基于分段模糊c-均值的连续密度HMM语音识别模型参数估计

The segmental fuzzy c-means algorithm for estimating parameters of continuous density hidden Markov models

摘要: 本文在分析了连续密度隐马尔可夫模型(CDHMM)的两种主要最大似然参数估计方法的基础上,引入模糊集思想,提出了分段模糊c-均值算法代替普通的分段k-均值算法进行CDHMM的最大似然参数估计。文中给出了其实现方法。实验结果证明其在语音识别中具有很好的性能。

Abstract: In this paper, we propose the segmental fuzzy c-means algorithm for maximum likelihood estimating parameters of continuous density hidden Markov models (CDHMM) to substitute for the common segmental k-means algorithm, based on the analysis of two main methods for estimating parameters of CDHMM. Experimental results demonstrate the efficiency of the new algorithm.