Let's consider the one-dimensional string first. Standard text-book physics considers the three usual boundary conditions here, namely Dirichlet (endpoints of the string are fixed), Neumann (endpoints are free) and Robin conditions (obtained e.g. when the endpoints are attached to a spring).
Now, for real sonic propagation, those boundary conditions won't apply exactly, but serve only as an approximation (as usual in physics). When a sound wave hits an object, it will likely be reflected only partly while a certain component is damped out. Moreover, this damping will likely depend on the frequency, the input direction, the geometry and material of the object and so on. That is a very tough proble in general.
For the first steps, however, it is usually sufficient to use ideal reflection and use zero-Dirichlet boundary conditions. The assumption there is, that when a sound wave hits a wall, it won't lead to any oscillation there, and thus will reflect the wave completely with a phase shift of 180°. Using Dirichlet conditions is also the easiest you can do: In a cartesian grid treatment is suffices to leave the boundary planes out of the propagation.