

@schwa I dunno, man. Claude tells me its code is production quality all the time.
Sure, it has a 30GB memory leak at runtime. But I guess since it was trained on all that Electron code, I guess it’s true?
@steve @hikari Yeah—context is everything. A bag of notes, even when you arrange them by “pitch height”, cannot be reliably named. Rootless voicings, dropping the third, and other such changes to chords in practice make this a fool's errand.
FWIW, chord recognition infers symbols within a fixed vocabulary, and your performance/accuracy is helped if you allow it to say "not in the vocab.” A polyphonic note detector that names the output pitches would be painful to use in practice.
@steve @hikari are we talking chord recognition (from audio) or “suggest some nice chords” tools? Both might share some components (e.g., transition probability matrices), but they are applied in very different ways.
I happen to have more experience on the recognition side, but am happy to try and answer questions about either.
(Edit: on second read, it might be neither! 😂)
Full disclosure: If you signed up recently, (a) you probably need to re-sign-up and/or let me know if you still don't have a confirmation email, and (b) you *may* receive the 015 edition newsletter to your inbox in the next day or so.
I'm still working on the email export in my Middleman project, but I wanted to publish ASAP so that my in-app newsletter links don't still point at a page with a 3-year-old "latest post". 😛