Digital speech recognition is an extremely important branch of speech recognition. Its application in real life is used more and more widely. HTK is a C language-based toolkit developed by CUED mainly used for speech signal reorganization, speech synthesis, character reorganization, DNA compositor and so on. From HTK's general principles and software architecture, this paper designs a digital speech recognition system based on HTK, and verifies its recognition efficiency. Then by changing the identification unit and MFCC dimension, and by increasing the number of gaussian mixture components, it considers effects of different factors on the performance of the system. Finally, through the comparing test, it verifies the right combination of the identification unit and the number of gaussian mixture components, and also proves that MFCC dimension can enhance the system's correct rate.