论文部分内容阅读
本文提出两种基于最小误识率准则的区分训练方法，将目标函数直接建立在Ｖｉｔｅｒｂｉ解码的基础上，以保证所得的隐马尔可夫模型（ＨＭＭ）参数能提高识别率。
This paper presents two discriminative training methods based on the minimum error rate criterion. The objective function is directly based on Viterbi decoding to ensure that the resulting HMM parameters can improve the recognition rate.