摘要
提高语音识别系统的鲁棒性是语音识别技术一个重要的研究课题。语音识别系统往往由于训练环境下的数据和识别环境下的数据不匹配造成系统的识别性能下降。为了能得到无噪音的语音识别特性,让语音识别系统在含噪的环境下获得令人满意的工作性能,根据人听觉特性提出了一种鲁棒语音特征提取方法。将小波变换和MFCC算法相结合,在MFCC的前端用小波包变换代替FFT和Mel滤波器组,同时在后端用临界小波变换代替DCT,最后得到鲁棒语音特征。通过实验结果分析表明,将方法用于抗噪声分析可以提高系统的抗噪声能力;同时特征的处理方法对不同噪声有很好的适应性。
Improving the robustness of speech recognition system is an important issue in speech recognition technology.The performance of traditional speech recognition system degrades seriously when the training environments and the testing environments are mismatched.In order to acquire satisfactory performance of speech recognition system under noisy environment,in this essay,a new robust speech feature extraction method based on properties of the human auditory system is presented,which is obtained by using wavelet transform to replace the FFT and Mel filter banks and using critical band wavelet transform to replace the DCT.The result show that the performance of speech recognition system can be improved greatly by using the new method under noise environment and the proposed method is highly applicable.
出处
《计算机仿真》
CSCD
北大核心
2010年第8期355-358,362,共5页
Computer Simulation
