BPM detection using Gamedev algorithm

Hello everyone, I tried writing to use Gamedev algorithm in matlab to extract the BPM of a metronome but something isnt working.

I don't get the right BPM for the metronome used.

Hopefully someone can inspect my code and find whats the problem.

http://archive.gamedev.net/archive/reference/programming/features/beatdetection/default.html

% CIRCULAR BUFFERING
filename='120 BPM 4_4 Wood Metronome HD.mp3';
format long;
[y,Fs] = audioread(filename);
yLeft=y(:,1);
yRight=y(:,2);

count=0;
len=length(yLeft);
numOfWin=len/(43*1024); % the length of window in samples (43*1024)
numOfWin=floor(numOfWin) ; %numOfWin- the num. of complete windows withn the song
newNumSamp=numOfWin*(43*1024);
yLeft=y(1:newNumSamp,1);
yRight=y(1:newNumSamp,2);

%CREATE LEFT & RIGHT MATRIX 
LReshapeData=reshape(yLeft,[1024,numOfWin*43]);
RReshapeData=reshape(yRight,[1024,numOfWin*43]);
EWin=zeros(1,43);

% %create the energ buffer for the first window:
 for segmentInd=43:-1:1   
     Eseg=sum((LReshapeData(:,segmentInd).^2)+(RReshapeData(:,segmentInd).^2) ); 
     EWin((43-segmentInd)+1)=Eseg;

 end


  for segmentInd=44:numOfWin*43 
    Eseg=sum((LReshapeData(:,segmentInd).^2)+(RReshapeData(:,segmentInd).^2) );
    avgE=sum(EWin)/43; 
    VarE=(1/43)*sum( (EWin-avgE).^2);
     C=(-0.0025714*VarE)+1.5142857;

      if EWin(1)> abs(C)*avgE 
          count=count+1;
      end
       EWin(1:end)=[Eseg,EWin(1:42)];%shift right the buffer in 1 segment
  end


 tsec=newNumSamp/Fs; % t=n*Ts- the total length in seconds
 tmin=tsec/60;
 numBPM=count/tmin;

[Federico Dentesano]

Hi have you finished the code? I am trying to understand how BPM tracking works and this code was helpful, I would like to see it complete if you have it..

Answer 1

[Dinesh Yadav]

The missing points in the above implementation are as follows:-

• The first sample points you have used for comparing average energy to "C" is 1:1024, but the second sample you have used for comparing is of 44th column i.e. 44032:43056 samples. Therefore, you are missing to compare energies of 2nd to 43rd column. Change your update "EWin" command code as shown below:

EWin(1: end)=[Ewin(2:43),Eseg];

• The second one is related to flooring of sample points. Even after flooring you should store those extra sample points and compute and compare energies of those also. Here the sample points after flooring have been discarded so even if there is a beat at the ending the above algorithm won't detect that.