中文说明：具体步骤： (Ι)利用汉明窗对腭裂语音信号进行分帧，得到分帧后的语音信号x(n)。 (Ⅱ)将加窗分帧后的语音信号进行离散傅里叶变换。因FFT变换点数过大会增大运算复杂度，过小会降低频谱分辨率，故本文选取882点。即得到每帧语音信号的线性频谱(Ⅲ)将上面得到的线性频谱通过32个具有三角形滤波特性的Mel频谱滤波器组，得到Mel频谱 S(l) (Ⅳ)对滤波器组的输出依次做对数变换、离散余弦变换，最终得到MFCC系数
English Description：
The specific steps are as follows: (1) divide the speech signal of cleft palate into frames by using Hamming window to get the speech signal x (n). (II) the speech signal after windowing and framing is transformed by discrete Fourier transform. Because the number of FFT transform points is too large to increase the computational complexity, too small will reduce the spectral resolution, so 882 points are selected in this paper. That is to get the linear spectrum (III) of each frame speech signal, pass the above linear spectrum through 32 Mel spectrum filter banks with triangular filtering characteristics to get Mel spectrum s (L) (IV), do logarithmic transform and discrete cosine transform on the output of filter banks in turn, and finally get MFCC coefficients