Sound Searching: MediaMined offers sound-object recognition

National Science Foundation
For Immediate Release
New software tool provides unprecedented searches of sound, from musical riffs to gunshots.

Audio engineers have developed a novel artificial intelligence system for understanding and indexing sound, a unique tool for both finding and matching previously un-labeled audio files.

Having concluded beta testing with one of the world's largest Hollywood sound studios and leading media streaming and hosting services, Imagine Research of San Francisco, Calif., is now releasing MediaMined™ for applications ranging from music composition to healthcare.

The company developed the tool with support from the National Science Foundation's Small Business Innovation Research program (IIP-0912981 and IIP-1206435).

"MediaMined™ adds a set of ears to cloud computing," says Imagine Research's founder and CEO Jay LeBoeuf. "It allows computers to index, understand and search sound--as a result, we have made millions of media files searchable."

MediaMined™ is a two-part process—in order for a sound library to be searchable, it first has to be indexed. Once that's done you can search the indexed library's musical or sound content by sound samples. Kind of like saying, "Find me something that sounds like this riff from Jim Hendrix" and submitting a few bars from Hendrix's stunning live version of "Gloria". If it's really intelligent and you're really lucky it will return tons of other riffs from Jimi Hendrix that you've never heard instead of other guitar players who tried to sound like Jimi Hendrix playing a riff.

But on a more serious note, this opens up all kinds of possibilities for music-matching and recommendation services beyond "Customers who bought this also bought these" and genre-based lists. The possibilities are vast and the only question is how long will it be before Google makes them an offer they can't refuse (kidding, sort of).

deckeda's picture

... what SoundHound does just for music. You hold up your iPhone (or anti-iPhone, for the sadly Apple-averse) up to a speaker in a restuarant (or simply sing into it) and it spits back the artist and song info. AS IF BY FREAKING MAGIC.

Michael Lavorgna's picture

Matching A = A is a much simpler proposition than matching A to X, Y and Z. The former method doesn't need to 'understand' in any way what's it's matching while the latter needs to be able to differentiate, sort and prioritize a whole slew of criteria.