With the development of computer technology and voice technology,inter- communtion with computer by using voice have been a symbol of modern scientific achievements, that also made speech recognition and speech synthesization a research direction of computer technology. Speech recognition technology is concerned with many research fields of varied subject ,they have contributed a lot to the development of speech recongnition. These characters include levity, dynamic, instantaneousness, continuum and so on .The process of computer recognize speech are essentially the same as that of human’s. Now the mainsteam of Speech recognition technology is base on the basic theory of statistic mode recongnition. The purpose of this article is to realize a isolated word recongnition system,introduce the arithmetic of hing-rated speech recognition and some principal assumption of improve it … This article introduce the basic conception, common method and character of isolated word speech recognition , and particularly discuss the procedure of gain eigenvector, the principle of HMM, the completion of Baum-Welch arithmetic and the virtue of HMM then improve the recognition speed, reduce operation complexity and satisfy the real-time requirement by using Viterbi arithmetic and Segmental K-means arithmetic . At the end the author expatiate some shortages of this system and some idea about improving them. Keywords Speech recognition Hidden markov models Segmental K-means arithmetic Isolated word MATLAB Emulation