search and all that. But a lot of the users don't care about that. What they want is a transcript. We recently launched this automatic chapters feature. So based on that transcript, now we automatically break the episode down into segments.
And then we give like automatic titles and all that. And people seem to be liking the ease with which they can navigate through the podcast. People are asking for more and more features in that kind of area. Can they deep link based upon your automated chapters? It definitely makes it helpful. It's a good question. You can deep link to the chapter specifically, but you can deep link to any sort of transcripts.
segment so so you can you can deep link to any any point in time so if you basically share from the beginning of the chapter yeah that will be the beginning of the chapter but i think actually it's a good feature I was just thinking about, yeah, Ilya, the fact that Spotify expects you to have...
in the metadata of the episode description, hour, hour, minute, minute, second, second. And then they do the deep linking based upon that in the metadata. And then Apple uses the ID3v2 tag where they expect you to put it within the chapter there. Notice that it's two completely different ways that they implement that and they have an expectation for RSS hosts to know these things and implement these things. And very few podcasts actually do either of those two things. That's very true.
And so from the podcaster's perspective, as podcast listeners, which we all are, but how does this benefit us who are podcast hosts? Yeah, I think another way to think about it is like, are podcasters a first class citizen as you guys think about developing it? Or are they tertiary or secondary to the consumer or the listener? I had a podcast before we started.
but it was in a different language. Right. And I had so much sort of that parasocial connection with podcasters. Our stand has always been that we have to play nice. Like for example, we will not, I think we will, we will never remove the ads from. even though it might be good for the listener, but it sort of goes against that other constituency that we serve, which is the podcasters. Like we don't serve them directly, as in like we don't have any B2B product for podcast.
right but we provide the almost like you could say like marketplace right like the platform for them to reach users who use our app and uh and we want to be fair to because they need to make money, etc. We have some ideas about... you know, features for podcasters to benefit from maybe get more data about listenership, like where people skip, what kind of parts of the transcript they highlight, these kind of things, like which deep links they share. YouTube has a lot.
Actually, YouTube has very good analytics for this kind of thing. But we don't have the critical mass yet for this to be valuable to podcasters. So even for podcasts like yours, you will probably have to be like hundreds of listeners of your show for this to make any meaningful sense. So that's later on the roadmap, but we have to reach that scale with consumers first. Well, I would think like as a podcast. I say, hey, if you want to share this portion of the show, go listen on Medicare.
telling my listeners, Hey, this is a great way that you can, you know, that problem when you're always trying to share that one moment and you can't share it and you're screenshotting and all that, you know, that could be something called. Actually, that's a good idea. Yeah, Allison's giving you guys all the good nuggets. You can take it away and market it. Totally. Yeah.
We haven't actually gone out to podcasters yet. We have gone out to a few who are more like they've been on our show, so they know us. But I think what we're building towards is that critical mass that Ilya is talking about. That's when we go to podcasters and say, like, look.
These many people are listening to your show on Metacast every week. And these are the things that are listening to. These are the interesting bits. And then from there, introduce the whole idea to them. One person or actually a couple of podcasters have come to us. and said that one thing that they have a challenge with is going through their own back catalog of all the things that they have talked about and finding those specific things. And Metacast is like perfect for that use.
So we don't right now have any B2B sort of or podcaster product. But eventually, if we are able to successfully build like a critical mass, we will get. Yeah, it makes total sense. And I think one of the things that we were thinking about in advance of this conversation.
your transcript-centric app, as you noted, and then feature sort of expansion from there to meet some basic needs around preferences for audio and things of that nature. How do you think about transcript adoption sort of across the industry?
and what that means for podcasters and how it helps them or maybe how the current implementation maybe within some of the big apps don't help them that much i have some opinions on that but i'm going to let you guys answer it first So our thesis is that transcript...
are super important they like an overlay on top of the audio like audio is just linear it's just it's binary you know it's bytes right Whereas Transcript is something you can search, you can process with LLMs, you can extract insights, you can build Q&A bots, you can do all sorts of stuff on it. But actually, I just read in the pod news, I think it was last week, that Apple has transcribed the entire back catalog of podcasts. So our thesis is that transcripts will become a commodity.
We hope that Apple will eventually just give it out to everybody and every podcast will just get it for free. You know, if not folks like us and like you, we will just transcribe the entire back catalog. But I would hope that there will be at some point a collection that can be used by anybody basically free of charge and can be built into the app.
Yeah. And at the same time, I think we're really thankful to Apple for starting on this path like about a year back, although it was really bad timing for us personally. uh them they have started like this movement towards everybody trying to get transcripts which is great for the industry and podcasting in general Having said that, none of the apps actually do a good job of using the transcripts in there.
For sure. Right. It's not like music, like in Spotify or YouTube, when you're listening to a song and the lyrics are like floating by. That's sort of how most of the apps are thinking about. scripts right now and i feel like it's totally useless even with folks at those big companies. And I said, hey, it's all well and good that you jam a non-diarized, no speaker label version of the transcript within an innovative application that I can't get to anywhere.
How am I supposed to use this thing? Or the read-along version within Spotify, which is great if you're kind of staring at the screen. To your point, most of us are listening, not staring. so getting back to the relevant sections are very challenging if you want to use it for knowledge retention or sharing and again another native application and with their web application they do a lot
to try to obfuscate or prevent Google and others from accessing some of that information. So it makes it very challenging, I think, for the discoverability nature of transcripts in an open web.
world and you guys have native applications so i'm curious to how you think about this right if if part of the value of transcripts is both within the native applications themselves but it's also me going to google and saying um tim ferris uh i forget what the example we had here earlier yeah exactly and hoping that i can find the thing was said
But if it's all trapped within native applications, or even if it weren't just in with the native applications, you wouldn't know who said what, because it's not diarized. It doesn't have speaker labels attached to it. So I'm curious because you guys are. in the guts of transcripts how you guys it's easy enough to go and allison forgive me for technical language but like you can go to whisper v3 turbo from open ai and you can get a wall of tech
That's very different from getting something that, you know, Tim Ferriss said this and Joe Rogan said that, and then the producer said this. And between the combination of the three, we can infer how each one of them feels about a time. But if it's a block of text, you don't know, did Elon say it or did Joe say it? You know, who's smoking the joint if you're not watching the thing, if you don't know who's making the joke.
uh so the context gets lost without some of that so i'm curious how you guys feel about that that's where ilia like in arnav where apple did do the 100 million but they're this this wall of block it's just the ux be done and there's technical complexities you guys know for that and why that is you know more challenging so i'm curious I think it depends on how you use this, right? So if you want to read instead of listening. then non-dierized transcripts are fairly useless.
if it's an interview, right? Because you don't know who said what. The segments may have both people overlap. And we have non-diarized transcripts basically because they're cheaper. Yeah, for sure. For now, right? Yeah, they are much cheaper than diarizing the transcript. But if you're listening and actually you want to remember specific things and you want to come back and listen to them again, diarized transcripts are nicer, but they aren't strictly necessary. They're not critical.
Yeah, that's critical. Yeah, because if I want to actually retain that segment and I want to listen to it again. I just go to my bookmark, a tap play, and I just listen to what was said. Because the transcripts are pretty crappy anyway. And the reason is not the tech. The reason is how we speak versus how we write. And I remember listening to a podcast when Trump was first.
And the person on the podcast was saying that, I think about New York Times or some other thing, where they interviewed Trump and they published the entire transcript of the conversation unedited.
uh on uh in the interview and it has all of the ums and ums and uh like stutter the disfluencies disfluencies everything right and and and but then he said a very important thing he said if you want to humiliate someone And this kind of stuck with me because if you open the transcript, unless it's a scripted show... transcripts are not readable. If you listen to the conversation between Rogan and Josh Maitkin, it's back and forth, some of the words that the transcript gets wrong.
It's just not readable in my opinion. So I think actually there needs to be an extra layer of processing on top of the transcript. So if you don't want to listen, if you want to just get the insights, then you run an LLM on top of the transcript and then you get those insights. And then actually it doesn't matter who said what, because what you're after is the actual note.
And who said it is less relevant. Can I give you a counterfactual there? Please. Where if there's two speakers and they're arguing with each other, who says what actually does make a difference? Where does Tyler Cowen feel about it versus Satya Nadella? the show. So I would argue that the context of the speaker actually matters a lot. I think we're probably talking about the same thing. The whole decision or discussion comes down to, do you want to read?
or do you want to listen right because if you want to read then even a diarized transcript, like Ilya says, is not that a nice experience to read through all of the pauses and kind of back and forth discussions and all that. It's not like a blog post or a book. where you would expect an edited, polished format, right? Versus if you're coming to listen to it, then I think it makes sense. then the job of the transcript is to quickly find which are the important things for me to listen to. Yeah.
i think lucus's point if you actually because i think he was alluding to what i said about running llms on top of the transcript so if it's two people arguing the llm will not know uh who said what whose position it is and if there's a two sort of positions that cancel each other. There is no insight that you can derive from this. From non-diarray transcripts. Yeah. I'll give you guys an example with that.
I try not to talk about deep guests on the show, but we're sort of contemplating something that's like consensus view versus. dissenting view and doing something like this across 10 podcasts that are all discussing Deep Seek or Starlink in Ukraine or whatever it happens to be. And knowing who said what matters in terms of who's in the consensus view versus who's in the dissenting view, and then being able to sort of go on a citation basis.
just like research oriented and say this definitely was this person who said it and i can cite it if i'm a journalist researcher academic have strong citations. Let's talk about that because the question to all of you, will the AI...
a couple of examples. So I have a film background in documentaries. We rely... heavily to shape our stories right but those were written by humans we paid a transcription service they wrote them they got it right now with the ai transcripts they're getting it wrong My client was recently reached out to by a journalist saying, hey, we want.
You said she didn't say that thing, but the transcript from her very own website that was AI generated said that she did. So that's a big problem, right? And so is there AI? to this do you guys believe or do we need a human touch on this because i don't want apple to go back take everything i ever said for 10 years on food heels and say i said something you didn't say yeah thing i didn't say right yeah We need to differentiate between LLM-based AI. versus transcription AI, right?
so what we are not doing we're not generating uh text using an ai in metacast at least what we're doing is we're taking the audio and generating In some cases, we're diarizing it. So like Lucas says, then we know who is like speaker labeling it basically. So we know who's speaking what. But that turns out to be a little bit more expensive. So in most cases, we just do a plane.
We're not letting an AI, like an LLM, come in there and say, polish this up, which then adds the risk of the LLM hallucinating and adding things that were never in the audio.
right which we see all the time exactly yeah so it's also it's like diction matters right like if you pick a particular word as a if this is a dan carlin history like you wouldn't want the alternative word where like multilateralism gets picked up and it's like something else is said instead the the llm decides there's a better word alice and i think a lot of what arnav's getting at is there's a
cost challenge here today so like the way deepcast does it is materially more expensive than many other folks because we take what we get from our service provider but we do pre-processing and post-processing to accomplish what we're going to get
So our cost is probably, you know, two to three X more than what many other folks are doing. And it's hard to scale that if you're not Apple, if you're not Spotify, who has an infinite reserve of money like Apple does, I don't know, 180 billion sitting in Dublin. Yeah, exactly. Especially if you're bootstrapped and there is no venture capital helping. So a lot of it is technology.
but we all know that like these costs are dropping like speech to text models are dropping roughly on 10 to 20 percent and then llms are dropping orders of magnitude allison so i think this is a matter of time yeah and you know to ilia arnab's point like
Maybe Apple doesn't. We all get to benefit from it. Or maybe the RSS hosts, because it starts becoming cheaper, they all do it and we get to benefit from it. But it's a point of time, and I think we're going to get there soon enough. The question is exactly. Yeah, and also I wanted to add that I think with Apple putting transcripts in the app and also apps like ours, and I think Pocket Casts are going to add transcripts too.
So all of that evolution in the industry, it will also drive creators to maybe get better about their transcript. and provide transcripts. At the time when they publish, like you said, the RSS hosts, right? so for example if you upload something transistor transistor can um this is just jackson company they can transcribe the episode for you and then they also have the editor uh when i mean the the editor user experience right where you can go and edit
maybe some wild inaccuracies that are there. Or if the diarization is not correct, you can adjust that. But also, I think most of the industry now uses Descript for producing their podcasts. And Descript is a transcript-based podcast editing tool. And the cool thing about Descript, unlike all of the other sort of transcriptions... you upload speakers individually.
So you actually have diarization built in by design because they're different files. You don't have to guess who said what. This is a file from Allison. This is a file from Lucas. And they don't overlap. So you transcribe them individually. And then as you produce the part. If you want a good transcript, you can actually edit it right there in Descript and export it and upload it to Transistor or any other host.
And then apps like ours or apps like Apple, they will actually use the transcript from the RSS feed. And then, so I think the best of both worlds is actually the combination of the two, the AI plus human. Yeah. I mean, ideally you want the human to like edit the whole thing and but it's a lot of work. But if at least, you know, the human checks that nothing wildly inaccurate was said, then it would sort of good enough.
I do think, by the way, Ilya, I think Apple's running in their own by defaults. And even if you send the transcript in the transcript object, they don't include it. You have to go back into Apple Connect. and tell it to use your transcript so there is still a physical human driven non-api element which is something apple should just you know hey if they're delivering a transcript they put the effort to get it right
and probably take that. On the other hand, they may be looking for consistency of experience. I'm not entirely sure what the justification was for not taking free transcripts from the RSS host. as opposed to choosing to do your own so like we are seeing a lot of the transcripts that are in the rss feed are actually not
Maybe the podcast, the tool, the host itself is generating. Not a human is generating it. And maybe they're not even in the format it says it is in. Like sometimes we've seen a lot of cases. where the RSS feed says, oh, here's the JSON transcript for the episode. Like it is JSON, but it's not the format. Yeah. So maybe Apple will catch up to that. To be fair, we are also, we pretty much.
In our mobile app, we actually use our version of the transcript right now. We don't use the creators. On the web app, we have started blending that over. And very soon, I think on our mobile app too, we'll start. using the transcript from the RSS feed if that format is what it says. It's accurate. The other challenge that we often see is maybe the transcript is there, but it's an hour long episode and the segments are like 15 minutes.
and that's totally useless in an app like ours where you actually need to see the segments individually as the conversation sentence like i'm saying this sentence you want to see that individual versus if lucas you say something else that needs to be a second uh thing that i should be able to share independently right so there are challenges in there maybe that's why apple is not doing it we are definitely not doing it yet because of those same
I want to move to something that Allison kind of alluded to tangentially around AI in particular. So she was alluding to AI in the context of transcripts. There are all sorts of applications of AI possibly in the world of podcasting. Obviously, Deepcast has the things that we're doing. I'm wondering where you guys see AI having an impact over it.
time horizons being what they are and AGI maybe being here in two years or whatever happens to be. But let's say like within the next six to 12 months, things you think AI could be doing for podcasters. the podcast apps, service providers, like where does AI play like positive force for change or better user experience from your guys? On the podcaster side, I feel like usage of AI is pretty obvious.
Because the most difficult part in creating a podcast right now is the post-production. Post-production is just a freaking nightmare, even if you use Descript with all of these transcripts and editing. So I really want to see the world where... And it automatically adjusts all of the interruptions. All that stuff you have to edit manually that Descript doesn't handle for you, right? AI could recognize. I think Descript already does that.
like sort of technical conversations, sort of administrative conversations within the podcast that you just need to cut out, all that stuff. I think Descript is almost there, but still there is... a lot of manual editing required just because they don't really do anything with the audio. Then the audio quality, I think, could be also done by AI eventually. So I think there's tons of stuff that can be done by AI, like automatic show notes. like automatic chapters, automatic transcripts.
so i feel like there is a lot to be done there and uh i think it will be done within the next few years adding to that ilia i think on the preparation side itself there's like huge we see the benefits already right like We had Jason Fried from 37 Signals in our episode. The first thing I did to prepare for it was go to ChatGPT and say, hey, tell me everything you know about Jason Fried. What podcast has he been on? Where can I listen to him? And all that.
There were like wild hallucinations in there. It gave me some episodes that the podcast didn't even exist. It just made up things. But even with all that, the benefit that I got out of that, rather than doing all that research myself. it saved me like hours right there yeah right in terms of like preparing about uh figuring out what to talk about or getting to know your guests better You do have to verify everything that you're getting out of ChatGPT, but I think it is saving a lot of time.
i want to also switch the listener side uh i feel like In the last two, three years, there has been a lot of apps that want to provide you the summaries, the clips, all of this stuff, anything that will make you not listen to a podcast. So it's basically like, oh, why listen if you can just read the summary, right? And we intentionally steered clear from that approach in Metacast because we actually think that AI does not replace the actual experience you get from.
listening to the host, developing the connection. And also like keeping yourself busy while you're driving or like doing dishes, like a summary doesn't help there. You actually do want to hear the live. real voice of the real person. Even, you know, yes, you can have those AI synthesized podcasts and maybe they are fine for some types of content like news.
Like when people debate each other about certain topics or when they talk about their career or, you know, other things that you develop an emotional connection with. So I feel like on the listener side, I almost want to keep AI out of it in a way, right? So, I mean, I want to give them the tools like we are developing where you can search, you can discover all of that. But the actual listening experience, maybe it can enhance the audio to make the audio.
But otherwise, just like get it out of the way and let people enjoy the beauty of podcasting as it is in raw form. Yeah. That connection to the podcaster is why you're listening to a podcast. right right like if i wanted to learn why we have leap years in our gregorian calendar i would google that and probably read a blog post or have chat gp tell me very quickly why we I want to listen to Neil deGrasse Tyson talking eloquently in a hilarious way about the historical mishap.
that has led to us doing this yeah and that's because i know that he has talked about this before i've heard that And I'm there for that connection to his voice and the way he describes this rather than just the facts out of it. Yeah. Yeah, it's true. The one thing I was curious with Ilya and Arnab on that position. Sounds like, Ilya, you have a really strong conviction around it is. Neil deGrasse Tyson has no trouble being discovered.
Right. So people go and they look for him. Do you think there's value in the sort of the AI derived stuff for discoverability on the open web through Google, through ChatGPT, through perplexity, and the list goes on and on? Audio, as you know, is not a format that those guys are dealing with at all. So some of this derived stuff might actually help with discovery because you're looking for a podcast to discuss.
IP law in the world of podcasting and whatever, you're not going to find that through any other mechanism. Maybe it's through the search of the transcript and you don't actually need the AI derived stuff. It's just the raw data itself. Do you think there's value in discoverability? I think definitely there is. And yeah, I think it's a very good point because on one hand, I think all of the native apps will have that functionality.
uh but um if you want to generalize this to like in open ai where you search and it recommends you oh but you can also listen to like these guys talk about that topic too right and it would give you like this specific thing of the open ecosystem of podcasting is where does it refer you to? Like which app, right? Which player? Or like does OpenChatGPT become like a player too where it can also play you this thing?
This is also a challenge of the open podcasting ecosystem where it's just so fragmented. Luckily, we have the RSS standards. But even with that, there's just like a wild... because difference between different players to be fair that's the same challenge with google also right like google is all like if it finds that lucas has talked about transcribing on a podcast. Is Google going to give the top ranking result to Apple Podcasts or Spotify or to Mac?
or transistor or deepcast, right? Yeah, it's a similar problem. That's the same challenge everywhere. And OpenAI is going to do the same thing. Whatever has more critical mass is going to get better. Yeah, this is an area where I have some conviction, right? Obviously, DeepCast doing some of these things, but...
Interestingly for us, some of our primary ingress points to an episode is not someone looking for a podcast or even an episode title. They've typed something like podcast about Ukraine war. Well, we have a Ukraine war top. so they come in through the ukraine war topic in order to discover stuff or podcast episodes featuring elon musk we have elon musk as something so it's an ingress point and because we're open webs we're browser only
eligible for Google to dig into it. Obviously, a lot of opinions on this stuff, which we're going to see on the show all the time. Two questions, two parts. You guys take them however you will. How is podcasting broken right now? And how is podcasting working really well right now? I think what works well is that we have this massive amount of great shows people are passionate about.
their podcasts uh like the technology is really good it helps you produce those podcasts like descript i think was a game changer and described an anchor that allowed people to host podcasts for free So being able to follow a podcast like FounderQuest. for example, like a company you don't even know existed. Like they build an error monitoring tool and they sort of, it's a building public podcast. I don't know how many people listen to that, maybe like a thousand.
Maybe a few thousand. But it's like this sort of super niche podcast where you get to connect with the founders as they build their thing. like it doesn't exist anywhere else in any other medium so this is what i really like like about podcasting you can find in like there is a niche for for for anything and there is a sub niche within that niche So it's really awesome. And it's also a free ecosystem that's not behind the walled garden. So I really hope that Apple keeps it that way.
because they're our last hope. Spotify and YouTube, they're trying to make everything to close the ecosystem. But what's broken, I think, is discovery. Especially discovery of the niche content and discovery of the back. I forgot who, I think it was also in Justin Jackson's podcast. They did the survey. How do people discover podcasts? They hear about podcasts on other podcasts. That's how they discover them, which is a not sustainable growth strategy for a new pod. The Gossip Girl movie, right?
Like how are people going to discover you unless you like, I don't know, like it's really rough. We saw that when we started our podcast. Like it's really hard to bring people, bring listeners to your podcast unless you're already big. have a big name. So I think the long tail is amazing. It almost feels like it's locked up behind a glass wall that we can't yet break because we don't have the right tools.
Arnab, how about you? I think I would say something similar-ish. The beauty is the decentralized nature. I don't need a sponsor to go create a show unlike I would need to for TV, right? There's only a few limited ways I can do that. I need somebody to trust me, believe me.
and maybe i'll need some experience beforehand to be able to have somebody trust me that i'll do a show versus a podcast anybody and people do this all the time right anybody you have a passion you want to talk about it go talk about it host the files on the internet it's kind of like the old blog based internet ecosystem right anybody can host anything and anything any app any person in the world can And I think that's also the problem with podcasting is the decentralized nature.
because there are not really good standards and i know there's the rss spec there's podcasting 2.0 they're trying to do some really good things bringing in like standardization into it but that is not there right So as a listener, like if you want to know, should I trust this show? Like how good do people like this show? Where would you go to to find that info?
five different places you could go leave a reader comment about it on reddit maybe there are different websites again it's all very decentralized right now so i would say it's the Like most things, the biggest superpowers are also often the biggest weaknesses.
And in podcasting, that's very true. I love it, Arnab, that you just made our little intro snippet right there, Alison. He said, on both sides, it's like it's a gift and a curse and it's a responsibility and it's a superpower. I know. I was thinking anyone can do it. God, anyone can do it. Exactly. Anyone can do it. Yeah, exactly. The cool thing about podcasting, right, is like it's also not censored. I think it was really obvious during the COVID time, right?
There are different narratives in different countries, right? And like YouTube, they would just shut down anything that's not approved by CDC. They would ban people. They would kick people out. They'd be famous. Whereas on podcasting, easy. Like Rogan, I mean, people like to hate Rogan, but I know.
He had some legitimate people on his podcast during that time, right? That gave you a different perspective. And kudos to Spotify for not shutting him down. But even if they did, you could always get the RSS feed. load it into your overcast or other players that support that and listen to that anyway so that's like the open internet and that's what i really love about The strong benefits of that open ecosystem are also its downfall. Anyone can say anything.
by the software, who I will not name, that I was using to host my website, my sales pages. Everything was shut down during that time, except guess what I got to keep? My RSS feed. Arnav and Ilya, thanks for joining us on Sound Strategy. Where can people find you, which we'll put in show notes, and then also where can they find Metacast? Metacast.app. has the links to download Metacast on both iOS and Android front and center on the homepage.
And yeah, and if they want to follow us individually, they can just search our, I guess, find us on LinkedIn by name. Metacast.app slash about page. And that's where you can find our links and everything else. We'll link them.
We'll definitely do that. Love having you guys here. Thanks for getting into the nitty gritty of sort of technology and allowing me Ilya to play like point counterpoint with you on some of these things. I think that's what we're hoping to get in the show. So I really appreciate it. aging in that way. So thanks to the both of you. And obviously thanks to Alison as well. This was a lovely chat. Perfect. Thank you.
Thanks for tuning into Sound Strategy with me, Lucas Dickey. Find everything from today's episode at soundstrategy.fm. Transcripts, takeaways, key quotes, and links to all of your favorite podcasts. I hope you enjoyed our deep dive into the wild world of podcasting today. Check our show notes for all guest details and contact info. Sound strategy wouldn't exist without these amazing guests. So a huge thanks to everyone.
And most importantly, thank you, my podcast-obsessed friends, for listening. Come back next week for another edition of Sound Strategy with Lucas Dickey.