摘要
将类关联特征(class-dependentfeature,CDF)用于隐马尔可夫模型(hiddenMarkovmodel,HMM)的建模,提出了一种新的HMM训练算法,与传统的HMM训练算法在理论上完全一致,但新算法避免了直接估计高维的状态输出概率密度函数(probabilitydensityfunction,PDF),可提高模型参数的估计精度。
Using the class-dependent feature (CDF) into the modeling problem of hidden Markov model(HMM), this paper presents a novel HMM training algorithm and demonstrates that the new algorithm is the same in theory as traditional algorithm, but without the necessity of estimating high-dimensional probability density function.
出处
《孝感学院学报》
2004年第3期74-77,共4页
JOURNAL OF XIAOGAN UNIVERSITY
基金
湖北省教育厅重点项目(2002A02004)
关键词
类关联特征
隐马尔可夫模型
训练算法
class-dependent feature
hidden Markov model
training algorithm
