文献类型：会议论文

英文题名：Research and Analysis of HMM Speech Recognition Model

作者：Zhu Shu-qin[1];Wei Shao-qian[1];Zhang Yin-xia[1]

第一作者：朱淑琴

通讯作者：Zhu, SQ[1]

机构：[1]Beijing Union Univ, Normal Coll, Beijing 100011, Peoples R China

第一机构：北京联合大学师范学院

通讯机构：[1]corresponding author), Beijing Union Univ, Normal Coll, Beijing 100011, Peoples R China.|[1141711]北京联合大学师范学院;[11417]北京联合大学;

会议论文集：International Conference on Artificial Intelligence and Soft Computing (ICAISC 2012)

会议日期：MAR 15-16, 2012

会议地点：Melbourne, AUSTRALIA

语种：英文

外文关键词：speech recognition; hidden Markov models; initial parameters estimation; frame relevance

摘要：Hidden Markov model is a kind of probabilistic model describing statistical property of the stochastic process, which obtained the quite widespread use in the speech recognition. The inappropriate choice of HMM initial parameters possibly causes model parameter restraining in the partial optimal solution or increasing the computation load. The research and analysis of speech recognition process based on the HMM algorithm is carried on in this paper. Initial parameters estimation of HMM train is especially discussed. When partitioning the characteristic parameters sequence, frame correlation of speech signal is considered. A large number of experiments about different initial parameters had been made to show that the method in this paper has a good result.