Speaker discrimination using several classifiers and a relativistic speaker characterization

Abstract

Automatic Speaker Discrimination consists in checking whether two speech signals belong to the same speaker or not. It is often difficult to decide what could be the best classifier to use in some specific circumstances. That is why, we implemented nine different classifiers, namely: Linear Discriminant Analysis, Adaboost, Support Vector Machines, Multi-Layer Perceptron, Linear Regression, Generalized Linear Model, Self Organizing Map, Second Order Statistical Measures and Gaussian Mixture Models. Moreover, a special feature reduction was proposed, which we called Relativistic Speaker Characteristic (RSC). On the other hand we further intensified the feature reduction by adding a second step of feature transformation using a Principal Component Analysis (PCA). Experiments of speaker discrimination are conducted on Hub4 Broadcast-News. Results show that the best classifier is the SVM and that the proposed feature reduction association (RSC-PCA) is extremely efficient in automatic speaker discrimination.