The guy who solved voice recognition at Google years ago told me that after looking at enough spectrograms with human speech he could tell exactly what was being said. "Oh, that shape- it's an 'ooo' noise."
Armed with that knowledge I started looking at more spectrograms of birds (I record birds around my house and pass through bird.net's model. Amusingly, I kind of learned eveyrthing in this article just by looking at lots of spectrograms- they are far, far easier for me to parse than any other musical representation.
I just found 'demucs' and am really enjoying doing source separation and then looking at the resulting spectrograms, then back at the original full musical spectrogram- it helps me parse out the instruments a lot better when they overlap.
There is also a product called 'Celemony DNA' which can do some editing in spectrogram space to remove individual notes.
Armed with that knowledge I started looking at more spectrograms of birds (I record birds around my house and pass through bird.net's model. Amusingly, I kind of learned eveyrthing in this article just by looking at lots of spectrograms- they are far, far easier for me to parse than any other musical representation.
I just found 'demucs' and am really enjoying doing source separation and then looking at the resulting spectrograms, then back at the original full musical spectrogram- it helps me parse out the instruments a lot better when they overlap.
There is also a product called 'Celemony DNA' which can do some editing in spectrogram space to remove individual notes.