[dodgrile]

I don't think finding individual notes / the inherent notes within a chord is likely that difficult. The problem with this sort of thing is the nuance that's really involved - a player with a decent ear will be able to tell if the same A note is played on an open string, on the lower, fatter strings, or further up the neck. With chords, you also need to start considering voice leading (where you're specifically picking chord inversions for melody, which on guitar will affect how / where you play a chord) and how the mechanics fit together overall (I probably don't want to play an open C chord followed by something on the 11th fret, for example).

[tyingq]

Aren't there a fair amount of songs where the guitar is in some non-standard tuning or played with a kapo, etc? I imagine a naive ML approach would spit out some tabs that would be difficult to play without that context.

[rtsil]

In addition to that, the same set of notes can represent different chords depending on the context and the chord's function, so instead of just picking up notes and recognizing them, the software has to actually understand the chord progression and the harmony.

[gerbilly]

Yeah exactly what I was going to say, you have to consider the chord voicing, the strum pattern, and whether the sequence of chords makes sense for a human to comfortably play.

Someone above mentioned: "Here comes the sun", which is a great example.

First gotcha, it sort of sounds finger picked, but it's not it's flat picked. Another unusual thing about that song is that it's played w/ a capo on 7th fret.

Now I'm sure some software could figure out the individual notes, but I wouldn't be surprised if it transcribed it as nonsense sequence of notes all over on the g, b and e strings, instead of arpeggiated chords as it really is played.

[omarhaneef]

Tell me you have a good ear without telling me you have a good ear.