用于语音识别的基于频谱调整的信道自适应方法  

Channel adaptation method based on spectral adjusting for speech recognition

作　　者：赵蕤[1] 王作英[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《清华大学学报（自然科学版）》2005年第4期441-444,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家"八六三"高技术项目(2001AA114071)

摘　　要：语音识别系统在实际应用时,其性能会因各种因素而下降,其中重要的一个因素是信道的不匹配。该文提出了一种新的信道自适应方法——频谱调整法。该方法在频域上定义一个分段线性信道归一化函数,根据最大似然准则利用梯度投影法求其最优参数后,对语音的幅度频谱进行归一化。实验表明,该方法可以利用很少的自适应数据使识别的字错误率下降10%左右。Channel mismatch is one of the main causes of degradation in speech recognition performance. This paper presents a channel adaptation method named spectral adjusting (SA) normalizes the distorted speech spectrum with a piecewise linear normalization function. The function parameters were obtained using the Gradient Projection algorithm based on the maximum likelihood criterion. Tests show that the method is able to reduce the word error rate by about 10% even with very short utterance.

关 键 词：信息处理 语音识别 稳健性 信道自适应 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]