Research and realization of speaker recognition based on embedded system

Speaker recognition is the process of automatically recognizing the speaker by distilling the speaker´s information from phonic signals. Compared with the other biometrics technologies, speaker recognition is more convenient, more accurate and more economical. Speaker recognition system is presented which uses linear prediction cepstrum coefficient (LPCC) as feature parameters, and devises an SVM classify system. Support vector machine (SVM) is a kind of novel machine learning method proposed by Vapnik, which have proven to be a powerful technique for pattern classification. SVM map inputs into a high-dimensional space and then separate classes with a hyperplane and devise nonlinear and high dimensional sample problems. The source code is optimized to be transformed into embedded devices. The selection of S3C2410A assures real-time running of our system. Experimental results show that SVM is suit speaker recognition based on embedded system, which provides an important method for its practical purposes.