基于mfcc，vq的说话人识别代码实现_自己写的

%测试代码，以两次为例进行测试 clc,clear,close all,warning off,feature jit off tic disp('Calculating reference template parameters~~~') % Rt = 0; %准确率参数 formatSpec1 = 'H:\\DSP\\sound2\\a_%d\\strawberry_%d.wav'; formatSpec2 = 'H:\\DSP\\sound2\\strawberry%d.wav'; % global constlen global Num count = 0; %计算匹配成功率用 Num = 17; FrameInc = 80; %帧移为80 for i = 1:Num ref(i).mfcc = Fofeveryone(i); end %第一个人第一次的语音长度 % fname1 = sprintf(formatSpec1,1,1); % [x,Fs] = audioread(fname1); % Fs = 48000 % x = x(:,1); % [x1, x2] = vad(x); % count = x2-x1; % constlen = sum(count(:)); %所有元素个数 % y = []; % for i=1:length(x1) % y = [y; x(x1(i)*FrameInc +1 :x2(i)*FrameInc)]; % end % m = mfcc(y); % ref(1).mfcc = m; % %第一个人后9次的语音 % for Qt=2:10 % fname1 = sprintf(formatSpec1,1,Qt); % x = audioread(fname1); % x = x(:,1); % %figure() % [x1, x2] = vad(x); % count = x2-x1; % count1 = sum(count(:)); % y = []; % for i=1:length(x1) % y = [y; x(x1(i)*FrameInc+1:x2(i)*FrameInc)]; %存在1的差距 % end % x = resample(y,constlen,count1); % 除法运算存在误差（不能除尽的bug） % if length(x) > constlen*FrameInc %消除误差 % x(length(x))=[]; % end % if length(x) < constlen*FrameInc % x(length(x)+1)=x(length(x)); % end % m = mfcc(x); % %m = m(x1-2:x2-2,:); % ref(1).mfcc = ref(1).mfcc+m; % end % ref(1).mfcc = ref(1).mfcc/10; %后四个人的语音信号 % for i=2:2 %五个人 % % for Qt=1:10 %次数 % fname1 = sprintf(formatSpec1,i,Qt); % x = audioread(fname1); % x = x(:,1); % %figure() % [x1, x2] = vad(x); % count = x2-x1; % count1 = sum(count(:)); % y = []; % for j=1:length(x1) %此处不能使用i，不然导致ref(i)超出矩阵维度 % y = [y; x(x1(j)*FrameInc +1 :x2(j)*FrameInc)]; % end % x = resample(y,constlen,count1); % % if length(x) > constlen*FrameInc % x(length(x))=[]; % end % if length(x) < constlen*FrameInc % x(length(x)+1)=x(length(x)); % end % % m = mfcc(x); % %m = m(x1-2:x2-2,:); % if Qt==1 % ref(i).mfcc = m; % else % ref(i).mfcc = ref(i).mfcc+m; %报错，错误使用 + 矩阵维度必须一致。 结构体了解一下 % % end % end % ref(i).mfcc = ref(i).mfcc/10; %五个人各自的特征向量 % end disp('Calculating test template parameters~~~') for j=1:Num test(j).mfcc = tFofeveryone( j ); end % disp('正在进行模板匹配...') % dist = zeros(Num,1); % for i=1:Num % for j=1:Num % % dist(i,j) = dtw2(test(j).mfcc, ref(i).mfcc); % % dist(i,j) = disteu(test(i).mfcc, ref(j).mfcc); % temp = abs(test(i).mfcc-ref(j).mfcc); % dist(i,j) = sum(temp(:)); % end % end % disp('正在计算匹配结果...') % for j=1:Num % [M,i] = min(dist(:,j)); %查找在第几行i % if i == j % i % count = count+1; % end % end % fprintf('测试模板的识别成功率为：%d\n', (count/Num)); for k = 1:Num % read test sound file of each speaker distmin = inf; k1 = 0; for l = 1:Num % each trained codebook, compute distortion计算其与每一个码本的距离（失真度） d = disteu(test(k).mfcc, ref(l).mfcc); dist = sum(min(d,[],2)) / size(d,1); if dist < distmin distmin = dist; k1 = l; end end if k == k1 count = count + 1; msg = sprintf('Speaker %d matches with speaker %d--------Successful identification', k, k1); else msg = sprintf('Speaker %d matches with speaker %d--------failed',k,k1); end disp(msg); end msg = sprintf('识别率：%d',count/Num); disp(msg); toc