PERFORMANCE OF CLUSTERING METHODS OF SPEECH BASED IDENTIFICATION SYSTEM: RLS, MODIFIED RLS, AND FUZZY CLUSTERING
Nurul Hidayat · 2010
In this paper, we discuss about the performance of three clustering methods of speech based object identification system. The three methods are Randomised Local Search (RLS), Modified Randomised Local Search (M-RLS), and Fuzzy Clustering. The data used in clustering is a magnitude of speech signal, which is the result of a pre-processing using Trispectrum Estimation Method. The performance of the three clustering methods are measured based on an experiment to recognise a word spoken by one person. It follows from the result that the performances of the three clustering methods depend on the number of references, while it is not the case for the number of cluster. The more the number of references used in learning process, the high the performance of the system, but the higher the computation cost. In general, the M-RLS method is better the other two.