The conversion of Pinyin into Chinese characters is an important aspect of Chinese information processing. This paper gives improvement on cache-based natural language algorithms on the basis of predecessors, with two formulas applied to two different occasions to meet different requirements. Also it shows how to achieve such a conversion by using hidden Markov model (HMM) and Viterbi algorithm is used to solve the model. This article gives a detailed way about the design. At the end we randomly choose the corpus to train on the computer and give test results with the analysis of the wrong reasons.