Replies: 1 comment
-
Yes, this kind f music is hard to get right for (down-)beat trackers since it contains strong onsets on the offbeats. Thus it often detects double the correct tempo, and in turn the downbeat positions. There's not much you can do to "improve" the algorithm. There's nothing wrong with it. It is just that the neural network has more or less strong activations the wrong positions. This should be solvable with more training data. The other option — as you mentioned — is to give the tracker hints about what to expect, e.g. the known tempo or hint towards known downbeat positions as suggested in #392. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
The beat tracker seems to struggle with music with emphasis on the offbeat.
Consider this popular song: https://www.youtube.com/watch?v=RILP53OR63k. You can hear the guitar/ukulele strumming on the offbeat. This causes
DBNDownBeatTrackingProcessor
to get confused about:Is this something that could be solved with more training data? Or can something be improved about the algorithm?
I noticed that the
DBNDownBeatTrackingProcessor
takes in amin_bpm
/max_bpm
parameter, so I could help it out by providing it the actual bpm, but I wonder if there's anything we can do to make it work more often by default in these types of songs.Beta Was this translation helpful? Give feedback.
All reactions