Abstract
We consider the extension of the greedy adaptive dictionary learning algorithm that we introduced previously, to applications other than speech signals. The algorithm learns a dictionary of sparse atoms, while yielding a sparse representation for the speech signals. We investigate its behavior in the analysis of music signals, and propose a different dictionary learning approach that can be applied to large data sets. This facilitates the application of the algorithm to problems that generate large amounts of data, such as multimedia of multi-channel application areas.