MATLAB: MFCC into feature vector

Audio Toolboxfeature extractionMATLABmfcc

Hello, right now im working on baby cry meaning using MFCC for feature extraction
this is my code for mfcc
[audioIn, fs] = audioread('Lelah2.wav');
coeffs = mfcc(audioIn, fs);
so the result is a matrix, not a feature vector
any suggestion to change the matrix into a feature vector ?

Best Answer

The mfcc function returns mel frequnecy cepstral coefficients (MFCC) over time. That is, it separates the audio into short windows and calculates the MFCC (aka feature vectors) for each window.
For example, in this scenario:
coeffs = mfcc(audioIn,fs);
[L,M,N] = size(coeffs);
  • L - Number of windows the function analyzed (aka number of feature vectors)
  • M - Number of coefficients (aka number of features in each feature vector)
  • N - Number of channels
Depending on your application, you may want to combine the feature vectors into a single statistical summation by averaging the coffecients, or you may want to feed each feature vector into your system separately.
For example, the following code gives the mean of the coefficients.
coeffs = mean(coeffs,1);
For more information, consult the documentation:
https://www.mathworks.com/help/audio/ref/mfcc.html#mw_7e1e2bdf-548a-45f3-a053-7627b9e8b2b9
Related Question