采用连续混合密度隐马尔柯夫模型的语音识别

SPEECH RECOGNITION USING CONTINUOUS MIXTURE DENSITY HIDDEN MARKOV MODELS

介绍了采用连续Ｍ元高斯混合密度的隐式马尔柯夫模型的语音识别实验系统，对语音特征矢量用非线性归一化算法进行预处理．另外，还提出一种基于语音知识的模型初始化方法．实验证明：语音的预处理和模型初始化方法都取得了较明显的效果．

This paper discusses a speech recognition system using the continuous mixture density Hidden Markov Model. We use a nonlinear programming method to proceed time warping about feature vector. Moreover, we present a method based on speech knowledge to initialize model.