Settings to reduce volume instability artifacts in instrumental output

[astrohoff]

I'm using this tool to create instrumental tracks to help me focus during work. But for some songs the output has significant volume instability artifacts that are more distracting than the vocals themselves. I'm looking for suggestions on how to reduce such artifacts.

My current settings are UVR-MDX-NET Inst 3 model with segment size 1024 and overlap 0.75. I'm testing with lossless music files.

I've tried using MDX23C as well as the ensemble model set suggested by sabassa on this thread #444 but that doesn't seem to improve this particular artifact in the songs I am testing with.

The song I've noticed the strongest volume instability on is Need Some Love by Rush. The chorus of that song is particularly bad, it sounds like someone is playing with the volume knob the entire time. I imagine it may be partly due to how the song was mastered (e.g. limiting) but even still I'm wondering if there are models/settings that can compensate for that.