Novel approach in speaker identification using support vector machines

This paper presents a novel approach on speaker identification using support vector machines (SVMs). To improve the performance of the identification, an extra training set is applied to train a discrete density hidden markov model (HMM). In testing session, first, the multi-class-SVM classifies each feature vector. Then, the HMM model is applied to make a decision with the classes sequence. HMM-based technique outperforms the conventional methods, especially when there are not enough training or testing data. While the proposed method doesnpsilat induce much computational complexities, it reduces the identification error rates up to 57.14%.