Silence Removal and End Point Detection MATLAB Code

Visit http://ganeshtiwaridotcomdotnp.blogspot.com/2011/06/final-report-text-prompted-remote.html for more detail about our project.
For the purpose of silence removal of captured sound, we used the algorithm specified in
"A New Silence Removal and Endpoint Detection Algorithm for Speech and Speaker Recognition Applications"
Our actual system was in JAVA but we verified the performance of this algorithm in MATLAB.

Inputs and Output
Before silence removal
After automatic silence removal
Here is the Matlab code : 
It first records sound for 5 seconds and removes the silence and then plays back.




//silence removal and end point detection 
THRESHOLD=0.3; // adjust value yourself 
TIME=5;
%capture;
Fs = 11025;
y = wavrecord(TIME*Fs,Fs);
%plot(y)
%wavplay(y,Fs);
samplePerFrame=floor(Fs/100);
bgSampleCount=floor(Fs/5); %according to formula, 1600 sample needed for 8 khz

%----------
%calculation of mean and std
bgSample=[];
for i=1:1:bgSampleCount
    bgSample=[bgSample y(i)];
end
meanVal=mean(bgSample);
sDev=std(bgSample);
%----------
%identify voiced or not for each value
for i=1:1:length(y)
   if(abs(y(i)-meanVal)/sDev > THRESHOLD)
       voiced(i)=1;
   else
       voiced(i)=0;
   end
end


% identify voiced or not for each frame
%discard insufficient samples of last frame

usefulSamples=length(y)-mod(length(y),samplePerFrame);
frameCount=usefulSamples/samplePerFrame;
voicedFrameCount=0;
for i=1:1:frameCount
   cVoiced=0;
   cUnVoiced=0;
   for j=i*samplePerFrame-samplePerFrame+1:1:(i*samplePerFrame)
       if(voiced(j)==1)
           cVoiced=(cVoiced+1);
       else
           cUnVoiced=cUnVoiced+1;
       end
   end
   %mark frame for voiced/unvoiced
   if(cVoiced>cUnVoiced)
       voicedFrameCount=voicedFrameCount+1;
       voicedUnvoiced(i)=1;
   else
       voicedUnvoiced(i)=0;
   end
end

silenceRemovedSignal=[];

%-----
for i=1:1:frameCount
    if(voicedUnvoiced(i)==1)
    for j=i*samplePerFrame-samplePerFrame+1:1:(i*samplePerFrame)
            silenceRemovedSignal= [silenceRemovedSignal y(j)];
        end
    end
end

%---display plot and play both sounds
    figure; plot(y);
    figure; plot(silenceRemovedSignal);
%%%play
wavplay(y,Fs);
wavplay(silenceRemovedSignal,Fs);
NOTE: Don't forget to adjust the microphone level and sound boost feature to achieve good results.

10 comments :

  1. Wow.. Excellent.. Can u help me compare the efficiency of this algorithm with the STE and ZCR methods?? Thank you

    ReplyDelete
  2. @Lakshmi,
    As shown in research paper, the comparision of efficiency is as follows :
    Phrases______STE______ZCR-STE__Proposed_Method
    Combination
    lock number__77.9531% 70.3720% 83.5565%
    Running Text_50.8391% 50.1231% 59.7181%

    ReplyDelete
  3. good work dear,thanks for sharing

    ReplyDelete
  4. Hi Ganesh,
    Thank you so much for sharing this code. It is brilliant and yet simple to understand. Do you mind to give me the paper entitled 'A New Silence Removal and Endpoint Detection Algorithm for Speech and Speaker Recognition Applications' since I dont have access to that book. My email address yusnita082@gmail.com but why is it your name is not one of the author?

    ReplyDelete
  5. what about remove silence from first and end of speech?

    ReplyDelete
    Replies
    1. plz take a look at the code and the plot ... this algorithm removes silence from every part of speech.

      Delete
  6. This comment has been removed by the author.

    ReplyDelete
  7. it doesn't work, i dont know why?

    ReplyDelete
  8. I`m trying to import a wave but the Threshold is diferent for each other. How I can ajust the Threshold acordly to the wav.
    And the line 10 of your code the "bgSampleCount=floor(Fs/5); %according to formula, 1600 sample needed for 8 khz"
    Where did you find 5 ? its because of Time = 5 ?

    ReplyDelete

Your Comment and Question will help to make this blog better...