BigChampagne's Ultimate Chart (a/o February 10th):
1. “What Doesn't Kill You (Stronger),” Kelly Clarkson
2. “Set Fire to the Rain,” Adele
3. “Turn Me On,” David Guetta
4. “Rack City,” TYGA
5. “Good Feeling,” Flo Rida
6. “Smooth Criminal (Glee Cast Version),” Glee Cast
7. “Young, Wild & Free,” Snoop Dogg & Wiz Khalifa
8. “Domino ,” Jessie J
9. “Sexy and I Know It,” LMFAO
10. “Ass Back Home ,” Gym Class Heroes
Lots of financial resources have been poured into recommendation startups, though this is still a riddle for entrepreneurs. Pandora has successfully approached the space with human reviewers and smaller artist lists, though a more brainier mindset continues. The latest comes from Songtweak, which is bubbling out of Georgia Tech's College of Computing.
The idea, at a top-level, is to identify portions of songs that listeners love the most, and use that intelligence to deliver tighter recommendations. "Our platform was built on the observation that it's quite common for two people to like a song for different reasons," cofounder Mike Genovese told Digital Music News.
On the service, users interact with a 'tweakbox' that receives input on elements like instrumentation, the beat, genre, aggressiveness, and the artist. The premise is that repeated tweaking will serve super-targeted results over time.
This is still a very early idea and a tad rough, but one challenge is to avoid plunging too deep into obscurity. That was the motivation behind aggregating the recommendations of hundreds of music bloggers, essentially an army of arbiters recruited to screen more promising recommendations.
Comments Closed
Maxwellian Friday, July 23, 2010
OK. so some super computer can tell me what I like, by seeing whether I like this part or that part? I don't think my tastes can be broken in 1s and 0s. Its so much more than that, I see this being limited in the ability to offer great discovery.
Mike Genovese Friday, July 23, 2010
Just to clarify, Songtweak is about trying to help you use songs you are familiar with to find new songs. Given a favorite song, it offers suggestions of similar songs that you may enjoy. The tweakbox lets you refine your results based on what aspects of the song you like or dislike. The tweakbox is provided to allow the user to have more control over the music discovery process.
More information can be found on our about us page.
SteveDb Tuesday, July 27, 2010
The problem is in the way the question is phrased to the machine leaves out the possibility of music working in ways that aren't really at the "part" component level. An example is: the best electronic music has a new musical object at least every 8 bars (Ex. Breeder's Remix of Orbital). Each addition is not that significant by themselves, but by the time the climax arrives it achieves real intensity. But this never happened very often, because it's a lot more difficult. It requires much more time to acquire and organize 4-16x as many musical ideas as a "minimal" type track.
So how do you tell a digital intelligence: make it have an "optimal musical development" every 8 bars?" As opposed to a hummable melody or a catchphrase sample? But this is something human reviewers would catch immediately.
Another more recent example is LCD Sound System's latest which didn't make much of an impression until one of Pitchfork's writers pointed out it was a hidden "parody" type composition, based on David Bowie. I listened again and was impressed by the subtle "meta" in the way the piece worked. How do you tell a machine intelligence to search for that? Compare it to every song ever written, how?
And then of course, for a lot of non musical people, a song is their favorite because that's when someone proposed, or the Sun came up after a full moon, or some other personal story happened to be soundtracked with a certain record. Definitely a digital difficulty defining that!
trevor Wednesday, July 28, 2010
I know there are differences, but people used to say the same things about Pandora. Like, how can you possibly make recommednations or fine-tune people's stations based on up/down ratings or musical attributes? It's far too subjective, music has too many nuances and emotional attachments, etc. etc.
Anyway now it has become a major way to listen, enjoy, discover, etc. And it really hasn't replaced other ways of listening, it iTunes, iPod, etc. So I think you have to let the experiments breathe a little to really see if there's an application for the general populace.
OUR SPONSORS
© 2011 All Rights Reserved
Follow Us