Perry Kivolowitz56m agoThank you for noticing!
Energy across frequency bands provided less help than one would think. Detecting the *onset* of instruments that inhabit those bands gave better results. Also, in many long hours of testing and development I stumbled upon one of the tells in AI music... in comparing the dynamics of a stereo signal, AI music uniformly showed less breadth than human produced music. When I started using that kind of signal, human-produced music really showed expressiveness.
Basic getting the beat right in a wide range of music is surprisingly hard. Classical music. Humans hear the beat intrinsically. Computers, not so much. Turns out it is an ongoing area of academic research.
on CadenceViz