Wwise Vorbis files seeking is limited to the seek table's granularity, specified in the Vorbis Encoder Parameters of the conversion settings (Edit... button). The target sample position is rounded to the closest seek position. This is true for both streamed and in-memory (soundbank) files.
The consequences of this can be noticed with "From Elapsed Time" virtual voices, Interactive Music and Seek actions.
"From Elapsed Time" virtual voices
When a "From Elapsed Time" virtual voice sound is virtual and becomes physical, playback restarts at the appropriate file position. In the case of Vorbis files, it actually starts at the closest seek point. If you were listening to marker notifications, it is possible that the same markers be notified again if the selected seek point is behind these markers.
Interactive music
Music segments playback can start anywhere depending on the transition rules. The audio clips it contains may thus be seeked to the appropriate sample. Again, in the case of Vorbis, the resulting seek position depends on its seek table. However, the interactive music system requires that seeking be sample-accurate. In order to make up for Vorbis seeking granularity, the sound engine may have to decompress and flush several chunks of data in order to start playback at the correct position. This may result in a small CPU spike. It is recommended to use finer seek table blocks for Vorbis sounds used inside music segments, if they are susceptible of being seeked into.
"From Elapsed Time" virtual voices are processed sample-accurately when these voices are part of the interactive music system. Thus, the comment above also applies for interactive music virtual voices.
Seek action
Seek actions are sample-accurate: the sound engine makes up for Vorbis seeking granularity, like it does when seeking a voice of the interactive music system.
