gemrb
Audio occlusion
Apparently the games supported audio occlusion, some sort of naive muffling that was used together with roll-off on supported cards.
To quote @skellytz, who is actively researching it:
The original BG2 supported EAX 2.0 and made use of its occlusion effects. Back when we used compatible sound cards, many sounds farther away from the center of the screen were muffled like this: Attached File bg2-occlusion.ogg
This effect is different from environmental presets, reverb, or fading. At the time, most other games used occlusion for physical obstacles like walls or underwater effects (example). BG2 used it for sound sources farther away from the center of the screen, even without any obstacles, which I felt was kind of an unpolished feature, to be honest. Still, it did process sounds farther away better than without occlusion enabled.
Looks somewhat difficult to model this for an iso perspective without incurring other oddities.
Simply taking the distance without having any obstacle to take into account (as in the example) isn't exactly natural sound, at least not for distances we are looking at here.
Taking the center of the screen again, and using in-line-of-sight to any source as a condition could possibly result in weird audio when PCs and sources are scattered around the screen, also when thinking about the many caves and scrolling over no-go areas on such maps (we may be in the void, so what do we hear, nothing, everything, ...?).
The best thing maybe is to check whether none of the PCs has a LOS of at least screen size distance, or maybe more, and then considering the source as muffled, e. g. for being behind a door in a room that currently nobody is in. But that's also modelling sound that is inaccurate when everybody is just around the corner, although I think I've heard this kind of implementation in many games.
Something that could mildly help here without doing so much math on our own is OpenAL's directional source and cone feature, e. g. specifying the orientation's 120° cone of a caster which should made the sound muffle when facing the edge of the screen, though I haven't tried it out as far as I can remember.
While we could experiment with using doors, wallgroups and/or the searchmap (for something closer to the base concept), I think it's easiest to just wait for more information.
A simple, physically oversimplified but cheap and forgiving approach could be to check whether the source is below fog or in an undiscovered area, since this could simply match the impression of not having anything in sight.