一种基于MVDR和CCBC的抗噪语音识别方法

提出了一种适用于抗噪声语音识别的方法，其特征提取过程基于最小方差无失真响应（Minimum variance distortionles sresponse，MVDR）谱估计方法，并对该特征进行频率弯折以提高其知觉分辨率，最后使用基于正则相关分析的谱变换补偿（Canonical correlation based on compensation，CCBC）法对该特征进行自适应处理，从而提高了系统的鲁棒性。在展览馆噪声、人群噪声和汽车噪声下，与基于传统Mel倒谱系数（MFCC）特征的系统进行了对比实验，结果表明使用本文方法的语音识别系统的识别率得到了显著的提高。

Robust Speech Recognition Based on MVDR and CCBC Approach

A new approach for the robust speech recognition is proposed.The approach employs minimum variance distortionless response(MVDR) spectrum estimation method to extract features and warps the spectrum on a frequency scale to improve the perceptual resolution.Then,the approach makes adaptive processing to the features by canonical correlation based on compensation(CCBC) method,and improves the robustness of the features.Compared with the system based on the traditional Mel frequency cepstrum coefficients(MFCC) feature,experimental results in the noise environment of the exhibition hall,the crowd and the car indicate that the recognition accuracy of the speech recognition system based on the new method is remarkably increased.