机构地区: 华中理工大学电子与信息工程系
出 处: 《华中理工大学学报》 1992年第1期47-51,共5页
摘 要: 本文提出了利用语音短时特征参数和相应距离量度公式来构造隐Markov模型(HMM)观察值概密函数的一种方法.文中以语音倒谱系数特征及其相应Euclidean距离为例,给出了概密函数表达式,导出了其识别和训练算法,讨论了算法实现的具体问题.实验证实了本文提出的构造HMM观察值概密函数方法是有效的. The short-time characteristic parameters and the corresponding distance measure formula of speech have been used to construct the density function for observed values of the Hidden Markov Model (HMM). An expression for the density function is given with the coefficients of the inverse-spectrum of speech as well as the corresponding Euclidean distance as examples. The recognition and training algorithms are derived and actual implementation of the algorithm is discussed. The feasibility of the method presented is proved by experiment.