基于MDI与进化计算相结合的HMM训练方法

MDI为HMM训练的优化准则之一，但传统的MDI是基于局部最优求解的，所得的解也是一个局部最优解，而进化计算则是基于全局搜索的。为此，提出了将MDI及进化计算相结合来训练HMM的方法。各个模型用个体来表示，个体的适应值采用模型的最小差别信息。实验结果表明，该方法所得的系统识别率高于传统的方法。

HMM training method based on MDI and evolutionary computation

Minimum discrimination information (MDI) is one of the optimization criteria for HMM training. With traditional MDI training method, it can only gain a local optimum solution for it is based on local search, but evolutionary computation is based on global search. Hence, a new training method is proposed based on evolutionary computation and MDI. Each individual in evolutionary computation represents a HMM, while the fitness value of each individual represents the minimum discrimination information. The experimental results indicate that the system's recognition rate trained with the proposed method is superior to the one trained with traditional training method.