I've been spending a good deal of time delving into the jAudio package from McGill University. jAudio is an open source package for extracting features from music. The features can then be processed by other packages such as Weka, M2K or ACE. The authors of jAudio have done a very good job of architecting, coding and documenting the jAudio system. The jAudio code has been just super to work on. I've added some code to jAudio to allow us to distribute our feature extractions over a large number of CPUs. Since feature extraction can be a very CPU-intensive task, being able to distribute the task is key. With jAudio we can process audio at about 10 seconds per 3 minute song so processing a 5,000 song collection will take about 14 hours, but with the extended jAudio it is just as easy to run this task on 10 CPUs and get it done in 2 hours, or run it on a 1000 node system and get the job done in just a couple of minutes.

You can read more about jAudio in Cory McKay's paper jAudio: Towards a standardized extensible audio music feature extraction system


Post a Comment:

This blog copyright 2010 by plamere