Arabic speech clustering using a new algorithm

Summary form only given. We present the implementation and the test of an efficient clustering algorithm for speech clustering in an unsupervised manner to cluster given acoustic vectors. The proposed algorithm is based on regulating a similarity measure and replacing movable vectors so that it makes it possible to mitigate the principal defects raised by the traditional methods (K-Means, dynamic clouds, Fuzzy C-Means, ...) which are needed for knowing a priori, the number of clusters, sensitivity to the choice of the initial configuration, and to choose a suitable similarity measure according to the shapes of the data. We present the results obtained on a personal basis made up of isolated digits pronounced in Arabic language. The obtained results and the comparison with the Fuzzy C-Means algorithm (FCM), one of the most powerful algorithms in the literature, give very promising results.