一种改进的基于LP倒谱特征的孤立词语音识别方法  

A Speech Recognition Method of Isolated Words Based on Modified LP Cepstrum

作　　者：侯雪梅[1] 张雪英[1] 赵高峰[1] 

机构地区：[1]太原理工大学信息工程学院,山西太原030024

出　　处：《太原理工大学学报》2006年第5期508-510,共3页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金资助项目(60472094);山西省留学回国人员科研基金资助项目(200224)

摘　　要：阐述了Mel谱失真测度的概念,指出用Mel频率尺度可较充分地反映人耳对频率及幅度的非线性感知特性。在此基础上,针对孤立词语音识别,对常规LP倒谱特征提取方法进行改进,即将LP倒谱按符合人耳听觉特性的Mel尺度进行非线性变化,得到LP Mel倒谱系数(LPM-CC)作为特征参数。识别网络使用RBF神经网络,进行了孤立词语音识别。实验结果表明此种方法抗噪性能好,识别效率高。The measurement of Mel spectrum distortion is a kind of warped frequency spectrum distortion measure. Using Mel frequency scale can reflect sufficiently the nonlinear perceptive characteristics of human hearings to frequency and amplitude. Aiming at speech recognition of isolated words, an improved algorithm for normal LP cpestrum feature is put forward in this paper. That is to say, LP cpestrum is made nonlinear changes by means of Mel scale according to auditory characteristic, and the LP Mel cepstrum coefficient （LPMCC） is used as feature parameexperiment shows that this method is good for robustness and effective on recognition

关 键 词：语音识别 LPMCC RBF 

分 类 号：TN912.34[电子电信—通信与信息系统]