How to perform k-means clustering on Spectrogram (Short Time Fourier Transform) to cluster frequency band.

https://www.researchgate.net/publication/261954000_Multi-band_multi-centroid_clustering_based_permutation_alignment_for_frequency-domain_blind_speech_separation

In this paper, it suggest using k-means clustering on spectrogram to cluster frequency bands. So, for each audio signal, I get a f x t matrix with f is frequency bin and t is time sequence for that bin. I don't know how to perform k-means on that matrix.

Answer 1

[KSSV]

Check the below example code:

x = rand(1000,1) ; y = rand(1000,1) ;    % random points 
N = 4 ;    % number of groups 
[idx,C,sumd] = kmeans([x y],N) ;  % apply kmeans 
figure
hold on
for i = 1:N
    plot(x(idx==i),y(idx==i),'.') ;
    plot(C(i,1),C(i,2),'*')
end

[Alvin Nguyen]

Hi, It seem you are perform k-mean on just 1000x2 matrix. The my main problem is, the matrix f x t will always has the same number of row f but the number of t will increase for longer file. So I can't get how to perform on such matrix.