基于Bark子波变换的MFCC特征提取

为提高低信噪比环境下语音的鲁棒性，提出一种改进的Mel频率倒谱系数(MFCC)特征提取方法。在传统MFCC特征提取的基础上，引入更适应人耳听觉系统的Bark子波变换，在快速傅里叶变换之前对语音进行预处理，并在MFCC提取方法中代替离散余弦变换；在语音预处理阶段，利用改进的Lanczos窗函数抑制旁瓣以提高语音鲁棒性。实验表明，与传统MFCC方法相比，在噪声环境下，改进方法具有更高的说话人识别率。

MFCC Feature Extraction Based on Bark Wavelet Transform

In order to improve the quality of speech in low Signal Noise Ratio（SNR）,an improved Mel Frequency Cepstral Coefficient（MFCC） feature extraction method is proposed.On the basis of the traditional MFCC feature extraction,the improved method introduces Bark Wavelet Transform（BWT） for more suitable to human ear＇s auditory system,it is used to make preprocessing before Fast Fourier Transform（FFT）,on the other hand,it is used to instead of Discrete Cosine Transform（DCT） in MFCC.In the pre-processing stage Lanczos window function is adopted to restrain the side lobe and to improve the robustness.Experimental results show that compared with the traditional MFCC,the improved method can improve the speaker identification accuracy in the noisy environment.