mfcc_fft (1).m

function Fn = mfcc_fft(signal)
%     windowSize = 10;
    An = NewPowerSpectrum(signal);
%     disp(An);
    [K,n] = size(An);
    
%     %分帧
%     windowStep = samplingRate/frameRate; % 采样率除以帧数 --> 一帧包含的采样数
%     n = fix((length(input)-windowSize)/windowStep);% fix朝零方向取整,n帧
    
    %a set of 13 coefficients is calculated as follows:
    Fn = zeros(13,n);
    for l = 1:13
        for k = 1:K
            Fn(1,:) = sum(An(k,:))/sqrt(K);
            Fn(l,:) = An(k,:)*cos(l*(2*k+1)*pi/(2*K))/sqrt(K/2);
        end
    end
    
%     disp(Fn);
    
%     mfccDCTMatrix = 1/sqrt(totalFilters/2)*cos((0:(cepstralCoefficients-1))' * ...
%                     (2*(0:(totalFilters-1))+1) * pi/2/totalFilters);
%     mfccDCTMatrix(1,:) = mfccDCTMatrix(1,:) * sqrt(2)/2;
%     ceps = zeros(13, n);% 生成13 x n的全零矩阵 准备
%     for start=1:13 % 对每列信号
%         first = start*windowStep + 1;
%         last = first + windowSize-1;
%         fftData = zeros(1,fftSize); %初始化列向量
%         fftData(1:windowSize) = preEmphasized(first:last).*hamWindow; % 预加重这列信号并加窗
%         fftMag = abs(fft(fftData)); % 每个都做FFT然后取模（列向量）
%         earMag = log10(mfccFilterWeights * fftMag'); % 三角滤波和对数运算
% 
%         ceps(:,start+1) = Fn * earMag; % 做DCT得到MFCC
%     end
end