In most cases the listener is positioned at either the player or camera, or at some (possibly parametrized) interpolation between the two. The sound changes based on the proximity of the listener to the NPC, and the attenuation provides the falloff properties. This scenario is widely used in 3D games and well understood in game audio. However, choosing where to implement the listener in the game may require special considerations for player sounds.
While some situations may benefit from being played back in 2D on the player (such as surround magic effects), there are potential pitfalls that could cause undesirable results, specifically, in the common scenario where the camera moves through the world to highlight action taking place elsewhere in the environment. If a sound playing from the position of the player is set to 2D when this happens, the sound will persist even though the player is no longer part of the visual scene. In this way it's usually safest to continue treating the player attenuations in the same way we've begun with NPCs, as 3D sounds.
While you may start out sharing several of the same attenuations, a unique attenuation ShareSet can always be applied at any level of the hierarchy. Additionally, you can create a custom attenuation that will apply only to a single object. Changing the mode of a sound objects attenuation from Use ShareSets to Define Custom applies a unique instance of the attenuation that is not shared (or sharable) with any other sound object.
Taking an attenuation ShareSet and defining custom properties for a sound object
While defining custom attenuations can be a good solution as an exception to the rule, there is more power and transparency to creating ShareSets which are all managed in one place.
Once you're gotten the feel for the way sound propagates in the game, you'll find yourself instinctively using attenuation ShareSets you've created for new sounds. By establishing well defined and flexible attenuations, a distance falloff model can be created that suits the game.