Matlab在语音识别中的应用.docx

1、Matlab在语音识别中的应用1.基于GUI的音频采集处理系统注：本实验是对“东、北、大、学、中、荷、学、院”孤立文字的识别！首先是GUI的建立，拖动所需控件，双击控件，修改控件的参数；主要有string Tag(这个是回调函数的依据)，其中还有些参数如value style 也是需要注意的，这个在实际操作中不能忽视。这里需要给说明一下：图中所示按钮都是在一个按钮组里面，都属于按钮组的子控件。所以在添加回调函数时，是在按钮组里面添加的，也就是说右击三个按钮外面的边框，选择View CallbackSelectionChange,则在主函数中显示该按钮的回调函数：function uipanel

2、1_SelectionChangeFcn(hObject, eventdata, handles)以第一个按钮“录音”为例讲解代码；下面是“播放”和“保存”的代码：以上就是语音采集的全部代码。程序运行后就会出现这样的界面：点击录音按钮，录音结束后就会出现相应波形：点击保存，完成声音的保存，保存格式为.wav。这就完成了声音的采集。2.声音的处理与识别2.1打开文件语音处理首先要先打开一个后缀为.wav的文件，这里用到的不是按钮组，而是独立的按钮，按钮“打开”的回调函数如下：function pushbutton1_Callback(hObject, eventdata, handles)其中p

3、ushbutton1是“打开”按钮的Tag.在回调函数下添加如下代码：运行结果如图：2.2预处理回调函数如下：function pushbutton2_Callback(hObject, eventdata, handles)运行结果如图：2.3短时能量短时能量下的回调函数：function pushbutton3_Callback(hObject, eventdata, handles)其回调函数下的代码是：2.4端点检测这里要先声明一点，为了避免在以后的函数调用中，不能使用前面的变量，所以其实后面的函数都包含了前面的部分。显而易见这样程序就会显得很冗长，这也是值得以后修改的地方。funct

4、ion pushbutton4_Callback(hObject, eventdata, handles)2.5生成模版本功能和上面重复的部分省略掉了，现在只补充添加的代码：2.6语音识别将打开的语音与提前录好的语音库进行识别，采用的是DTW算法。识别完后就会在相应的文本框里显示识别的文字。代码如下：程序运行前后的对比图：GUI的整体效果图：总结实验已经实现了对“东、北、大、学、中、荷、学、院”文字的识别，前提是用模版的语音作为样本去和语音库测试，这已经可以保证的正确率，这说明算法是正确的，只是需要优化。而现场录音和模版匹配时，则不能保证较高的正确率，这说明特征参数的提取这方面还不够完善。特征

5、参数提取的原则是类内距离尽量小，类间距离尽量大的原则，这是需要以后完善的地方。也需要优化，先生成一个模版库，然后用待测语音和模版库语音识别，让这个模版库孤立出来，不需要每次测试都要重复生成模版库，提高运算速率。以后有机会可以实现连续语音的识别！附件这是全部代码文件mfcc.mat 文件是程序运行过程中生成的；test 文件夹里面存放了录音的模版：这里是6个.M文件，如下：1 WienerScalart96.mfunction output=WienerScalart96(signal,fs,IS) % output=WIENERSCALART96(signal,fs,IS)% Wiener f

6、ilter based on tracking a priori SNR usingDecision-Directed % method, proposed by Scalart et al 96. In this method it is assumed that% SNRpost=SNRprior +1. based on this the Wiener Filter can be adapted to a% model like Ephraims model in which we have a gain function which is a% function of a priori

7、 SNR and a priori SNR is being tracked using Decision% Directed method. % Author: Esfandiar Zavarehei% Created: MAR-05 if (nargin=3 & isstruct(IS)%This option is for compatibility with another programmeW=IS.windowsizeSP=IS.shiftsize/W;%nfft=IS.nfft;wnd=IS.window;if isfield(IS,IS)IS=IS.IS;elseIS=.25;

8、endend% .UP TO HERE pre_emph=0;signal=filter(1 -pre_emph,1,signal); NIS=fix(IS*fs-W)/(SP*W) +1);%number of initial silence segments y=segment(signal,W,SP,wnd); % This function chops the signal into framesY=fft(y);YPhase=angle(Y(1:fix(end/2)+1,:); %Noisy Speech PhaseY=abs(Y(1:fix(end/2)+1,:);%Specrog

9、ramnumberOfFrames=size(Y,2);FreqResol=size(Y,1); N=mean(Y(:,1:NIS); %initial Noise Power Spectrum meanLambdaD=mean(Y(:,1:NIS).2);%initial Noise Power Spectrum variancealpha=.99; %used in smoothing xi (For Deciesion Directed method for estimation of A Priori SNR)NoiseCounter=0;NoiseLength=9;%This is

10、a smoothing factor for the noise updatingG=ones(size(N);%Initial Gain used in calculation of the new xiGamma=G; X=zeros(size(Y); % Initialize X (memory allocation) h=waitbar(0,Wait.); for i=1:numberOfFrames%VAD and Noise Estimation STARTif i=NIS % If initial silence ignore VADSpeechFlag=0;NoiseCount

11、er=100;else % Else Do VADNoiseFlag, SpeechFlag, NoiseCounter, Dist=vad(Y(:,i),N,NoiseCounter); %Magnitude Spectrum Distance VADend if SpeechFlag=0 % If not Speech Update Noise ParametersN=(NoiseLength*N+Y(:,i)/(NoiseLength+1); %Update and smooth noise meanLambdaD=(NoiseLength*LambdaD+(Y(:,i).2)./(1+

12、NoiseLength); %Update and smooth noise varianceend%VAD and Noise Estimation END gammaNew=(Y(:,i).2)./LambdaD; %A postiriori SNRxi=alpha*(G.2).*Gamma+(1-alpha).*max(gammaNew-1,0); %Decision Directed Method for A Priori SNRGamma=gammaNew; G=(xi./(xi+1); X(:,i)=G.*Y(:,i); %Obtain the new Cleaned value

13、waitbar(i/numberOfFrames,h,num2str(fix(100*i/numberOfFrames);end close(h);output=OverlapAdd2(X,YPhase,W,SP*W); %Overlap-add Synthesis of speechoutput=filter(1,1 -pre_emph,output); %Undo the effect of Pre-emphasis function ReconstructedSignal=OverlapAdd2(XNEW,yphase,windowLen,ShiftLen); %Y=OverlapAdd

14、(X,A,W,S);%Y is the signal reconstructed signal from its spectrogram. X is a matrix%with each column being the fft of a segment of signal. A is the phase%angle of the spectrum which should have the same dimension as X. if it is%not given the phase angle of X is used which in the case of real values

15、is%zero (assuming that its the magnitude). W is the window length of time%domain segments if not given the length is assumed to be twice as long as%fft window length. S is the shift length of the segmentation process ( for%example in the case of non overlapping signals it is equal to W and in the%ca

16、se of %50 overlap is equal to W/2. if not givven W/2 is used. Y is the%reconstructed time domain signal.%Sep-04%Esfandiar Zavarehei if nargin2yphase=angle(XNEW);endif nargin3windowLen=size(XNEW,1)*2;endif nargin4ShiftLen=windowLen/2;endif fix(ShiftLen)=ShiftLenShiftLen=fix(ShiftLen);disp(The shift l

17、ength have to be an integer as it is the number of samples.)disp(shift length is fixed to num2str(ShiftLen)end FreqRes FrameNum=size(XNEW); Spec=XNEW.*exp(j*yphase); if mod(windowLen,2) %if FreqResol is oddSpec=Spec;flipud(conj(Spec(2:end,:);elseSpec=Spec;flipud(conj(Spec(2:end-1,:);endsig=zeros(Fra

18、meNum-1)*ShiftLen+windowLen,1);weight=sig;for i=1:FrameNumstart=(i-1)*ShiftLen+1; spec=Spec(:,i);sig(start:start+windowLen-1)=sig(start:start+windowLen-1)+real(ifft(spec,windowLen); endReconstructedSignal=sig; function Seg=segment(signal,W,SP,Window) % SEGMENT chops a signal to overlapping windowed

19、segments% A= SEGMENT(X,W,SP,WIN) returns a matrix which its columns are segmented% and windowed frames of the input one dimentional signal, X. W is the% number of samples per window, default value W=256. SP is the shift% percentage, default value SP=0.4. WIN is the window that is multiplied by% each

20、 segment and its length should be W. the default window is hamming% window.% 06-Sep-04% Esfandiar Zavarehei if nargin3SP=.4;endif nargin2W=256;endif nargin4Window=hamming(W);endWindow=Window(:); %make it a column vector L=length(signal);SP=fix(W.*SP);N=fix(L-W)/SP +1); %number of segments Index=(rep

21、mat(1:W,N,1)+repmat(0:(N-1)*SP,1,W);hw=repmat(Window,1,N);Seg=signal(Index).*hw; function NoiseFlag, SpeechFlag, NoiseCounter, Dist=vad(signal,noise,NoiseCounter,NoiseMargin,Hangover) %NOISEFLAG, SPEECHFLAG, NOISECOUNTER, DIST=vad(SIGNAL,NOISE,NOISECOUNTER,NOISEMARGIN,HANGOVER)%Spectral Distance Voi

22、ce Activity Detector%SIGNAL is the the current frames magnitude spectrum which is to labeld as%noise or speech, NOISE is noise magnitude spectrum template (estimation),%NOISECOUNTER is the number of imediate previous noise frames, NOISEMARGIN%(default 3)is the spectral distance threshold. HANGOVER (

23、 default 8 )is%the number of noise segments after which the SPEECHFLAG is reset (goes to%zero). NOISEFLAG is set to one if the the segment is labeld as noise%NOISECOUNTER returns the number of previous noise segments, this value is%reset (to zero) whenever a speech segment is detected. DIST is the%s

24、pectral distance. %Saeed Vaseghi%edited by Esfandiar Zavarehei%Sep-04 if nargin4NoiseMargin=3;endif nargin5Hangover=8;endif nargin3NoiseCounter=0;end FreqResol=length(signal); SpectralDist= 20*(log10(signal)-log10(noise);SpectralDist(find(SpectralDist0)=0; Dist=mean(SpectralDist); if (Dist Hangover)

25、 SpeechFlag=0; else SpeechFlag=1; end2 mfcc.mfunction cc=mfcc(k)%-% cc=mfcc(k)计算语音k的MFCC系数%-% M为滤波器个数，N为一帧语音采样点数M=24; N=256;% 归一化mel滤波器组系数bank=melbankm(M,N,22050,0,0.5,m);figure;plot(linspace(0,N/2,129),bank);title(Mel-Spaced Filterbank);xlabel(Frequency Hz);bank=full(bank);bank=bank/max(bank(:); %

26、DCT系数,12*24for i=1:12 j=0:23; dctcoef(i,:)=cos(2*j+1)*i*pi/(2*24);end% 归一化倒谱提升窗口w=1+6*sin(pi*1:12./12);w=w/max(w);% 预加重AggrK=double(k);AggrK=filter(1,-0.9375,1,AggrK);% 分帧FrameK=enframe(AggrK,N,80);% 加窗for i=1:size(FrameK,1) FrameK(i,:)=(FrameK(i,:).*hamming(N);endFrameK=FrameK;% 计算功率谱S=(abs(fft(Fra

27、meK).2;disp(显示功率谱)figure; plot(S);axis(1,size(S,1),0,2);title(Power Spectrum (M=24, N=256);xlabel(Frame);ylabel(Frequency Hz);colorbar; % 将功率谱通过滤波器组P=bank*S(1:129,:);% 取对数后作离散余弦变换D=dctcoef*log(P);% 倒谱提升窗for i=1:size(D,2) m(i,:)=(D(:,i).*w);end% 差分系数dtm=zeros(size(m);for i=3:size(m,1)-2 dtm(i,:)=-2*m

28、(i-2,:)-m(i-1,:)+m(i+1,:)+2*m(i+2,:);enddtm=dtm/3;%合并mfcc参数和一阶差分mfcc参数cc=m,dtm;%去除首尾两帧，因为这两帧的一阶差分参数为0cc=cc(3:size(m,1)-2,:);3 getpoint.mfunction StartPoint,EndPoint=getpoint(k,fs)%UNTITLED 此处显示有关此函数的摘要% 此处显示详细说明 signal=WienerScalart96(k,fs);sigLength=length(signal);%计算信号长度t=(0:sigLength-1)/fs;%计算信号对应时间坐标FrameLen = round(0.012/max(t)*sigLength);%定义每一帧长度FrameInc = round(FrameLen/3);%每一帧的重叠区域，选为帧长的1/31/2tmp=enframe(signal(1:end), FrameLen, FrameInc);