This paper explores automatic feature learning methods to classify percussive components in nonnegative matrix factorization (NMF). To circumvent the necessity of designing appropriate spectral and temporal features for component clustering, as usually used in NMF-based transcription systems, multilayer perceptrons and deep belief networks are trained directly on the factorization of a large number of isolated samples of kick and snare drums. The learned features are then used to assign components resulting from the analysis of polyphonic music to the different drum classes and retrieve the temporal activation curves. The evaluation on a set of 145 excerpts of polyphonic music shows that the algorithms can efficiently classify drum components and compare favorably to a classic “bag-of-features” approach using support vector machines and spectral mid-level features.
Click to purchase paper as a non-member or you can login as an AES member to see more options.
No AES members have commented on this paper yet.
To be notified of new comments on this paper you can
subscribe to this RSS feed.
Forum users should login to see additional options.
If you are not yet an AES member and have something important to say about this paper then we urge you to join the AES today and make your voice heard. You can join online today by clicking here.
