Monday, April 7, 2008

Startup Echo Nest's search tech 'listens' to online music

By Brendan Lynch

A Somerville startup founded by former MIT Media Lab researchers has unleashed a "musical brain" onto an unsuspecting Internet.

The Echo Nest Corp.'s musical brain is a web-crawler that can listen to music as well as analyze text. The company released its first application programming interface, or API, demonstrated in the website thisismyjam.com, last week. The website is intended to showcase what the musical brain can do, according to CEO Don Rose.

Music is the No. 2 search category on the Internet, behind sex, according to Rose, who has set a lofty goal for the Davis Square company.

"I think our competition is Google," Rose said. Rose, a co-founder of record company Ryko Corp., and former president of the American Association of Independent Music, was brought in a year ago, after a mutual Media Lab acquaintance introduced him to the co-founders.

Thisismyjam.com isn't meant to bring in money, according to Rose. The company plans to license subsequent APIs -- it has about six -- focused on music search, personalization, and recommendation, to companies such as Apple Inc., Rolling Stone, Rhapsody or any other company that deals with online music. The company estimated the online music market at $7 billion and growing by 40 percent a year.

The Echo Nest was founded in the summer of 2005, focusing on R&D until last week's API launch. The company is angel-funded by MIT professor and Media Lab founder Barry Vercoe, also an audio adviser to the One Laptop Per Child project. The company also won a $150,000 National Science Foundation grant in 2006. The seven-employee company is hiring two software engineers.

Thisismyjam.com is an example of the API the Echo Nest is calling "Analyze," a web-based application that quickly reads a music file and creates an XML file describing the music's tempo, beats, key and other attributes, for use in visualization, games and remixing. Users can take song clips and make a mix of song clips that slow down or speed up to blend into one another.

"We know what the beat is," co-founder and co-chief technology officer Tristan Jehan said. "We know what the rhythm is. We can do what a DJ would do, which is align the beats in the best way possible."

But the brain doesn't only listen to the music. At the Media Lab, Jehan worked on a team doing acoustic-music analysis, and fellow co-founder and co-CTO Brian Whitman worked on a team doing contextual music analysis. Over a coffee break one day, the pair realized the best method was both methods. For example, software that only listens to music might recommend Christian rock to someone interested in rock, Jehan said.

"It's going to sound similar, but culturally there is a huge gap," he said.

John Wilbanks, director of Science Commons, a Cambridge-based semantic web nonprofit, said there's value in pairing semantic text search with the audio-listening technology. He said similar pairings have worked with biotech trial databases.

"Semantics go back down into what the math thinks is noise and find something meaningful," he said. "The weakness of semantics can be counteracted by statistics and vice-versa."