基于因子分析的隐马尔可夫模型及其训练算法

虽然基于对角协方差矩阵高斯分布的隐马尔可夫模型(HiddenMarkovModelBasedonDiagonalGaussiandistributions,HMM-DG)目前在现代大词表连续语音识别系统中得到了广泛的应用,但HMM-DG在帧内特征相关(intra-framefeaturescorrelation)建模方面存在缺陷。该文将因子分析方法与HMM-DG的混合高斯建模相结合,提出了一种具有弹性的帧内特征相关隐马尔可夫模型框架—基于因子分析的隐马尔可夫模型(HiddenMarkovModelBasedonFactorAnalysis,HMM-FA),并导出了HMM-FA的训练算法。仿真实验表明:在相同的条件下,HMM-FA的性能优于HMM-DG。

A Hidden Markov Model Based on Factor Analysis and its Training Algorithm

Currently,the Hidden Markov Model Based on Diagonal Gaussian distributions(HMM-DG)is the most popular and successful model in speech recognition.However,there are well known shortcomings in HMM-DG particularly in the modeling of the correlation among feature-vector elements(intra-frame features correlation).This paper investigates the use of mixture Gaussian models and factor analysis in HMM,proposes a Hidden Markov Model Based on Factor Analysis(HMM-DG)and derive s an expectation-maximization(EM)algorithm for maximum likelihood estimation.The theoretical analysis and simulation shows that the HMM-FA can achieve better performances over HMM-DG with the same amount of training data.