In recent weeks the HoverDerby broadcast booth has had nearly total sound isolation from the promenade below, where everyone else is talking during a show. The one exception was in one spot where our camera avatar would stand. It has been a vexing mystery. We finally figured out why. Our observations may help you, too.
Yesterday Torley explained one feature of the sound occlusion algorithm that is critical to understand. When trying to determine how a sound will be affected by the surroundings, the algorithm looks only at the collision volume nearest your ears and not any others that are farther away.
In our case, the collision volume nearest the camera avatar's ears was a simple 2D "camera pad" that the av stands on. Sound would naturally leak around that camera pad if it were the only collision volume. The totally sealed envelope of collision volume around the camera pad gets completely ignored.
The solution was simple: take away the camera pad and have the camera avatar rest directly on the sound envelope.
In our particular case, a nearby desk is still causing the same problem. We are going to experiment with ways to deal with that.
I hope this observation helps you figure out your own audio occlusion solutions. The key observation is that only the nearest collision volume matters. (This speeds up the realtime algorithm, by the way.) So make sure your nearest collision volume (and its audio material selection) is the one you intend to have the biggest impact on how a visitor experiences your scene.