The Google Image Labeler is a game where you and a web partner try to agree on a label on an image.  The more images you and your partner identicaly label the more points you get.   Of course, at the same time, google is collecting all sorts of human-applied labels that it can then use to improve its image search function.  It's a great way to get humans to help classify images without having to pay them

It seems that a similar technique could be used to help apply labels to music.  The labels could then be used to help cross the semantic gap, and allow people to search for music based upon content-based keywords.  Teaching computers how to extract meaningful acoustic labels from audio is a hard job.  Getting humans to provide the labels by getting them involved in an engaging game, where the labels are generated as a sideproduct would be a great way to collect this data.  The resulting labels would be extremely useful for researchers and music search engine creators.   Of course, audio and images are very different media.  A game that works well with images won't work well with audio. So the trick is to figure out how capture  the labelling process in an engaging activity.  Sounds like a good masters thesis for some music technologist.
Comments:

I think Google just licensed the idea/game from a CMU research project. http://www.espgame.org/

Posted by Jeremy P on September 05, 2006 at 01:52 PM EDT #

Post a Comment:
Comments are closed for this entry.

This blog copyright 2010 by plamere