Wade Roush
We’ve hit the time of year dreaded by all public radio fans: pledge drive season, when stations hold your favorite NPR and American Public Media programs hostage until enough listeners pay up.
Until recently, there’s only been one way around this semiannual torture ritual, and it’s only been available to listeners of KQED here in the San Francisco Bay Area. It’s the pledge-free stream, a password-protected online version of the station’s programming that’s available to listeners who give $45 or more.
But now there’s another way to enjoy a radio-like stream of news and information shows, without pledge drives, at any time of the year—and at zero cost (to users, anyway). It’s a smartphone app called Swell, from the Palo Alto, CA, startup Concept.io.
You can think of Swell as your own personal NPR station, or perhaps as the Pandora of news and talk. Yes, that’s an analogy I’ve used before, in a January article about the on-demand radio app Stitcher. But Swell has adopted a set-it-and-forget-it approach that makes it even more consciously Pandora-esque than Stitcher. It’s like turning on your favorite public radio station, and letting little software-driven station managers inside your phone worry about the exact content you’ll be served.
When you start up Swell—which is available only for iOS devices at the moment, but is coming to Android soon—it automatically picks a program from the Swell catalog based on your past listening behavior. (The catalog isn’t drawn just from NPR—it’s also heavy on programs from APM, TED, the BBC, Disney/ABC, Fox, and Bloomberg.) If you like the app’s selection, you can keep listening until it ends, at which point another show will start. If you don’t like it, you can swipe left to skip straight to the next program.
Over time, as the app observes which shows you listened to and which you skipped, it gets better at picking stuff you’ll like. (Yes, this is exactly the sort of mechanism that screens out anything that might jostle your existing world-view; Upworthy co-founder Eli Pariser has called that the filter bubble problem. But let’s face it, public radio is pretty uniformly lefty to begin with. And hey, you can always get Fox.)
This play-or-skip model is more or less how Pandora works too, but because Swell is about ideas rather than music, the technology under the hood is fundamentally different, says G.D. “Ram” Ramkumar, CEO and co-founder at Concept.io. Oakland, CA-based Pandora (NASDAQ: P) has built a huge “music genome” database to identify the traits of the songs and artists you like, on the theory that you’ll also like other songs with similar traits. Swell, by contrast, relies on algorithms from probability theory related to the multi-armed bandit problem and a strategy called “exploration and exploitation.”
It’s too complicated to explain here, but it boils down to tracking your listening patterns and identifying the program choices that best predict your preferences. There’s also an element of old-fashioned collaborative filtering, in the mold of Amazon or Netflix recommendations. For example, if you liked This American Life, and other people who liked TAL also liked The Moth Radio Hour, Swell knows you’ll probably like The Moth too.
“This notion of picking the source that helps the most, and combining that with an element of collaborative filtering, is new technology that didn’t exist before Swell,” says Ramkumar. “The point is, as you continue to listen, it understands and learns at the level of programs, topics, and genres.”
Here’s a cute video that Swell produced to explain the whole thing. (Article continues below video.)
Concept.io released Swell in late June. So far, according to Ramkumar, it’s being used mainly by people who are running or exercising with earbuds, or commuting in their cars while playing the audio stream over their vehicles’ Bluetooth audio systems. Those happen to be the exact scenarios the company had in mind when it built the app, he says: “It’s about making productive use of your time, and staying engaged in the time you are driving and exercising.”
Ramkumar’s previous company, Snaptell, also built a consumer app: a visual product scanner that used computer-vision algorithms to identify products from cameraphone snapshots. Amazon bought the company in 2009 and incorporated the technology into an augmented-reality app called Flow. Ramkumar spent a while at Amazon’s A9 research operation as “chief architect for visual search,” then went on to an entrepreneur-in-residence position at Charles River Ventures, where he started looking for another area of consumers’ daily lives that he could help to improve through mobile technology.
“We settled on the commute as the critical time period,” Ramkumar says. “If you want to listen to music, there are good alternatives, but if you don’t, there are limited choices. There is the choice of searching TuneIn or iTunes, but search is not convenient when you are driving.”
After abandoning a couple of early experiments—such as a text-to-speech engine that recited Wikipedia articles based on nearby landmarks—Ramkumar’s team turned its focus to … Next Page »
Comments | Reprints | Share: 
