Thursday, December 27, 2007

Acoustic Recognition: Toward teaching a machine to label music

The New Scientist Technology Blog posts about Major Minor, an on-line music labeling game by Dan Ellis and Michael Mandell at Columbia University (USA).  The goal is to build up a database of sounds that have been tagged by users (all humans, presumably) with descriptive words.  The scoring system seems very well thought out to generate appropriate tags, weed out inappropriate ones, and to motivate participation.  

Why does the world need such a database?  Well, just like with humans, in order to teach a machine (e.g. a computer) to recognize some characteristic of a sound, you have to feed it many, many examples of sounds with that characteristic for it to learn from.  In this case, the researchers use the Internet to provide the means to efficiently and effectively get people to the "test site" and their well-constructed, slightly addictive game for the incentive to get the test subjects to willing participate and give up their neural processing time to help out.  Good luck to the team at Columbia U!

No comments: