Original video name: Answer to @hate5six channel about the algorithm for changing cameras
This is the first video th-cam.com/video/vNET8XRbH7c/w-d-xo.html
Librosa is a python library for Music Information Retrieval. It's maintained by Brian McFee, you can check him talking about it in this video: th-cam.com/video/MhOdbtPhbLU/w-d-xo.html
Librosa allows you to get the time instants of the beats as an float array. The documentation is right here: librosa.github.io/librosa/generated/librosa.core.clicks.html
I think this approach is better than trying ML algorithms, because it's a really really simple turnaround. I don't know exactly how we could train a ML model for this.