基于多相关分组的HMM训练算法

A HMM training algorithm based on grouping multiple observations by multiple correlation coefficient

在用多观察序列训练HMM理论的基础上,提出了一种基于对多观察序列按多相关系数分组的HMM训练算法(简称基于多相关分组的HMM训练算法).该算法避免了直接计算条件概率的困难,与传统的Baum-Welch算法相比,既考虑了训练序列之间的相关性,又不增加计算量.

This paper proposes a HMM training algorithm which is based on grouping multiple observations by multiple correlation coefficient .It avoids computing the conditional probabilities directly and considers the correlativity between successive observation vectors and need not increase computation works any more compared with traditional BaumWelch algorithm.