matlab.rar_LPCC的matlab实现_SVM 语音_mfcc svm c++_语音识别 svm_语音识别SVM

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Content: T2&BIC(Zhou & Hansen) %% Data: 2009.12.28 %% Author: Bill See(CQUPT) %% Acknowledgment:<Efficient audio Stream Segmentation via the Combined %% T2 Statistic and BIC> %% %% 改进的T2&BIC程序，将原有的程序添加了一个最大窗，超过这个最大窗的时候 %% 程序以滑动窗口运行，直到出现△BIC大于零的值。其中△BIC的值的计算仍与原程序 %% 相同，即调用BICone()函数；除此之外在Shift.m的基础之上，以2分钟为单位再进行 %% 二次窗口的划分，与SV_BIC的窗口相同。 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% tic [y,fs,bits]=wavread('test1.wav'); %news3.wav y=y(:,1); %仅提取单声道的音频 allLen = length(y); %音频的长度 allTime = allLen/fs; %音频的时间 %设定每一帧的长度，和帧的增幅 frameLen = 256; frameMov = 128; SV_BIC_SegT=0; %用来记录每个跳变点的时间 clipLen=floor(120*fs); %将整个语音以段的形式来处理 staClip=1; %设定段的起始点 numClip=0 %段的个数 k=1; T2num=0; labeta=1.05; labetaAck=1; wMin=101; wMax=2100; %窗口 最大值 wShift=300; % wGrow=100; winbod=20; wMar=43; wLap=300; wAck=400; delta=25;% 多次试验证明25是理想的数值 deltaH=5; pp=0; newP=0; while staClip+clipLen<=allLen+clipLen if staClip+clipLen<=allLen s = y(staClip:staClip+clipLen); else s = y(staClip:allLen); end numClip=numClip+1 mel=melcepst(s,fs,'m',12); %用来存储该段内的音频，（分帧以后） En=shorten(s);% 求短时帧能量 z=zcr(s); %求短时过零率 melL=[mel,En,z]; % clear En z; % mel2=deltamfcc(mel,11); %一阶差分MFCC % mel2=dct(mel); %利用dct.m计算一阶差分MFCC % mel=[mel,mel2]; % clear mel2; mel=[mel,En]; [N,fd]=size(mel); k=0; wStart=1; wDyn=wMin; %动态窗值，即窗口的末尾值 while wDyn<N wSignal=mel(wStart:wDyn,:); if wDyn>N wDyn=N; k=k+1; end tempinx=computT(wSignal,wStart,delta); if tempinx<wDyn-winbod & tempinx>wStart+winbod detBIC=computeBIC(wSignal,tempinx-wStart,labeta,fd); else detBIC=-1; end while wDyn<=wStart+wMax & detBIC<0 & wDyn<N & k<2 wDyn=wDyn+wGrow; if wDyn>N wDyn=N; k=k+1; end wSignal=mel(wStart:wDyn,:); tempinx=computT(wSignal,wStart,delta); if tempinx<wDyn-winbod & tempinx>wStart+winbod detBIC=computeBIC(wSignal,tempinx-wStart,labeta,fd); else detBIC=-1; end end while wDyn>wStart+wMax & detBIC<0 & wDyn<N & k<2 wStart=wDyn-wShift; wDyn=wStart+wMax/2;%floor(wMax/2);%floor(3*wMax/4); if wDyn>N wDyn=N; k=k+1; end wSignal=mel(wStart:wDyn,:); tempinx=computT(wSignal,wStart,delta); if tempinx<wDyn-winbod & tempinx>wStart+winbod detBIC=computeBIC(wSignal,tempinx-wStart,labeta,fd); else detBIC=-1; end pp=0; if tempinx>wDyn-winbod wDyn=wStart+wGrow; pp=1; end end if detBIC>0 & pp==0 tempinx2=ackT(mel,tempinx,wStart,wDyn,wAck,delta); detBIC2=computeBIC(wSignal,tempinx2-wStart,labeta,fd); if detBIC2>0 temp=(numClip-1)*(clipLen-wLap)/frameMov+tempinx2; T2num=[T2num,temp]; wStart=tempinx2+wMar; wDyn=wStart+wMin; fnum=ceil((T2num(end-1)*128)/(clipLen-wLap)); lnum=ceil((T2num(end)*128)/(clipLen-wLap)); if fnum==lnum newS=T2num(end-1)-(numClip-1)*(clipLen-wLap)/frameMov+86; else newS=86; end newE=T2num(end)-(numClip-1)*(clipLen-wLap)/frameMov-86; newSignal=mel(newS:newE,:); newP=[newP;LDA(newSignal,newS,newE,labetaAck,fd,deltaH)+(numClip-1)*(clipLen-wLap)/frameMov]; else detBIC=-1; wStart=tempinx-wMin+1; wDyn=wStart+wMin; end end end staClip=staClip+clipLen-wLap; end T2num=T2num*128/fs; T2num=T2num'; % for m=1:N plot(m,tt(m),'k');hold on; end j=2; m=length(T2num); while j<=m if T2num(j)-T2num(j-1)<3.2 T2num=[T2num(1:j-1);[];T2num(j+1:m)]; m=m-1; else j=j+1; end end T2num newP=newP*128/fs; clear mel; fprintf('This is the total time of improved T2BIC :\n'); toc