The dimensions are higher after combining MFCC with LPCC, which affect the performance of recognition. To solve the problem, genetic algorithm parameters for initial feature dimension reduction, to improve recognition performance. Firstly, we extract MFCC and LPCC of the speech signal. Then we reduce the dimensionality based on genetic algorithm. Finally, recognition is presented based on SVM with the low dimensional data. Simulation results show that compared to traditional PCA, the genetic algorithm recognition can increase the recognition rate by 12.2%. Meanwhile, recognition rate was reduced by 1.23% compared with the initial feature,but the recognition time increased 4.5 times.