How to solve dimension problem

I got script like this and for some reason it works with some audio samples and with other it doesnt.

asc

ascacsSamples have same length exactly same number of samples but for some reason it gives me this error.

% 

rm 

assignment because the size of the left side is 13-by-1 and the size of the right side

is 3-by-1.

Error in stFeatureExtractionmfcc (line 61)

mfccFeatures(:,i) = zeros(numOfFeatures, 1);

Error in Mfccshoww (line 39)

Curframe{i} = stFeatureExtractionmfcc(frame, fs, 0.09, 0.03);

[Signal, fs] = audioread('long_sample_1.wav');
if (size(Signal,2)>1)
    Signal = (sum(Signal,2)/2);% convert to MONO
end
% win = 0.31;
% step = 0.2;
%Testsamplemfcc = stFeatureExtractionmfcc(Signal, fs, win, step);


 %spectrum model
% %  fs = 44100; 
% %  T = 1/fs;
% %  L = length(Features);
% %  f = fs*(0:L-1)/L;
% %  t = (0:L-1)*T;
% % 
win = 0.31;
%win = 0.310090703;  %% length of a single bark match with the window and step
step = 0.1;
curPos = 1;
windowLength = (win * fs); 
Stepf = round(step * fs); 
d = cell(1,1);
Length_signal = length(Signal);
numberOfFrames =  floor((Length_signal-windowLength)/Stepf) + 1;

 
  % = Featuresaudio(1:windowLength);
% % Featuresaudio 
% [Features] = getDFT(Featuresaudio, 44100);
 
Ham = window(@hamming, windowLength);
for i = 1:numberOfFrames
    frame  = (Signal(curPos:curPos+windowLength-1));
    %frame  = frame .* Ham;
    Curframe{i} =  stFeatureExtractionmfcc(frame, fs, 0.09, 0.03);
    %Curframe{i} = mean(Curframe{i},2);
    Curframe{i} = Curframe{i};   % in case of transposing
    d{i}= ((Curframe{i}-mfccFeatures{1}).^2);
    d{i} = sum(d{i});
    d{i} = sum(d{i},2);

    curPos = curPos + Stepf;
   
end

d = cell2mat(d);

idx = zeros(1,1)
if all(d(:) > 200)
        fprintf('Dog wasnt found in any frame')
else 
   idx  = find(d(:) < 170);
         %fprintf('Dog detected in frame number %d\n',idx(:))
end

for i = 1:length(idx);
     idx(i) = idx(i) *0.1;
 end 
 
 idx = round(idx,2);

  fprintf('Dog detected at second %0.2f\n',idx(:))  

Answer 1

[Matt Gaidica]

It appears stFeatureExtractionmfcc() is returning multiple values. What is in that function?

[Kamil Kacer]

function mfccFeatures = stFeatureExtractionmfcc(signal, fs, win, step)
% function Features = stFeatureExtraction(signal, fs, win, step)
%
% This function computes basic audio feature sequencies for an audio
% signal, on a short-term basis.
%
% ARGUMENTS:
%  - signal:    the audio signal
%  - fs:        the sampling frequency
%  - win:       short-term window size (in seconds)
%  - step:      short-term step (in seconds)
%
% RETURNS:
%  - Features: a [MxN] matrix, where M is the number of features and N is
%  the total number of short-term windows. Each line of the matrix
%  corresponds to a seperate feature sequence
%
% (c) 2014 T. Giannakopoulos, A. Pikrakis
% if STEREO ...
if (size(signal,2)>1)
    signal = (sum(signal,2)/2); % convert to MONO
end
% convert window length and step from seconds to samples:
windowLength = round(win * fs);
step = round(step * fs);
curPos = 1;
L = length(signal);
% compute the total number of frames:
numOfFrames = floor((L-windowLength)/step) + 1;
% number of features to be computed:
numOfFeatures = 3;
mfccFeatures = zeros(numOfFeatures, numOfFrames); %
Ham = window(@hamming, windowLength);
mfccParams = feature_mfccs_init(windowLength, fs);
for i=1:numOfFrames % for each frame
    % get current frame:
    frame  = signal(curPos:curPos+windowLength-1);
    frame  = frame .* Ham;
    frameFFT = getDFT(frame, fs);
    
    if (sum(abs(frame))>eps)
        % compute time-domain features:
       %Features(1,i) = feature_zcr(frame);
       %Features(2,i) = feature_energy(frame);
      % Features(3,i) = feature_energy_entropy(frame, 10);
        % compute freq-domain features: 
        if (i==1) frameFFTPrev = frameFFT; end;
      %  [Features(4,i) Features(5,i)] = ...
         %   feature_spectral_centroid(frameFFT, fs);
       % Features(4,i) = feature_spectral_entropy(frameFFT, 10);
    %    Features(7,i) = feature_spectral_flux(frameFFT, frameFFTPrev);
     %   Features(8,i) = feature_spectral_rolloff(frameFFT, 0.90);
        MFCCs = feature_mfccs(frameFFT, mfccParams);
       mfccFeatures(1:13,i)  = MFCCs;
        
      %  [HR, F0] = feature_harmonic(frame, fs);
       % Features(22, i) = HR;
        %Features(23, i) = F0;        
        %Features(23+1:23+12, i) = feature_chroma_vector(frame, fs);
    else
        mfccFeatures(:,i) = zeros(numOfFeatures, 1);
    end    
    curPos = curPos + step;
    frameFFTPrev = frameFFT;
end
mfccFeatures(13, :) = medfilt1(mfccFeatures(13, :), 3);

[Matt Gaidica]

You would need to provide all the dependencies and your audio sample. This is a mess to debug for you otherwise.

[Walter Roberson]

       mfccFeatures(1:13,i)  = MFCCs;

That does make the array have at least 13 rows

      mfccFeatures(:,i) = zeros(numOfFeatures, 1);

but that tries to assign to only 3 rows.

The difference is that the problem occurs when

if (sum(abs(frame))>eps)

is false. In other words, it happens when there is a frame that is effectively silent but which is not the first frame.

[Kamil Kacer]

So it works as long as there is no silence in recording right ?

[Kamil Kacer]

Can you provide me with solution ? please

[Walter Roberson]

No, the situation is undefined, the code can only extract features when features exist, and does not define what the output should be when no features exist.