view mirex2012-matlab/cell2sparse.m @ 116:91bb029a847a timing

Reorder the calculations to match the series of vector operations in the most recent bqvec code, just in case it's the order of vector calculations that is saving the time rather than the avoidance of std::vector

function spCQT = cell2sparse(Xcq,octaves,bins,firstcenter,atomHOP,atomNr)
%Generates a sparse matrix containing the CQT coefficients (rasterized).
%
%The sparse matrix representation of the CQT coefficients contains all
%computed coefficients at the corresponding time-frequency location
%(similar to a spectrogram). For lower frequencies this means, that
%each coefficient is followed by zeros stemming from the fact, that the time
%resolution for lower frequencies decreases as the frequency resolution
%increases. Due to the design of the CQT kernel, however, the coefficients
%of different octaves are synchronised, meaning that for the second highest
%octave each coefficient is followed by one zero, for the next octave down
%two zeros are inserted, for the next octave four zeros are inserted and so
%on.
%
%INPUT:
%   Xcq         ... Cell array consisting of all coefficients for all octaves
%   octaves     ... Number of octaves processed
%   bins        ... Number of bins per octave
%   firstcenter ... Location of the leftmost atom-stack in the temporal
%                   kernel
%   atomHOP     ... Spacing of two consecutive atom stacks
%   atomNr      ... Number of atoms per bin within the kernel
%
%Christian Schörkhuber, Anssi Klapuri 2010-06

if 0
%% this version has big memory consumption but is very fast
    emptyHops = firstcenter/atomHOP;
    drop = emptyHops*2^(octaves-1)-emptyHops; %distance between first value in highest octave and first value in lowest octave
    spCQT = zeros(bins*octaves,size(Xcq{1},2)*atomNr-drop);

    for i=1:octaves
        drop = emptyHops*2^(octaves-i)-emptyHops; %first coefficients of all octaves have to be in synchrony
        X = Xcq{i}; 
        if  atomNr > 1 %more than one atom per bin --> reshape
            Xoct = zeros(bins,atomNr*size(X,2)-drop);
            for u=1:bins %reshape to continous windows for each bin (for the case of several wins per frame)
               octX_bin = X((u-1)*atomNr+1:u*atomNr,:);
               Xcont = reshape(octX_bin,1,size(octX_bin,1)*size(octX_bin,2));
               Xoct(u,:) = Xcont(1+drop:end);
            end
            X = Xoct;
        else
            X = X(:,1+drop:end);
        end
        binVec = bins*octaves-bins*i+1:bins*octaves-bins*(i-1);
        spCQT(binVec,1:2^(i-1):size(X,2)*2^(i-1)) = X;

    end
    spCQT = sparse(spCQT); %storing as sparse matrix at the end is the fastest way. Big memory consumption though!

else
%% this version uses less memory but is noticable slower
    emptyHops = firstcenter/atomHOP;
    drops = emptyHops*2.^(octaves-(1:octaves))-emptyHops;
    len = max(((atomNr*cellfun('size',Xcq,2)-drops).*2.^(0:octaves-1))); %number of columns of output matrix
    spCQT = [];

    for i=octaves:-1:1
        drop = emptyHops*2^(octaves-i)-emptyHops; %first coefficients of all octaves have to be in synchrony
        X = Xcq{i}; 
        if  atomNr > 1 %more than one atom per bin --> reshape
            Xoct = zeros(bins,atomNr*size(X,2)-drop);
            for u=1:bins %reshape to continous windows for each bin (for the case of several wins per frame)
               octX_bin = X((u-1)*atomNr+1:u*atomNr,:);
               Xcont = reshape(octX_bin,1,size(octX_bin,1)*size(octX_bin,2));
               Xoct(u,:) = Xcont(1+drop:end);
            end
            X = Xoct;
        else
            X = X(:,1+drop:end);
        end
        X = upsample(X.',2^(i-1)).';
        X = [X zeros(bins,len-size(X,2))];
        spCQT = sparse([spCQT; X]);  
    end

end