摘要
本文从LPC系数的统计特性出发,基于线性均方误差估值的思想,提出了一种用于语音识别的新的LPC距离量度——LMSE距离.在一个IBM-PC机语音处理系统和一个VAX-11/750机语音处理系统上,做了大量孤立词识别实俭,证实了这种距离的可行性.文中还将它与著名的Itakura距离进行了比较.
According to the statistical properties of LPC coefficients a new LPC distance measure, the LMSE distance, used in speech recognition and based on the idea of the linear mean square error estimation is proposed. Experimental results obtained on an IBM-PC speech processing system and a VAX-11/750 system are reported. It is shown that the performance of the distance proposed is good. The LMSE distance has also been compared with the famous Itakura distance.
出处
《华中理工大学学报》
CSCD
北大核心
1990年第4期67-70,共4页
Journal of Huazhong University of Science and Technology
关键词
语音编码
距离量度
线性均方误差
Speech recognition
Speech analysis
Distance measure
Linear mean square error estimation
