How do you discover new music and artists? Probably through friends, the radio, or by wandering the aisles of a Virgin Megastore. Time consuming? Yes. Effective? Rarely. Two companies have recently entered the scene with the intent to turn your musical affinities into recommendations for other titles, albums, and artists, and, not only could it revolutionize the way consumers find music, but it has implications in a host of other consumer arenas.
Recommendation engines run by companies like Savage Beast and Siren Systems have developed highly advanced methods to determine what kind of music is similar to whatever your musical taste du jour may be that are far more intuitive and intelligent than a traditional text-based search. They analyze hundreds of attributes of songs in order to best categorize each selection, including beats per minute, instruments used, gender of vocalists, and content of lyrics, then use that data to tailor a list of songs or albums that a user already knows and enjoys.
Savage Beast's president Tim Westergren explains that they look at approximately 400 different traits, which they refer to as "genes." (A tenor sax solo within a song, for example, can have seven or eight genes of its own to help describe the piece.) These genes are stored in the Music Genome Project, a proprietary database that currently holds data on over 350,000 songs. Siren Systems uses their SongScore algorithm to compare 700 data points in songs before compiling its personalized list for users via the consumer-facing Soundflavor Web site, which is significantly smaller at this stage of the game with about 5,000 songs.
Both Savage Beast and Soundflavor use a combination of human analysts and automation to catalog songs—which generally takes about 15 minutes per song to complete—and both are adamant about the importance of using trained professionals to catalog music. "Our intention is not to be the arbiters of what's good or bad," says Westergren, so they train analysts, who are also musicians, for approximately 40 hours and often run duplicate analysis on songs to ensure that they have been appropriately and objectively handled. Soundflavor chose a more automated route in an effort to ensure that there was nothing subjective in the categorizing and analyzing of content. "We spent two years in R&D creating a metadata model that is objective," says Pete Budlong, cofounder and president of Siren Systems. "I call it metadata on steroids…so people do not decide if a song is ‘happy' or ‘sad.'" Analysts for Soundflavor use a text-based guide for each song and are both trained on the tool and tested before they are allowed to use it independently.
Because the Soundflavor Web site was preceded by a relationship with KROCK radio, the site had an initial focus on alternative rock, but it is following a genre map for the future. They are also allowing artists to submit music for free for as long as the site is in beta, and Budlong predicts they will charge a nominal fee thereafter—likely in the area of $20 per song. As for Savage Beast, "Our focus right now is to get all of the stuff on the charts," says Westergren, and the database grows from there. The company also accepts unsolicited submissions from artists.
According to Westergren, Savage Beast's approach is, "sort of a musicological study of not just what gives a song its style, but how it might correlate with other's preferences." They offer their database to retailers via the Web or in-store kiosks that are fully customizable to each retailer. Best Buy, a Savage Beast marquee customer, operates approximately 60 Web-enabled touchscreen kiosks that essentially act as a smarter version of the traditional listening station. Customers insert a CD into the kiosk and the recommendation engine offers similar albums. All songs can be previewed and Westergren anticipates that in the near future the touch screen will also offer options allowing customers to create their own CDs.
Soundflavor's approach is extremely different. Budlong views the future of music discovery as, "more active and less passive, with the consumer in the driver's seat." Hence the next iteration of the company's site, expected for release this spring, will retain more information about what users like and do not like, so each visit will result in a more and more tailored experience. All results lists are also relevancy ranked, à la Google, so users do not have to scroll through pages of unwanted or irrelevant material. To use Soundflavor, anyone on the Internet can visit the site and play with predesigned playlists or explore based on favorite songs, albums, genres, or artists. Available playlists, or Soundflavors, are based on activities and moods, such as "Remedy for a Bad Day," "For the Working Man," "Pumpin' Iron," or "Instrumental Only."
Westergren views recommendation engines as, "the logical next generation of search" and he believes that, "any product area where you can describe the product with a consistent set of details and rich analysis is a candidate for this," citing other possible categories like film, food, and books. Budlong expresses similar expectations for the use of recommendation engines, particularly in the entertainment space, and he predicts that, "in the long run, a metadata model combined with collaborative capabilities is the one that will win out."