The title and description strike me as two different problems. Which is it?
The title has a quick answer:
I've had the same problem as a live-performance Audio Engineer. The talent had some cheap in-ear monitors, which are just glorified earbuds (actually I think his *were* off-the-shelf earbuds), and it took me a while to figure out why I could hear the click track (metronome) in the PA after I double- and triple-checked my settings to make sure that I was not sending that input there at all. Turns out it was bleeding from his 'buds into his singing mic!
The answer, then, was to A) turn the 'buds down, or B) turn the click off, or C) live with the click being audible to the audience. I don't remember what we actually did. If that's your problem, then that's your answer too.
The description probably has a similar answer (turn something down) but for a vastly different reason. That reason comes up more for analog gear than it does for digital, but it does still appear sometimes for digital: headroom.
In analog gear (which includes mics and headphones), and in digital integer formats (I believe OBS is 16-bit integer, same as a CD), you have a hard maximum volume. If you try to exceed that max, it shaves off the peaks of the wave so that it never goes above that level. Of course that changes the signal in a way that (most of the time) sounds bad. (it's a fundamental part of the classic "fuzzy guitar" sound though, so it's not *always* bad) The technical term for that is "clipping", while less technical people often call it several things, including "peaking" because of a badly-labeled light somewhere.
The difference between the level that you're using and the maximum level is called "headroom", and the way to get a decent level at the end without ever abusing it in the middle somewhere is called "gain structure".
In a perfect world, you can just combine all of the gain controls from the mic, through all of the processing, to the speaker, and get a single number. It doesn't matter what the individual settings are, as long as they all combine to the right result.
But in the real world, every point in the chain has a maximum signal level and a noise floor. So you have to manage all of those individual gains so that you have the correct level at every point in between, not just at the end. If you get it wrong somewhere, then nothing you do after that point can recover. You have to fix the problem itself. That management is called "gain structure".