Abstract: Support vector data description (SVDD) defines multi-class data by their respective hyper-spheres. The computational complexity of the quadratic programming problem is reduced significantly and it is easier to solve multi-class classification problems. Thus, SVDD has attracted more and more attention in the field of speech recognition research. For the problems of the feature vectors of speech samples overlapping and updating, the conventional SVDD for multi-class classification was improved. On the one hand, the spatial position of the samples was fully used to construct the decision function in overlapping domain of hyper-spheres; On the other hand, based on class incremental learning the dynamic change of support vectors was implemented. Simulation experimental results indicate that the proposed method reduces modeling time obviously and has better recognition performance.

Key words: support vector data description, multi-class classification, decision function, incremental learning, speech recognition system simulation