At this layer of note information, fact is the data will be either sent to a MIDI device or a plugin as a MIDI event, so you might as well keep it in the native format and not convert to a frequency.
Microtonal music, or even just plain regular traditional music that's tempered to be more in tune with itself, should be trivially easy to do with electronic instruments, and yet due to some historical decisions to base the one near-universal music protocol we've been using for the last 40 years on a piano-centric representation, it's a lot harder than it ought to be.
I think we should not disregard microtonal music just because it isn't cleanly backwards compatible with a protocol that's over 40 years old.
Here's a pretty comprehensive-looking list of software instruments supporting microtuning:
I think it's not sent over MIDI so it only works if everything is running on the same computer.
In fact, the VST2 and VST3 SDKs support microtonal offsets per note event (measured in cent). Unfortunately, only few plugins and hosts support it.
This doesn't need to be a trivially human-readable format, but I think maybe it would be nice to allow for multiple pitch representations: floating point frequency, floating point fractions of a semitone in any arbitrary EDO, fractional frequency as used in just intonation, etc..
You could also allow for annotations, like maybe a note pitch is represented as "63.86" in 12-EDO, but it's not just some weird in-between note, it's actually the E that's 5/4 above C, so the file could have an annotation that says that's what the note actually means in this context.
I've heard the recent midi 2 can get around this but it's still a pain with the 40 years of midi equipment that surrounds us.
One way that works for multitimbral synth (which a lot of 90's romplers are) is to only play one note at a time on each channel, and use multiple channels for polyphony. This allows you to individually control the pitch of every note.
The one-note-per-channel trick works okay, but it's kind of awkward. If you want to use all 16 channels, it means setting every channel to the exact same patch, which is tedious. Also, you have to know what the pitch bend range of the synth is if you want to bend by the exact right amount. So, eventually MPE was adopted as an official MIDI standard to provide an easier, more user-friendly and standardized way to do one-note-per-channel MIDI.
MIDI 2.0 has added per-note pitch bend. For anyone making microtonal instruments right now, though, I'd say MPE is probably the best/easiest option.
If you want to make changes, you give ChatGPT the MP3 file and say "switch out the trumpet for a piano and make the singer a guy".
Music however.. i adore music. A big (but not essential) thing i love about music is the story that got the artist to that point. Pain, joy, emotion is transferred. Yea, often it's not essential so maybe AI Music has a place in my life, but easily 60% of my music is loved partly, if not heavily, because of the emotions that got the artist to that point.
I too am not interested in outsourcing this.
Like midi note transcription or waveform modeling will be cool machine learning tools. An end to end composition and mastering bot may make something passable like other current derivative music out there, being generous by saying passable.
Show me the leaps and bounds in self-driving cars that were only 3-5 years away 8 years ago.
There is still need for better daw, music transcription, and audio file formats in the next century. Lol humans want to make art in their free time not play around with pretend chatbot personalities.
P.s. people still record to TAPE and it sounds awesome even though there are albleton plugins