1

I need to classify music (songs) into genres (rock, french house, trash metal, etc). My idea was to extract features from the songs (bmp, zero crossing, etc) and then apply known classification algorithms. I want to start with something simple and then improve from there if needed.

What features are important for songs genre classification? What are the best scalable tools to extract those?

Marsellus Wallace
  • 155
  • 8
  • 2
    These authors circumvent the need for extracting features by clustering compressed versions of the music files. Not exactly what you wanted, but pretty neat: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.111.7971 – bdeonovic Feb 04 '14 at 18:04
  • 2
    I'm not sure this should be categorized as a Statistics question. What features are important to music classification requires expert knowledge of what makes different music different. You should choose quantifiable features that allow you to distinguish one genre from the next (obviously), but you would be better off asking a musician that. – Underminer Feb 04 '14 at 21:04
  • 1
    Thanks Benjamin, I'll take a look! @Underminer A musician might not know about the importance of 'Mel-frequencies cepstrum coefficients' (just to name one) for songs classification... Also, I'm asking about audio feature extraction tools as well. – Marsellus Wallace Feb 04 '14 at 21:31
  • 1
    You'll be surprised, but simple BPM + spectral shape is pretty good at genre guessing. I mean, it won't distinguish some of rap, triphop and rnb (eh, I wouldn't be able to do it that easy), but still... Otherwise, just browse google for music analysis software to find the keywords. Which of those are characteristic of music genre is something your model should tell, not you. – sashkello Feb 04 '14 at 22:03

0 Answers0