研究了用于CHMM的MCE学习方法 .提出了简单形式的识别函数、误差函数、损失函数等的定义方法以及总损失函数的最小化方法 .通过对非特定人汉语连续语音识别实验 ,证明了提出的函数形式和损失函数最小化方法对CHMM的MCE学习的有效性 This paper studies the MCE learning method used in CHMM and proposes a simple form of the definition of the identification function, the error function, the loss function, and the method of minimization of the total loss function.Through the experiment of non-specific Chinese continuous speech recognition, The functional form and the loss function minimization method to the effectiveness of CHMM’s MCE learning