This is my first time working with Mel-frequency cepstral coefficients and I have two questions....

  1. Are some of the 13 coefficients strictly related to some sound properties like frequencies, amplitude, pitch or something ? For example if I'm interested only in sound pitch, are there some coefficients that carry that information and I can exclude the rest of the coefficients that carry other information to reduce the size of the vector ? Or those coefficients means nothing individually and they must be used all together as a whole ?

  2. I'm trying to learn Chord recognition methods. I was thinking for calculating MFCC from a short Wave file where chord is played. (It is just a chord played on piano, there is no ambient noise or other instruments). And after I get the 13 MFCCs, cluster them to a centroid which describes the appropriate chord. (24 centroids, 12 for minor chords, 12 for major). I'm just wondering, are the MFCCs good descriptor for the present pitches in the sound ?

Thanks in advance


MFCC's are related strictly to spectrum shape and what's more - these are relaying on the mel scale. Additionally each coefficient is a DCT value from fitted cosinusoids to log-energies. Thus interpretation of pitch would be troublesome for you.

Chord recognition is usually based on chromagram calculation or so called chroma features. Check out work and publications of guys from QMUL, i.e: Real-time Chord Recognition for live performance - in here you will find another references.

You might also find their software useful: Sonic Visualizer together with lot's of so called Vamp plugins. You can check out the code and even use command line tool to extract features you want.

| improve this answer | |
  • $\begingroup$ Thanks for the answer.... Does this mean that I can use MFCC for musical instrument recognition (if it is a trumpet, organ or violin for example) ? $\endgroup$ – Alexander Aug 12 '14 at 22:10
  • $\begingroup$ @Alexander: Indeed you can use MFCC's for that purpose as a most simplistic approach. Also please keep in mind that you might want to use the same tones for achieving of the best performance. $\endgroup$ – jojek Aug 13 '14 at 7:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.