To me, those three terms mean similar things, but with varying degrees of specificity.
Balanced to me means
generally even or
safe; no cardinal sins, no funny business. It's the most malleable of the three, because - and this is why I personally don't believe in a specific "target" curve to define balance or accuracy - you can actually push or dip frequencies quite a bit before they become
unbalanced. As Jude said on his IE900 review,
balanced (and he even extends it to
reference) represents a host of curves, rather than a rigid, target one.
If you have Michael Bublé singing through a generally-flat headphone, for example, and you boost the low-end a couple dB, his voice will still technically sound correct and
balanced. The human singing voice can change quite a bit from one day to another, which is why vocal takes are generally done in one session. So, little variations here-and-there won't trigger that
incorrect or
unbalanced receptor in our head. So, for me,
balanced headphones or IEMs are ones that you instantly accept as "within the ballpark of
correct" without any wild dips or peaks that catch your ear. And, as always,
correct for me doesn't align with any specific curve. It's just that barometer drilled into our heads from hearing instruments and sounds everyday.
Now,
reference gets a lot more specific. I think a lot of you know by now my personal definition of
reference. A
reference IEM should not add any colour to the track, so the changes between one track to another can be as transparent (or obvious) as possible. If I play 10 different tracks and I hear the same lift in the bass, or brassy-ness to the vocals, or glare in the treble, etc., then that IEM isn't a
reference IEM. And, for an IEM to be
reference, all of its parts have to be
neutral, which, to me, has the same meaning as
reference, but applied to the individual parts of an in-ear, i.e. its lows, mids, highs, imaging, dynamics, etc. A monitor with a neutral low-end has a bass that complies with the tracks its playing, but if the mids and highs aren't similarly neutral, then it isn't a reference IEM.
Contrary to popular belief, it applies to imaging and dynamics just as well. The only experience I've had with a strictly-Harman-tuned IEM would be Moondrop's Starfield, which Ryan kindly gave me simply for that. I found it to match a lot of what I described in the
balanced bit, where it was generally safe and pleasant across genres. But, it didn't have the resolution, space or dynamic range to be considered
reference. Those categories weren't
neutral to me, because the Starfield could not capture the full scope of the tracks I was playing most of the time. On the other hand, though, you can also have in-ears like the ODIN, which
always sound
big and
grand, or the EVE20, which sounds giddy and excited
all of the time. An ideal
reference IEM should grow and shrink - spatially and dynamically - with the track. So, again, it has to get all those qualities right in order to be given that title, so to speak.
In reality, I don't believe there's a fully-reference IEM out there right now, especially once you consider personal biases as well. One of my references, for example, is 64 Audio's A18s, which has a slight softness to its treble and a slightly elevated mid-bass. I don't mind the former, and I actually need the latter when I'm getting fatigued and I need the pump the volume a bit higher. So, there are an infinite amount of scenarios that an IEM has to fulfil to be considered fully-reference, thus the idea is to just find one as close to it as possible for the most amount of time. At least, that's the conclusion I came to a while ago.
I think, regardless of what IEM you play your music through, pitch nowadays is
extremely muddy waters anyway. I don't know whether or not you've seen a lot of the videos about it online, but 99% of the music we listen to these days is technically off-pitch. Most music use a system called equal temperament, which splits the octave (from C1 to C2, for example) into a set of notes with equal intervals. This is the principle by which pianos are tuned, and how pianos in music production software is tuned as well. The issue is, notes don't naturally have equal intervals. So, if you take, say, the A major triad (A, C, E) and the D major triad (D, F#, A), the A note in both chords should have different frequencies. Here's a compilation of Jacob Collier explaining it online:
He, in particular, is someone who takes advantage of this to create lots of microtonal contrasts in his music. I believe his track,
Hideaway, starts and ends with two different versions of the A chord. He starts in the "less accurate" equal temperament to create a feeling of something familiar, but
slightly off, then ends it with a harmonically-correct version to create a sensation of safety and homecoming. He does this in his stunning,
stunning arrangement of
Moon River as well, where he goes up and down by intervals that are less than a semitone. So, rather than going from G to G#, he goes from G, to G half-#, then to G#. Again, it's stunning stuff.
It's also why, if you look at the basses of Henrik Linder from Dirty Loops, you'll see lots of squiggly lines on his fretboard, rather than straight, equally-divided ones:
Funnily enough, this goes back to what I said about a
balanced tuning. It's not just our in-ears that can be pushed a couple dB's here or a couple dB's there and still sound correct; music
itself can be pushed cents here-and-there without anyone batting an eye!
So, yeah, if you wanna get
really in-depth and mull over these sorts of things, there's a literal
sea out there left to explore.