Brian McFee

Ahh, I see. @lostanlen is correct - this is not copypasta, but instead an exception being raised one layer deeper in the call stack than the user might be expecting....

Thinking through the beat tracking implementation a bit, I'll likely make the following changes: 1. Convert the dynamic program to populate a boolean array instead of position information. After the...

While working on the beat tracker, I'm taking some time to clean up some internal API and comments to be more useful and transparent.

Circling back on a very old issue #44 (closed back in 2019 when we merged PLP) while my head is back in this code. And yes, I recognize that it's...

Update on beat tracking before i break on this for a bit. The multichannel extension / refactor is almost done. I'm running into some numba dispatching problems though, as it's...

This still needs testing, but multichannel beat tracking is now functional. With jit optimizations, we ended up being just marginally faster than the old implementation, so it's possible that there...

:laughing: Rewriting the beat tracker has revealed an off-by-one error that has been trimming one too many beats since ... forever! Here's the problem. The current (main branch) implementation does...

I think aside from linting and final cleanups, this is ready for CR. One thing that we should discuss though is how we want to handle user-provided tempo in multichannel...

In pinning down the vectorized tempo API, it's probably worth thinking about the (hypothetical) variable-tempo case as well. Mainly I'm trying to avoid painting ourselves into a corner here in...

Remaining TODOs: - [x] Add docstring examples for sparse / multichannel mode - [x] Consider refactoring the beat tracker dp inner loop to explicitly unroll the vectorized cumulative score calculation...