Mufin: An Automated Music Recommendation Engine That Actually Works

Whenever we get introduced to a site that features some kind of “digital fingerprinting technology” to power a music recommendation engine, it’s hard not to be skeptical. Fingerprinting works well for identifying the same song in multiple places – for example, duplicate songs in a music library. But when it comes to music recommendations, automated systems rarely work well, which is why Pandora relies on a team of 50+ analysts to drive its recommendation engine, and most other sites rely on social suggestions and meta data.

Today sees the private beta launch of Mufin, an automated music recommendation engine that actually seems to work(and it’s really, really cool when it does). The technology is sophisticated, but this isn’t surprisingly giving Mufin’s pedigree: the site is an offshoot from the Fraunhofer Institute, the German research organization that originally created the ubiquitous MP3 audio compression algorithm. You can grab an invite to the private beta here.

After searching the site’s 3.5 million song database for one of their favorite songs, users can press “Similar Tracks” to view a short list of songs that have been determined by the site’s powerful algorithm to contain similar characteristics. The algorithm analyzes 40 characteristics of each song, including tempo, sound density, and variety of other factors. Users can search for songs both from Mufin’s homepage and through an embeddable MySpace widget (with Facebook on the way).

In practice, these recommendations seem to work very well – at least more than 75% of the time in my testing. Many of my searches for popular songs (such as Here Comes The Sun) yielded results that sounded very similar, though many of these matches were in a different language or were recorded by obscure bands. To rule out the possibility of pre-determined matches, I tried a few more obscure recordings. The Darkness’ timeless classic “I Believe In A Thing Called Love” yielded “Sucker” by Europe. And Kyu Sakamoto’s “Sukiyaki” was matched with “Twilight Time” from The Platters – a song I’ve never heard of, but is unquestionably similar. Note: I’ve linked to YouTube clips for the above songs so you can hear the similarities – there doesn’t seem to be a way to embed Mufin results yet.

Unfortunately, while the technology behind the site is impressive, it still needs quite a bit of work. Many of the matches seem to be accurate, but the site offers no explanation as to why a given song was recommended. Mufin says that it will be including Pandora-like explanations detailing which characteristics have been deemed similar between each song, but for now you’re left in the dark.

The site also has a number of licensing issues to work out. Mufin has partnered with Universal Music Group, but many songs in the database are lacking preview clips, which make some matches useless as there’s no easy way to listen to them. Other songs in the catalog have previews but are missing links to online music stores like iTunes or Amazon, so there’s no convenient way to buy them legally. And results are riddled with duplicates from ‘Greatest Hits’ albums.

Beyond these issues, Mufin has a lot of potential, and could provide a viable alternative to Pandora or sites like once it cleans up its interface and squares away its licensing issues.