Audio Feaure Extraction Matrix

Hi guys! I am currently working on drum transcription through the query of beatbox. It requires audio feature extraction and the respective classification training and testing. I am currently having problems computing just one extracted feature, MFCC, eve more when trying to put others into a matrix, like Zero Crossing Rate and Spectral Centroid/Temporal Centroid. My idea is to grab a percussive audio signal (kick), split it into 5 successive frames of an equal interval of that same signal.

Grab Frame 1 until Frame 5 and extract features for each of those frames (MFCC, Spectral Centroid,etc). If we were to put it practically, 5 frames (per signal), times 5 features equals to 25 values.

Next step would be to take a mean of the respective frames' features. Input all the kicks into a matrix (319x5) for all the 319 events of kicks, times the 5 features. Repeat for all the other supervised labelled classes.

After I want to introduce them in a Support Vector Machine, train the data and test it against the Test database of pre-recorded drum hits.

Here is what I am attempting:

mfcc_size = 13;
TRAINING = [];
Y = 0;
n=0;
% repeat from here....
% change directory
wavfiles = dir('*.wav')
%copy this...?
for file = wavfiles'
  n=n+1;
  fprintf(1,'Processing %s\n',file.name)
  a = file.name
  f = mirframe(a);hold off
  s = mirspectrum(f,'Frame','Window','hanning','Mel');hold off
  m = mirspectrum(s,'Mel');hold off
  c = mirmfcc(m,'Rank',1:13);
  temp = mirgetdata(c);
  set(gcf, 'name',file.name); colorbar
  TRAINING(:,n) = temp(2:13);
  Y = [Y 1]; %where this is the class label for just one instrument (kick)
% eval(['print -djpeg SpectrviewMFCC_' num2str(file.name) '.jpeg']);
end
% to here...?

_____

To train, extract MFCCs from training data = TRAINING / Create labelled data of Y(class labels = Y /

svsmstruct = svmtrain(TRAINING, Y) / extract MFCCs from testing data = Test/ RESULTS = svmclassify (svmstruct,TEST)/

Any thoughts and directions on structure, syntax, solutions for inputting data and extracting features followed by tips for training/classifying within MatLab?

P.S I am currently using MIRToolbox, to do all the processing. I tried to implement Weka with Matlab, but no avail, as I am still not able to compute my audio feature matrix and output data to train.

Best regards, I hope to hear from you all, which have been a very supportive community so far! Antonio