Hey mayhem happy new year! Brief update on the Essentia benchmarking - I'm still finishing up the analysis but what I'm seeing right now is very promising. The traditional onset detection approach is kind of doing what I expected - when it's off it's usually off by a factor of 2 or so because it's struggling to determine which onsets are the appropriate fundamental period of the track. That's because this is generally very dependent
2025-01-03 00315, 2025
igold94[m]
on the time signature of the track. The deep-learning approach implemented by essentia uses the spectrogram for the entire track and classifies to one of 256 possible BPMs (ie: all whole number values). Because it has the full context of the track as it makes its decision it's often way more spot on in these edge case situations. Waiting to get the full tally but it seems more accurate than spotify's data ATM
2025-01-03 00345, 2025
monkey[m] has quit
2025-01-03 00349, 2025
BrainzGit
[listenbrainz-server] 14Suvid-Singhal opened pull request #3110 (03master…ui-improvement-tooltip): UI Improvement: Improved tooltip styling in settings page https://github.com/metabrainz/listenbrainz-server…
