Brian McFee

@lostanlen 's point above does make me wonder a bit about how we should define chroma features for JI-based representations. I can think of two ways to do this (there...

> What is the appeal for (2) in ET? specifically, between `bpo=36; n_chroma=12` versus `bpo=n_chroma=12`? > Is it meant to have longer CQT atoms in the time domain and thus...

> In light of what you're saying, i would be in favor of disallowing (2) for JI-based chroma (`n_chroma 1`. Yeah, it does seem like the simplest solution here. It...

Coming back to this, I was thinking a bit about how far the crystal algorithm gets us toward the JI ratios listed here: https://github.com/librosa/librosa/files/1466980/srutis.pdf as mentioned in #641 discussion. The...

Following up on this, I got in touch with Marc Sabat, who provided the C++ code from the 2007 paper. After digging through it and comparing to my implementation above,...

Aside, here's a one-liner to use sympy to map floats back to simplified ratios: ```python frac = lambda x : sympy.nsimplify(np.array(x), rational=True, full=True, tolerance=1e-3) ``` It doesn't explicitly let you...

I've been doing a bit more hacking on this, trying to build up some intuition for how the crystal growth algorithm actually works in practice. A couple of observations below....

> Historically speaking 7/4 was most likely to be treated as an augmented 6th and today it's used largely as that or a minor seventh so I would personally be...

Circling back on this, I think we're mostly settled on the frequency generation and dsp aspects of this. The remaining problems (in my head, at least) have to do with...

Cool - it's not urgent, so I'm happy to wait.