录音文件信噪比计算，实现语音活动检测

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 功能：实现语音活动检测 % 算法实现：可参见文献 % 输入参数说明：语音信号，一帧信号的长度 % 返回参数说明：语音是否活动的标志 % 作者：姜莎 % 日期：2006.08.24 % 修改日期：2006.10.09 % 参考文献：sohn, "A voice activity detector employing soft decision based noise spectrum adaptation" % 特殊说明：初始值采用第一帧信号的fft的值的模的平方做初始能量，并加上小的初始定值 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% clear; signal= wavread('D:\graduation project\project1\voice\jzy5.wav'); L=256; %一帧信号的长度 N=floor(length(signal)./L); %信号总的帧长 input_signal=signal(1:L.*N); %采用第一帧信号的fft的值模的平方做lambda的初始值 firstframe = input_signal(1:L); fftfirstframe = fft(firstframe); lambda = ones(L,1).*(abs(fftfirstframe).^2); lambda0 = ones(L,1).*((100./(2.^15)).^2); %避免在静音的时候出现snr无穷大的情况 beta = lambda+lambda0; %计算信噪比所采用的背景噪声的能量 %================================= % snrbuffer = zeros(10,1); snrthr=5; % snrtempcnt=0; %================================= snr = zeros(N,1); snrtemp=zeros(N,1); proratio = 10; %语音和噪声的概率比probabilityratio，给定为10 for i = 1 : N signalframe = input_signal((i-1).*L+1 : i.*L); framesigfft = fft(signalframe); snr = (1./L).*sum(abs(framesigfft).^2./beta); snrtemp(i) = snr; if i <= 20 speechactive(i) = 1; else if speechactive(i-1)==1 if snrtemp(i) > snrthr speechactive(i) = 1; else logi = (snrtemp(i-19 : i-1) > snrthr); if sum(logi) >= 3 speechactive(i) =1; else speechactive(i) = 0; end end else if (snrtemp(i) > snrthr && snrtemp(i-1) > snrthr) speechactive(i) = 1; else speechactive(i) = 0; end end end if speechactive(i) == 0 lambda =(1./(1+snr.*proratio)).*abs(framesigfft).^2+(snr.*proratio/(1+snr.*proratio)).*lambda; beta = lambda + lambda0; end end length=300; times=floor(N/length); b=zeros(times); for i=1:times b(i)=sum(snrtemp(length*(i-1)+1:length*i),1); end %========================== %==========================