Justin Salamon

@beasteers it's generally hard for me to evaluate / discuss ideas when presented in the form of code, because it requires a lot of time to go over the logic...

Related, I still think the core library shouldn't have any image/plotting dependencies, and their loading can be optional (along the lines of librosa.display).

I'm not sure I fully understand the question, but, CREPE was trained on a combination of speech and other instrument sounds, and is designed to work on any pitched sound...

Thanks for the contribution. However, I believe that if we want to add an audio reading dependency, it should be `soundfile`, which is more mature and supports a wide variety...

OK cool, I'll try to get to this some time soon. I agree about the character limit, 80 seems harsh given today's screens, I like black so happy to use...

That's not a bad thing to have, but it's not a regression test AFAIK. For regression testing we should generate CREPE output data from a small selection of varied content...

@maxrmorrison would it be possible to carve out the relevant viterbi code only, so as to avoid having the entirety of librosa as a (heavy) dependency?

@jongwook what's your justification? The greater the dependency chain, the greater the technical debt (=more maintenance), and the greater the chances of users running into issues during installation. It also...

I'd be in favor of carving out only the needed functionality, but @jongwook it's your final call.

The choice of step size depends entirely on your application/needs, and is just a way of controlling the temporal resolution of the output. Since many audio analysis tasks (e.g. speech...