Can we please send some AI resources to tempo detection? It’s embarrassingly bad and hasn’t improved in a decade.
Curiously I’ve not had such a bad experience. Perhaps your audio is to sparse, or the arrangement with extraneous articulations off-beat is driving tempo detection off mark?
I wouldn’t call it “embarrassingly bad” but I agree it could use some improvements. 
I separate drum percussion stems ( Spectralayers ) and run the tempo detection on the isolated track. It’s OK … but easily fooled!
Is there a better tool to create a tempo click track? Proper question, I dont know the answer.
Thanks
I just do it manually. First set musical mode of all tracks to OFF, make sure the event(s) (or rather the transients) starts on an actual first beat of a bar (if that is the case), and of course make sure that the meter is set properly, and then go into Warp Mode. Set the grid to 1 bar, and then, while in Warp Mode, move all the following barlines manually while looking at the transients or MIDI notes, to get a correct tempo map. If there are meter changes along the way, I just pencil them into the meter track. Then turn musical mode for all tracks ON again. Dom Sigalas explains this in a (Cubase) video somewhere.
I’ve tried automatic tempo detection sometimes, but I have found that it sort of works for parts of a song or musical sequence, but then some hiccup causes it to stray off the path. So I always end up doing it manually. I have pretty strong opinions on how I want my tempo map to be, so I wouldn’t really trust automatic detection anyway – at least for now.
1 Like
I do it manually most of the time too. I’m a stickler for details. Still, I would like Steinberg to improve tempo detection, BUT considering the other things that I want them to spend resources on too, I don’t consider it a top priority. Also, how often do I need to construct a tempo map, compared to how often do I need some other feature that I want them to develop?