摘要
将最大互信息理论用于语音识别，最大互信息估计作为目标函数。在隐马尔可夫模型参数调整过程中运用了泛化概率下降方法，保证了统计意义上实现目标函数的优化。最大互信息估计用于连接数字语音识别，识别率得到了提高。
The theory of maximum mutual information is applied into speech recognition , maxi- mum mutual information estimation is used as object function . General probability descent method is used to ensure the optimization of object function in statistics during the process of HMM pa- rameter modification. The recognition rate is increased when maximum mutual information esti- mation is applied into connected digit speech recognition .
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
1998年第3期82-85,共4页
Journal of Beijing University of Posts and Telecommunications
