A text-dependent voice lock is designed. The Mel frequency cepstrum coefficient（MFCC） and its firstorder difference is extracted as the features of voice. The speaker’s voice feature model is trained by using vector quantization algorithm. Identification effect of system is analyzed

under different number of code in the codebook

different SNR of speech sample combination via design experiments. The results demonstrate that the identification accuracy can reach above 95 %. while code number N is 48

under certain SNR. The algorithm can run normally after transplanted to embedded platform

can meet requirement of engineering practicality.