Software Designed to Help You Pick New Tunes

LAS VEGAS -- Most people find out about new music on the airwaves, in magazines, or from friends. Software may provide another avenue, by analyzing the music you already like and recommending similar tunes.

At least that's what some researchers and vendors believe, and they've been developing technologies to make it happen. Among them, Gracenote said at CES today that by midyear it will offer a product for online music stores that will let them make smarter music recommendations for their customers. And a project partially funded by the European Union said this week that it is ready to start licensing a handful of similar technologies to service providers and consumer electronics makers.

The basic goal is to go beyond the names of artists and genres to help people find music they like, instead analyzing the properties of the music, such as its rhythm, tempo, and energy level.

Early efforts relied mainly on signal processing techniques to uncover low-level similarities in music, such as its tempo and mood, says Xavier Serra, who is managing the E.U.-funded Semantic Interaction with Music Audio Contents (SIMAC) project at Barcelona's Pompeu Fabra University. That approach was sufficient to roughly group tunes with similar properties, but it might still have linked a fast-paced classical overture with a thumping techno beat.

More recent efforts incorporate other data as well, such as input from music fans and reviewers, which is appended to songs stored in giant databases that contain millions of tunes.

Micro Genres

Gracenote says it has its own team of experts who tag songs with one of 1600 "micro genres" used to link similar music styles from a variety of roots. For example, a company spokesperson notes, "Classic Motown and Psychedelic Pop are fairly different musically and are traditionally presented under separate categories of R&B and Rock, but are still strongly related and complementary from other perspectives."

The company already offers products for identifying the tracks on a homemade music CD and for organizing related tunes into a playlist. When its Gracenote Discover product comes out later this year, it hopes online stores, MP3 makers, and others will use it to help listeners find new music.

The SIMAC project also uses information from fans and reviews alongside signal processing to uncover related music. One of three products it hopes to license to the music and consumer electronics industries goes a step further: It draws on a separate project called FOAF, or Friend of a Friend, which is developing a way to make home pages on the Web readable by computers, so people can track down others with related interests.

Merging information extracted from a person's music collection with other factors in their FOAF profile, such as their age and socioeconomic background, as well as their explicit music preferences (the system can still incorporate genres like jazz and reggae), will allow the system to filter results more closely and produce better matches, Serra says.

Risky Business?

Such products come with a risk for service providers. Offering too many questionable matches can undermine customers' confidence in an online store, Gracenote admits. And the systems can require some intervention from end users just to get the style of music right, never mind to filter out the songs they don't like. "If you're a hip-hop fan, you might get 80 percent hip hop, but you might also get 20 percent techno, so you give the user the possibility to filter the selections," Serra says.

Other commercial vendors are also tackling the challenge. Predixis makes a plug-in for Winamp that analyzes the acoustic attributes of digital music to organize a big digital music collection into genres. Researchers at Sun Microsystems labs have developed "Search Inside the Music," which also analyzes songs by melody, tempo, and rhythm to group related tunes into playlists.

"It can also examine your music collection, get a sense of your listening tastes, and suggest new songs based on your preferences," says a Sun spokesperson.

Using computers for a task as subjective as matching music tastes sounds like a tall order. But if they pan out, the new systems could help people find little-known tracks they really like but might never have discovered otherwise. And it could be another step in the Internet's promotion of smaller artists without big-name recording contracts.