Digital artefacts are frequently created through mp3 compression–it is not dither noise (I assume).
They are the result of rounding errors and alike. They only stay for a short time which corresponds to the frame size and are isolated peaks in the spectrum. Just like short, unmodulated sine tones.
Here’s a an example that I’ve encoded with a very low bit rate. You will notice that the artefacts disappear after isolating the center (what is common to both sides)
Removing vocals is quite the opposite, all differences are preserved including digital noise.
However, my Remove/isolate 2D Stereo Toolkit does sometimes out-perform the built-in voice removal because it works with 8192 sample chunks that do overlap by 5/7 or so. This can soften those tones a bit. Increasing the frame size (in a text editor) increases this effect too, but the down side is that you’ll get a longer reverb tail.
You should try my tool, it returns the center-free audio in stereo and thus the artefacts are less audible in general.
https://www.dropbox.com/s/tkonxx1njg1lzcu/rjh-stereo-tool.ny
It would be an interesting project to make a “Digital Noise” filter, perhaps one of these days…