I am happy to report that our massive revision has been accepted with minor revision! We are almost converged.
Paper deadline coming up soon: EUSIPCO 2011!
And I have learned of this interesting pattern recognition competition for music signals. In the music instrument recognition task, the authors have decided that features extracted from 40 ms segments are the best. There is already plenty of evidence that making such a naive assumption will limit classifier performance in such a task. For the music genre classification task, they have computed MPEG-7 features from 20 segments of unknown length. Strange. Still, this might be a pretty cool exercise to motivate the relevant portion of my class on artificial intelligence.