HMM识别孤立词的研究与实现

STUDY AND IMPLEMENTATION OF ISOLATED WORD RECOGNITION USING HMM

本文介绍了隐式 Markov 模型(简称 HMM)识别语音的基本原理,对在训练孤立词模型过程中采用的 Baum-Welch 算法和 Viterbi 算法进行了研究,导出了参数估计的整套算式,提出了解决 HMM 在计算机上实现时出现的问题的方法及其实现算式。作者将 HMM 应用于汉语数字的识别,进行了不同算法的比较和不同初值条件的试验,给出了相应的识别结果。

This paper describes the general principle of Hidden Markov Model for isolated wordrecognition,and also gives a series of formulas estimating parameters in the training ofMarkov model for isolated word,based upon the discussion of Baum-welch algorithm andViterbi algorithm.In accordance with the problems in the implementation of HMM,themethods of solution are proposed.HMM is successfully applied to Chinese digit recognition,and experimental results of thecomparison between two algorithms and different initial values are presented in the paper.