Self Organizing Maps for Reducing the Number of Clusters by One on Simplex Subspaces

This paper deals with N-dimensional patterns that are represented as points on the (N - 1)-dimensional simplex. The elements of such patterns could be the posterior class probabilities for N classes, given a feature vector derived by the Bayes classifier for example. Such patterns form N clusters on the (N - 1)-dimensional simplex. We are interested in reducing the number of clusters to N - 1 in order to redistribute the features assigned to a particular class in the N - 1 simplex over the remaining N - 1 classes in an optimal manner by using a self-organizing map. An application of the proposed solution to the re-assignment of emotional speech features classified as neutral into the emotional states of anger, happiness, surprise, and sadness on the Danish emotional speech database is presented