Support Vector Machine Applications In Speaker Recognition

The speaker recognition system has a better prospect and can be applied widely, but, at present, almost all the speaker recognition systems are text-dependent. Although there pour out some text-independent speaker recognition systems, most of them have critical restrictions on the testing content. That is to say, they are not really text-independent. So, this paper aims to find a method to realize text-independent speaker recognition systems. At first, we use hidden Markov Model to denoise. The wavelet analysis is a useful tool in the research of temporal and frequency domains, especially in the high frequency conditions. It has been widely and successfully in audio processing and feature extraction because of its updating localization. We have created Wavelet Neural Network for pattern recognition on the combination of wavelet analysis and BP Neural Network.. The support vector machine (SVM) developing on the foundation of statistic learning theory (SLT) is known as a proper theory or tool and there are small examples, such as data containing noise. In this paper, we use the SVM as one of the classifications, discuss some properties of the SVM, and also impose a neural approach of combining multi-classifications.