Abstract: In the process of constructing the voiceprint recognition system, an important way to improve the recognition rate is to make the features extracted from the speech signal contain as many speaker’s personality characteristics as possible. To explore the characteristic parameters of different orders that affect the performance of recognition system, based on GMM-UBM baseline system in this paper, the recognition rates of the feature combinations of various dimensions in noiseless environment are studied. By means of increment-subtraction component method, the relative contribution degrees of various dimensions of feature parameters to the recognition rate are calculated quantitatively. According to the contribution degrees, various dimensions of feature parameters are weighted reasonably, and the weight coefficients based on fitting contribution are obtained, which can be used to improve the Mel frequency cepstrum coefficient (MFCC). The simulation experimental results show that the accuracy of voice print recognition is increased by feature parameters with improved weight coefficients.