How to apply pca to audio files (pitch features)?

Hello,

I applied pitch feature extraction to audio files. Now I want to apply PCA (for dimensionality reduction) to the pitch features. How can I do that? The problem is that I cannot save the features in csv or mat file. Also, the new audio feature replace the previous audio features thus giving me only feature vector i.e; comp_PCA1 gives me feature vector of 32600x1. But my files are 94 so the output should be 32600x94.

My code is given below:

clc
clear all
path_to_directory= dir ('**/*.wav'); % path to the directory 
for i = 1:length(path_to_directory)
    path= strcat(path_to_directory(i).folder, '\', path_to_directory(i).name);
    [signal, Fs]= audioread(path); % audio read here
     pitch = pwelch(signal);   % pitch feature extraction part
     [coeff,score] = pca(pitch); % pca for dimensionality reduction
     Comp_PCA1 = score(:,1);
% %      y (i,:)= coeff;   
end

[Walter Roberson]

Comp_PCA1(:,i) = score(:,1);

However you would have trouble if the files are not exactly the same number of samples in each. What do you want to do in that case?

[IKRAM HUSSAIN]

Hey Walter Roberson,

Thanks for your reply

I also tried the above code and your updated line, but the got this error: ''Unable to perform assignment because the indices on the left side are not compatible with the size of the right side''

Also, is it possible to save only sepific no of feature only i.e 16000? and discard the rest.

because the features (pitch) dimensions of diiferent audio files are not the same so can't be saved in one variable.

[W.Khan]

Thank you for your answer.

Yes, I have the same problem. "Unable to perform assignment because the indices on the left side are not compatible with the size of the right side."

When I run the syntax below, then Its save 32000x60 (mean 60 features have same dimension); but the rest have 16000 features.

Comp_PCA1(:,i) = score;

[Walter Roberson]

path_to_directory= dir ('**/*.wav'); % path to the directory 
for i = 1:length(path_to_directory)
    filename = fullfile(path_to_directory(i).folder, path_to_directory(i).name);
    [signal, Fs]= audioread(path); % audio read here
     pitch = pwelch(signal);   % pitch feature extraction part
     [coeff,score] = pca(pitch); % pca for dimensionality reduction
     this_Comp = score(:,1);
     nR = size(this_Comp,1);
     if i == 1
         Comp_PCA1 = this_Comp;
     elseif nR <= size(Comp_PCA1,1)
         Comp_PCA1 = [Comp_PCA1(1:nR,:), this_Comp];
     else
         Comp_PCA1(:,i) = this_Comp(1:size(Comp_PCA1,1));
     end
end

If I got everything right then this should make Comp_PCA1 correspond in length to the shortest of the files.

[W.Khan]

Dear Walter Roberson,

Thank you very much. It worked perfectly!

[Ridwan Alam]

Great! Please mark this problem as solved or accept Walter's answer. Thanks!

[Walter Roberson]

(I had only posted as a Comment before, so it could not be Accepted.)

Answer 1

[Walter Roberson]

path_to_directory= dir ('**/*.wav'); % path to the directory 
for i = 1:length(path_to_directory)
    filename = fullfile(path_to_directory(i).folder, path_to_directory(i).name);
    [signal, Fs]= audioread(path); % audio read here
     pitch = pwelch(signal);   % pitch feature extraction part
     [coeff,score] = pca(pitch); % pca for dimensionality reduction
     this_Comp = score(:,1);
     nR = size(this_Comp,1);
     if i == 1
         Comp_PCA1 = this_Comp;
     elseif nR <= size(Comp_PCA1,1)
         Comp_PCA1 = [Comp_PCA1(1:nR,:), this_Comp];
     else
         Comp_PCA1(:,i) = this_Comp(1:size(Comp_PCA1,1));
     end
end

If I got everything right then this should make Comp_PCA1 correspond in length to the shortest of the files.