@Wayne-Z 2017-11-17T21:25:27.000000Z 字数 2301 阅读 2455

韵母声纹的归一化图示

语音处理

韵母声纹的归一化图示
- 韵母
- 总结

本文对韵母进行了另一次梳理，将分析的频率主要集中在了0-2500HZ的区间，同时对程序解析的窗口进行了二分，使得处理更加精细。
代码如下：

fs=44000; %语音信号采样频率为44100
file = 'a-1.wav';
wavelength = length(audioread(file));
term = floor(wavelength/1100);
samples = [1, 2200];
T = 1:term;
P = zeros(2048,term);
for id = 1:term
    x = audioread(file,samples);
    t = (0:length(x)-1)/fs;
    y = fft(x,4096);           %对信号做2048点FFT变换
    f = fs*(0:2047)/4096;
    p= abs(y(1:2048));
    [pks,locs,widths,proms] = findpeaks(p,'SortStr','descend');
    s = widths.*pks;
    idx = 1;
    Sum =s(1);
    %P(locs(1),id) = s(1);
    for j = 2 : length(s)
        if s(j)/Sum > 0.10
            Sum = Sum + s(j);
            idx = [idx , j];
            %P(locs(j),id) = s(j); 
        end 
    end
%     for k = 1 : 2048 
%         if any(k == idx)
%             P(k,id) = p(k);
%         end
%     end
% 用于显示单纯的振幅峰值
% 筛选出0-1000,1000-2000,2000-3000的频率
ids = [];idm = [];idl = [];
    for i = 1:length(idx)
        if f(idx) <=1000
            ids = [ids,idx(i)];
        elseif  f(idx) <= 2000;
                idm = [idm,idx(i)];
        else
            idl = [idl, idx(i)];
        end
    end
%将各个区间内的多条化为一条
    As = sum(s(ids));
    Am = sum(s(idm));
    Al = sum(s(idl));
    Fs = floor(dot( s(ids),locs(ids))/As);
    Fm = floor(dot( s(idm),locs(idm))/Am);
    Fl = floor(dot( s(idl),locs(idl))/Al);
    Ps = dot(s(ids),pks(ids))/As;
    Pm = dot(s(idm),pks(idm))/Am;
    Pl = dot(s(idl),pks(idl))/Al;
    if ~isnan(Fs)
        P(Fs,id) = Ps;
    end
     if ~isnan(Fm)
        P(Fm,id) = Pm;
     end
     if ~isnan(Fl)
        P(Fl,id) = Pl;
     end
    if 1100*(id+2) <= wavelength
        samples = [1100*id, 1100*(id+2)]; 
    else 
        samples = [1100*id,wavelength];
    end
end
surf(T,f,P,'edgecolor','none'); axis tight; 
view(0,90);  % 通常看时谱图时都采用这个角度，若需要可以采用
set(gca, 'YLim',[0 3000])
xlabel('Time (0.025s)'); ylabel('Hz');