In Wwise, the term "spatial audio" refers to a variety of settings and configurations that relate to sound in simulated 3D environments, such as attenuation, diffraction, reflection, and so on. There are many aspects of spatial audio that require CPU processing power so you will have to test different options and decide on the appropriate balance between quality and performance.
In Wwise Authoring, many of the options are available on the Positioning tab in the Property Editor (refer to Positioning Tab: Actor-Mixer and Interactive Music Objects and Positioning Tab: Audio and Auxiliary Busses).
Spatial audio uses CPU processing power for numerous calculations related to emitter and listener positioning, sound attenuation, sound reflection off of 3D objects, diffraction, and so on. There are many possible combinations of spatial audio Effects and settings, so there are no simple guidelines for CPU optimization. However, the following suggestions might be helpful if you notice that spatial audio is consuming a lot of CPU resources in the Profiler:
Ensure that the spatial geometry is as simple as possible. Minimize the number of triangles and the number of diffraction edges.
Use finite attenuations for sounds. As the maximum attenuation radius increases, more CPU processing power is required to search for paths. When a finite attenuation curve reaches its maximum distance, the curve drops below the audibility threshold set in the project settings.
Avoid playing sounds if they are too far away to be heard, and stop them when they move beyond the attenuation radius and will not return.
Use rooms and portals to isolate independent sections of a scene.
When initializing spatial audio, you can set several Spatial Audio Initialization Settings to control CPU usage:
Use CPU load balancing.
Increase the movement threshold required to calculate reflections and diffraction, which reduces CPU usage at the cost of accuracy.
Reduce the number of primary rays used in the raytracing engine.
Reduce the maximum reflection order. First-order and second-order are sufficient in most cases.
Reduce the diffraction order. Start with 4, but even 2 or 1 can be usable.
For more information on spatial audio, including suggestions on CPU optimization, refer to the following documents: