I recorded the same piece of music from different instruments and I want to compare their timbre by means of feature extraction. As you can see from the image, the recording is composed of different sections with very different amplitude values. Furthermore, the recordings present a slightly different RMS energy.

I am going to divide the recording into short frames and extract the features from each frame. I want the instruments to be comparable. For this reason I performed a RMS normalization on the whole recording before windowing and feature extraction, in such a way that each recording has a RMS value equal to 1. Am I right, or is there a better approach? enter image description here

  • $\begingroup$ if the above plot contains all the instruments at "known" intrevals. Then it is better to apply a single normalisation to each instrument on their own whole interval of existance.. But the plot already is quite fine what is the problem ? $\endgroup$ – Fat32 Jan 11 '16 at 14:52
  • $\begingroup$ The above plot refers to the performance of a single instrument. Other instruments will have similar plots but with slightly different amplitudes. $\endgroup$ – firion Jan 11 '16 at 14:57
  • $\begingroup$ so why are you normalising their powers ? $\endgroup$ – Fat32 Jan 11 '16 at 15:10
  • $\begingroup$ Because some features are not invariant to amplitude changes, so in order to compare them a normalization is needed. I used RMS normalization because that's what is typically done in literature, but since there is a section in the performance with a very different RMS (wrt. other sections) I would like to know if this could be a problem or not. $\endgroup$ – firion Jan 12 '16 at 9:44

Your Answer

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

Browse other questions tagged or ask your own question.