Others

Further details are given here, as accuracy 1 on clean and noisy data (all 3199 instances and 465 noisy instances), accuracy 2 on clean and noisy data (all 3199 instances and 465 noisy instances), overall accuracy on clean data (i.e. mean of accuracy 1 and accuracy 2) and overall accuracy on clean and noisy data (i.e. mean of accuracy 1 and accuracy 2, considering all 3199 instances and 465 noisy instances).
Detailed statistics on each dataset (`ballroom', `loops' and `songs') are also given.


NB: Slight changes in percentages after fixing a few ground-truth errors (on the `song' dataset) on November 16th 2004.