Study On Speaker Recognition Based On Improved Grid Search Parameters Optimization Algorithm Of SVM

As one biological certification technology, speaker recognition has the advantages of lowcost, high security and convenience and has been attracted people’s attention. In allalgorithms about speaker recognition, Support Vector Machine (SVM) algorithm is suitablefor solving the problem of small sample classification and has been one of the hottest researchobjects. In this paper, aiming at the problem that SVM training data could not be too big; thecharacteristic dimension reduction algorithm of Principle Component Analysis (PCA) ismentioned. The algorithm can reduce the number of training data and improve the efficiencyof SVM on the condition of decreasing the effect on the recognition rate of system as far aspossible. What’s more, aiming at the problem that the time of searching the optimal vector inParameters optimization algorithm of SVM is too long, an improved algorithm is proposed inthis paper. The algorithm can decrease the time of parameters optimization on the conditionof ensuring the system recognition rate.In this paper, firstly, the system structure of speaker recognition is analyzed. And thenthe system is separated to the two parts of feature extraction and model matching and studiedbased on the function of system modules. The part of feature extraction is studied on reducingthe dimension of features. After analyzing the advantages of PCA in characteristic dimensionreduction from theory, the efficiency of the algorithm is proved with simulation analysis. Thepart of model matching is studied on Parameters optimization algorithm. And then the basicprinciple of Parameters Optimization Algorithm is discussed in detail. What’s more, theperformance of Parameters Optimization Algorithms is tested and the effect of ParametersOptimization is analyzed. At last, aiming at the instantaneity of the existing algorithms, oneimproved Parameters Optimization Algorithm is proposed. And then the speaker recognitionsystem based on the improved Parameters Optimization Algorithm of SVM in this paper isproposed. Though the simulation analysis, the efficiency of the proposed system is proved.In this paper, though the method of PCA, characteristic dimension is reduced and thecomplexity of system is decreased. From the two views of theory and simulation, based onanalyzing the performance of traditional SVM in Parameters Optimization Algorithm, oneimproved algorithm is proposed to decreasing the time of Parameters Optimization in SVM. Finally, the speaker recognition system based on PCA and improved Parameters OptimizationAlgorithm is analyzed in simulation and the efficiency of the system is proved successfully.