状态输出概率密度为对角协方差矩阵高斯分布的隐马尔可夫模型(HMM-DG)在帧内特征相关建模方面存在缺陷.本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(HMM-FA),并导出了HMM-FA的训练算法.理论分析和仿真实验都表明:在训练数据相同的条件下,HMM-FA的性能优于HMM-DG. DGM (Gaussian Dispersive Hidden Markov Model) with state output probability density is defective in intra-frame feature correlation modeling.This paper combines factor analysis with HMM-Gaussian mixture modeling , This paper proposes a flexible framework of intra-feature-dependent Hidden Markov Model - Hidden Markov Model (HMM-FA) based on factor analysis and derives the training algorithm of HMM-FA. The theoretical analysis and simulation experiments Both show that HMM-FA performs better than HMM-DG under the same training data.