We present a method for significantly improving the results of drum loop slice classification. An onset detector is used to slice loops of percussion only audio. Low level features are extracted from the audio slices and the slices are classified into one of seven percussion classes by a previously trained PART decision table. This general classification algorithm shows only an adequate performance. The user is then allowed to correct incorrect classifications. Each corrected classification is combined with a subset of the original classifications and a nearest neighbour algorithm reclassifies the remaining slices according to the corrected local model. The resultant algorithm converges on a 100\% correct solution, with nearly 40% fewer re-classifications than a non-assisted approach.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
