本文是对HMM最大距离训练方法的一种改进 ,该方法采用了更合理的模型距离定义 ,能更有效地利用训练数据集中的区别信息 ,使有限的训练数据得到更好的应用 ,达到提高语音识别系统性能的目的 .导出了HMM模型参数的迭代公式 .基于TIMIT数据库的非连续语音及连续语音实验结果表明 ,改进训练方法在降低错识率上较原来的方法有明显改 This paper is an improvement of the HMM maximum distance training method, which adopts a more reasonable model distance definition, can make more effective use of the difference information in the training data set, make the limited training data be better applied, And identify the performance of the system.Experimental results of HMM model parameters are derived.The experimental results of discontinuous speech and continuous speech based on TIMIT database show that the improved training method has obvious improvement in reducing the error rate compared with the original method