Colin Raffel

The high-level rule is to try to stick to external conventions, i.e. if in all past datasets/papers/competitions the multipitch time series are uniform, then we can assume that. But if...

Ok, in that case, if the code is broken for non-uniform timescales, we should throw an exception when a non-uniform timescale is passed in.

We could add a separate utility function that converts non-uniform time series to uniform.

This seems like a reasonable solution to me at a high level.

PR welcome!

cc @rabitt @justinsalamon Well, the difference starts with the fact that their annotation format is quite different. multipitch annotations specify, on a fixed timing grid, "what fundamental frequencies are present...

> It could be clarified in the docs, IMHO. I agree. If you or @justinsalamon or @rabitt want to make a PR, it would be welcome. > Especially pros/cons of...

Not sure how I missed this but thanks for flagging it @a-pillay . My only suggestion is just to provide a new argument rather than overloading midi_file since this certainly...

I wouldn't call that trouble - it's a single if statement, and adding a single if statement is better than making an argument's meaning ambiguous, because the arguments are user-facing.

@a-pillay feel free to pick up this PR if you have time.