At the moment I've got a script that generates a bunch of bars, and bakes a narrow range of frequencies to each bar's Z scale. It ends up being a pretty simple spectrogram that I composite onto the bottom of the video.
The only problem is that there's a lot of "bleeding" across frequencies. It looks like the same kind of thing you see on a spectrogram if you set the window to something really small.
If I have an audio file with a 440Hz tone, you can bake frequencies as low as 100Hz or has high as 1100Hz and still be able to see a little of the 440Hz tone influencing it, and there's a lot of really significant interference down until around 200Hz and up to 700-800Hz. Is there any way to make blender more accurate when baking sound to F-Curves?