Forget Pandora – Brooklyn’s Clio Uses Machine Learning To Recommend Music

The Music Genome Project conceived by Pandora is the foundation for one of the most successful streaming radio apps of all time. But any serious audiophile who spends more than few days on the service quickly begins to notice the same tracks and artists repeating.

Part of the reason the catalog feels shallow is because its expensive to license new tracks. But another reason is that the Music Genome relies on human experts to evaluate each track, an extremely time consuming process. It’s why Pandora has less than 1,000,000 songs while competitors like Rhapsody have ten times that.

The team behind Clio, a new service for analyzing music, is taking the exact opposite approach. Greg Wilder, the founder and chief scientist, is a pianist who left the conservatory scene to build a start-up instead. He sold his concert grand, quit his university job and taught himself how to code. The algorithms he created can pick out the hook in a pop song, the back beat being put down by a drum or the tenor of the lead singers voice. It uses these elements to match similar tracks.

“We’re just looking for the patterns that appear to our mind as music. It’s all math, in a certain sense,” says Wilder. This ignores, of course, that much of music is the context of history and genre. But Clio sees its value in breaking huge libraries of music down, freeing human curators up to make the fine distinctions. “Human beings are great, but they just don’t scale,” says COO Alison Conrad, herself a former music theorist.

Traditionally this kind of work relied on keywords, which meant that over time huge amounts of subjective tags and identifiers built up around the libraries of tracks. “That made it really expensive for companies to integrate their collections. Since we are relying on the music itself, not the external metadata, we can organize large collections much faster and cheaper.”

The company just launched a new website and announced its service is being piloted by ten major organizations. That includes big production libraries like APM, which provides the background scores for everything from TV commercials to Hollywood blockbusters. They are also being tested by one of the premier streaming services – Rhapsody, Pandora, Spotify – although the company won’t confirm exactly who yet.

Forget Pandora – Brooklyn’s Clio Uses Machine Learning To Recommend Music