One of the most ambitious of these services, dubbed Soundflavor, is launching Thursday. The service, produced by a company called Siren Systems, promises to deconstruct songs into more than 700 component parts and then make music recommendations based on how listeners' tastes match these musical elements.
Many new music recommendation services are popping up with the goal of helping listeners find their way through the multitude of music options online.
While leading people to undiscovered content, the companies may drive the demand for digital music. But can computers make decent recommendations, and are companies willing to risk their objectivity by using paid search and advertising?
"Right now, the digital music market is like a big haystack being dumped over your head," Siren President Pete Budlong said. "A recommendation engine with good relevancy is like a strong magnet; you can pull a needle to the surface."
While still very far from perfect--and sure to face the continuing skepticism of listeners loath to believe that computers can perform the subjective task of recommending music--the new tools could be an important part of driving on sprawling services like Apple Computer's iTunes, analysts say.
"There's a lot of interesting potential here," said Michael McGuire, an analyst with GartnerG2, a division of the Gartner research firm. "It's going to be crucial for online music in general to have this kind of third party that stands outside the music services and outside the record labels, providing that kind of search."
Several companies hoping to perform this task in the music business rose briefly to prominence in the late 1990s, attracting considerable attention and venture capital funding. Most of those companies were ultimately --including Microsoft, in--and folded quietly into other services.
However, the idea of "collaborative filtering"--finding other people who have similar tastes and then trying what they like--has proven to be a lasting legacy.
Amazon.com's recommendation engine is based in large part on that idea. It tracks what a buyer purchases and views, and then makes recommendations based on the actions of other people who have made similar buying decisions.
Several of the newest tools are also grounded in this idea. The, for example, allows people to view the play lists created by other subscribers to the service, discovering music through others' listening habits.
Apple's iTunes has a rudimentary version of this, posting play lists from celebrities such as Moby and Sting. An independent company has bolstered this with a new Macintosh-only iTunes plug-in called Goombah, which analyzes a listener's downloads, looks at other people's play lists and makes recommendations based on the listening habits of people with similar tastes.
RealNetworks' Rhapsody has built in a different kind of system, allowing Web log authors to post Rhapsody downloadable play lists directly to their blogs. These play lists can only be used by other Rhapsody subscribers, for now.
The human touch
Soundflavor takes a different approach. The company's founders spent several years developing a list of attributes they thought would accurately describe music--everything from beats per minute to whether a saxophone or female vocalist was used in the mix. They then hired actual people to listen to songs and rate them along those categories.
Their assumption was that using so many attributes might result in some surprising but still interesting recommendations, such as finding a rock lover a jazz song that matched his tastes in almost every respect other than genre, for example.
Soundflavor is launching Thursday with information on just 4,000 songs in its database, largely focused on alternative rock. This is because its initial test customers were rock radio stations, Budlong said. That database will expand as new songs are analyzed, the company says.
A Soundflavor account is free; people pay for the songs they download or CDs purchased.
Relying on human ears is expensive, as previous companies have found, however.a large staff of musicians to rate and review songs but ultimately laid most off, as the company's funds dwindled.
Ultimately, Siren hopes to strike relationships with song distribution services such as iTunes, Napster or RealNetwork's Rhapsody. It may also ask music labels to pay for the costs of analyzing their music, if they want their music in the recommendation database more quickly, Budlong said.
Those revenues are important for business reasons. But that also raises the danger of turning ostensibly objective recommendation tools into the equivalent of paid search, where advertisers or sponsors get favored treatment, Gartner's McGuire said.
"It will be interesting to see how immune or roped off the system is from the pernicious effect of things like paid search and advertising," McGuire said. "They will need to go to great pains to let people know that there is money exchanging hands."