“Water, water everywhere, nor any drop to drink,” reads by far the most famous — and widely quoted — line from Samuel Taylor Coleridge’s epic poem “The Rime of the Ancient Mariner.” It describes the predicament of being surrounded by a quantity that you’re searching for, but being unable to make proper use of it.
Right now, the world is drowning in podcasts. In February 2018, there were an estimated 500,000 active podcasts in existence. Today, that number is in excess of 1.7 million, with a total of more than 43 million episodes. And yet, for all that, podcast discoverability is, to put it nicely, horrendous.
This is where a new startup enters the picture. Podz, co-founded by a team who first met while working at Yahoo, seeks to find a way to solve the findability conundrum that besets today’s podcasts. More than that, though, it’s got far grander, far more significant designs: To do for the world’s audio archives what Google has done for search. Namely, to organize it and make it universally accessible and useful.
“We like to say that we’re living in the golden age of audio,” Seye Ojumu, chief technology officer at Podz, told Digital Trends. “But some of the tools that we have for finding things [to listen to] still feel like the Stone Age.”
Audio, Ojumu points out, is currently in the zeitgeist. Podcasts are huge. Clubhouse is blowing up. AirPods and other wearable listening devices are flying off shelves like they’re going out of style. To put it simply, people like listening to stuff. Or, as Ojumu phrases it, “increasingly, people are finding themselves in audiocentric, passive consumption experiences.”
Currently, there are a few ways people typically discover podcasts — and none of them are quite scalable enough to measure up to the scope of the challenge. They’re recommended by friends, they’re advertised on podcasts we already listen to, or they’re sufficiently notable that they pop up in the pop culture ether, the way a planet-sized hit such as Serial did a few years ago.
There are a few search capabilities in podcast apps, too, but these are fairly basic. While they may be fine for finding, say, that podcast hosted by your favorite comedian, that’s about their limit. Sure, they might extend to helping you find the episode of said podcast where, for instance, Joe Rogan interviews Elon Musk — but what if, having enjoyed Rogan’s conversation with Musk, you now want to find all other podcast interviews with the Tesla and SpaceX CEO? That’s tougher, but still achievable.
Now, what if you were interested in something Musk said during one of his conversations — maybe about the simulation hypothesis — and wanted to find every time Musk had spoken about this topic on a podcast? Or, to add yet another wrinkle, what if you wanted to listen to every time someone with comparable clout to Musk has chimed in on this topic. Or every person with comparable clout who is in conversation with an upbeat host with a British accent, in a podcast recorded in 2021.
One of the many problems faced by people trying to crack podcast discovery is that podcasts tend to be big and sprawling. As with any conversation, people skip around from topic to topic. While everything else in tech seems to focus on brevity — one-click shopping, 10-second TikToks, nuanced conversation distilled into tweets — podcasts remain defiantly long-form. Great for listeners, poor for searching.
In an attempt to “fix” podcast discoverability, Podz has created an A.I. trained on 100,000 hours of audio, which scours through the most popular 5,000 podcasts (that’s the overwhelming majority of podcasts most people listen to) and creates the most engaging 60-second sample snippets to populate an audio news feed.
As with Spotify’s music recommendations, over time it will get smarter as it learns your tastes. The user can jump from one sample to the next as if they were watching Instagram Stories or swiping through potential dates on Tinder. If they like the sound of a particular podcast, they can dive in and listen to it.
“We basically annotate all of the audio, so that we know — to a first approximation, and it’s slightly editorial — but we have a baseline understanding of the category that the audio is in,” Ojumu explained. “We have a baseline understanding of the hosts that are actually hosting the show. And, as we get better at this … we’re going to have an idea of the guests who are speaking. With those pieces of information, you now have a way to navigate from one piece of content, from one short-form audio to another short-form audio.”
If that was all there was to it, Podz would be a neat idea. But that’s not all there is. Despite its name, Podz isn’t only interested in solving the podcast discovery problem — that just happens to be the most prevalent example of this challenge at hand. Thinking this is all about podcasts is like thinking that Apple is a company whose raison d’être is to sell the iPhone 12 Pro Max. Sure, that’s what it wants now, but that is one small, short-term goal in a much larger mission. What Podz really wants to do — and, if it can pull it off, this is a multibillion dollar idea — is to make the audio space as searchable as the text space.
“Today, [the focus is] primarily going to be podcasts,” Ojumu said. “But it could be anything– the original source could be video, the original source could be a speech a person has given. Anything where there’s a person or persons talking.” He described the team’s eventual goal as building a “general-purpose audio search where you can search all of the audio that has ever been indexed.”
While this is, at present, still a pipedream, the goal is to develop A.I. tools that can comb through all audio and extract just the bits you want, sorted by speaker, subject matter, emotion, factual content, ideas, etc. “We could [theoretically] index everything said that was recorded by anybody, anywhere,” said Ojumu.
Right now, this is “fly before you walk” stuff for a startup — albeit a well-funded one, with investors including Katie Couric and Paris Hilton. But even if it’s not Podz that ultimately cracks this problem, it’s going to be one startup or another that does it. And when they do, the results could be significant. “[If you did this, it would be possible to] understand a little bit about the person who’s saying [something], understand a little bit about the nuance, understand the context, be able to put it into time and place, and then be able to intelligently search that,” Ojumu said.
The rewards could be massive — both for users everywhere and for whichever company that manages it. “If you have that, you would be Google, right?” he said. “You’d be Google for audio.”