基于STRAIGHT谱的非特定人数字语音识别  

Research on STRAIGHT Spectrum in Speech Recognition

作　　者：姚建霄[1] 张歆奕[1] 

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2011年第1期56-60,共5页Journal of Wuyi University(Natural Science Edition)

摘　　要：介绍了STRAIGHT算法的原理,并选择STRAIGTH谱作为语音识别的特征参数.采用对应点映射方法以提高同一发音不同样本参数间的匹配效果,在特征匹配的方法上选择差别子空间法,将二者结合应用于非特定人汉语数字0～9的语音识别,实验结果表明,基于STRAIGHT谱的非特定人数字语音识别可以达到97%的识别率.The principle of the STRAIGHT algorithm was introduced and the STRAGHT spectrum as feature index for speech recognition was chose. Corresponding point mapping was used to improve the effect of matching different sample parameters and the difference subspace was used for feature matching. These methods can be applied to digital identification of non-specific persons. The result shows that the STRAIGHT spectrum-based digital voice recognition can achieve a high recognition rate.

关 键 词：语音识别 STRAIGHT谱 差别子空间 对应点映射 

分 类 号：TN912.34[电子电信—通信与信息系统]