Mauch MIREX 2010

function [spec] = chromaframe(x,spark);

%  function [spec,mfcc,cq,hpcp,chroma] = chromaframe(x,spark);

% function to return spectrogram, constq, hpcp and chromagram

% for one single frame of arbitrary length

% with the possible inclusion of an adaptive thresholding stage.

fmin = 110; 

fmax = 880;

if nargin<2

    spark = sparsekernel(fmin, fmax, 36, 44100/16, 0.054);

    spark(1025:end,:) = [];

end

% make x a row vector

x = x(:)';

x2 = x;

fs = 44100;

% downsample by a factor of 16.

%x = x(1:16:end);

% x = resample(x,11025,44100);

 x = resample(x,2756,11024);

%  x = resample(x,11024,44096);

len = length(x);

% make windowing function

w = hanningz(len);

w = w(:)';

%  w = ones(1,length(x)); % or no windowing... shouldn't be better..

% take fft

spec = abs(fft(w.*x,2048));

spec = adapt_thresh(spec(1:1024),16,15);

%  %spec = spec(1:1024);

%  

%  %cq = adapt_thresh(abs(spec*spark));   

%  cq = abs(spec*spark);

%  hpcp = zeros(1,36);

%  

%  for k=1:36 

%      hpcp(k) =  sum(abs(cq(k:36:end)));

%  end

%  

%  %hpcp = adapt_thresh(hpcp,2,3);

%  

%  % for k=1:round(abs(log2(fmax/fmin))),

%  %     l(k) = sum(hpcp(k:3:end));

%  % end

%  % 

%  % [val,ind] = max(l);

%  % 

%  % chroma = hpcp(ind:3:end);

%  % 

%  for i=1:12 % sum bins.. 

%      chroma(i) = sum(hpcp((i-1)*3+1:3*i));

%  end

%chroma = adapt_thresh(chroma,3,2);

%  % modulated complex lapped transform

%  % spec = abs(fmclt(x',1024)');

%  % spec = adapt_thresh(spec);

%  

%  % % try to treat everything as gaussians

%  % n = 1:length(cq);

%  % mu = sum(n.*cq)/sum(cq);

%  % sigma = sqrt(sum((n-mu).^2)/sum(cq));

%  % cq = exp(-0.5 * ((n - mu)./sigma).^2) ./ (sqrt(2*pi) .* sigma);

%  % 

%  % n = 1:length(hpcp);

%  % mu = sum(n.*hpcp)/sum(hpcp);

%  % sigma = sqrt(sum((n-mu).^2)/sum(hpcp));

%  % hpcp = exp(-0.5 * ((n - mu)./sigma).^2) ./ (sqrt(2*pi) .* sigma);

%  % 

%  % n = 1:length(chroma);

%  % mu = sum(n.*chroma)/sum(chroma);

%  % sigma = sqrt(sum((n-mu).^2)/sum(chroma));

%  % chroma = exp(-0.5 * ((n - mu)./sigma).^2) ./ (sqrt(2*pi) .* sigma);

%  % 

%  % n = 1:length(spec);

%  % mu = sum(n.*spec)/sum(spec);

%  % sigma = sqrt(sum((n-mu).^2)/sum(spec));

%  % spec = exp(-0.5 * ((n - mu)./sigma).^2) ./ (sqrt(2*pi) .* sigma);

%  

%  %x2 = resample(x2,8000,44100);

%  %spec = mfcc(x',8000,100);

%  %spec = abs(sum(spec'))';

%  

% x2 = x2(1:4:end);

% spec2 = abs(fft(hanning(length(x2))'.*x2,2048));

% spec2 = spec2(1:1024);

%  mlmx = fft2melmx(1024,round(44100/16),40);

% [size(mlmx) size(spec2)];

%  mfcc = 20*log10(1+spec*mlmx');

%  spec = 20*log10(1+spec*mlmx');

%  spec = adapt_thresh(mfcc,2,3);

%  x2 = resample(x2,1378,44096);

%  len = length(x2);

%  w = hanningz(len);

%  w = w(:)';

%  % take fft

%  spec = abs(fft(w.*x2,1024));

%  spec = adapt_thresh(spec(1:512),16,15);

  spec = spec(1:512);

%  mlmx = fft2melmx(512,round(44100),40);

%  spec = spec*mlmx';