摘要： 给出了两种离散HMM参数估计，一种是首先估计HMM的转移概率和直接估计离散符号集的分布，然后假设各状态都是正态分布，再从符号集的离散概率分布，来估计正态概率分布的参数（GMM）；另一种是将离散符号设定为正态分布，用与连续HMM相似的方法，来估计离散HMM正态概率分布的参数。

关键词: 离散HMM, 分布估计, 参数估计, 高斯混合模型, HTK

Abstract: Two methods of parameter estimation of discrete observation in HMM are proposed in this paper. One estimates the transition probability and the distribution of discrete symbol set at first. Each state is assumed to be from a Gaussian distribution. Then it does both discrete HMM probability distribution estimation and discrete GMM parameter estimation based on symbol. The other is that all discrete symbols are assumed to be from a Gaussian distribution. Then it does the parameter estimation of discrete observation in HMM, which is similar to parameter estimation of continuous HMM.

中图分类号: 

• TP391.41