Application Research Of Hyper-Sphere Support Vector Machine In Speech Recognition

With the rapid development of science and technology, the technology of speech recognition has made significant progress and applied to the various fields. Compared to other ways,SVM(Support Sector Machine) method has better generalization ability and classification accuracy, and more suitable for speech recognition.Recently, SVM theory has been continuously improved and developed. A variety of new and improved algorithms are constantly emerging. This paper details the deformation of the SVM—HSSVM(Hyper-Sphere Support Vector Machine). The main work can be concluded as follows:(1) Briefly introduces the basic principles of speech recognition, analyzes of the current advantages and disadvantages of mainstream speech recognition and details the SVM theory. The core idea of SVM is to put the nonlinear transformation of low-dimensional space to a high-dimensional space to make it a linear transformation through the kernel function. And construct an optimal linear classification surface in this high-dimensional feature space.(2) HSSVM has a faster speed and classification results by using hyper-sphere to divide the data. This paper introduces its principles, derivates of the algorithm, and applies it to speech recognition.(3) This paper respectively identifies the data before and after the normalization. The experimental results show that the use of normalized data to recognize is better than that of no normalization results. This paper constructs three types of support vector machines non-specific speech recognition system based on "lvl", "lvR" and "HSSVM" to study the influence on recognition speed and recognition rate of different multi-classification and the same kernel function, kernel parameters and the penalty factor. The experimental results show that the run speed and recognition rate of HSSVM in speech recognition is better than that of "lvl" and "lvR" methods.