Abstract: Recognition accuracy and adaptability is two of the most important capacity of a speech recognition system. A general method to improve speech recognition accuracy often focuses on improving acoustic model to obtain a higher recognition accuracy, which often leads to a complex model and difficult model training. In addition, in the case of unfixed position of speaker to microphone, its performance is often very bad. A new speech recognition method is brought forward,which uses multi\|microphone to recognize Chinese syllables separately, and data fusion technique to process the recognition results. The preliminary experiment shows it not only can improve the system adaptability to environment, but also can reach a higher system recognition accuracy with a lower accuracy of each acoustic model.