基于HMM算法的语音识别系统设计及其混合编程实现

Speech Recognition Study Based on MFCC and CDHMM and Its Realization Using a Mixed Program Method

绍了连续元高斯混合密度的隐马尔可夫语音识别系统,提出了一种基于分段平均法的初始化模型参数方法;将可靠性度量方MK法结合到重估算法中去,简化了算法复杂度。系统的实现采用与混合编程,提高了整个系统的开发效率。

This paper introduces the theory and implementation of CDHMM speech recognition. It provides an improving K-means initialization method based on character of speech signal. The technique of reliability measure is used to resolve the problem of parameters combination, which can considerably reduce the calculation burden. It can speed up the whole system development by using a mixed program mode with C++ and Matlab. ;;