2

here's an interesting question:

Suppose I have an audio recording of a C major chord (C-E-G) played on the piano, which I would like to separate into three separate audio files - one with only C, one with only E, and one with only G playing (even a single audio file of a single note would suffice). I know software like Melodyne is capable of doing such things but would love for anyone to guide me in the right direction here.

What I've already tried: writing a neural network to do this (but honestly that feels like overkill.. this shouldn't be that challenging), and playing with the STFT and trying to figure something out from there.

Any help would be greatly appreciated!

jason green
  • 33
  • 2
  • 1
    The STFT approach could work. Start with the spectrums for each note individually. Then look at pairs to see how the notes interact. There's also a technique known as a [wavelet transform](https://en.wikipedia.org/wiki/Morlet_wavelet) which may be useful. – user3386109 Jul 14 '22 at 21:59

0 Answers0