This sounds like a great use of hardware accelerated raytracing. The problem itself is very similar. Sound attenuation can be modelled as transparency.