The dither code was broken in Shairport Sync and also less than ideal anyway. Fixed and improved, but it presents a slightly heavier processing load.
Dither is added whenever you use the software volume control at less than full volume. See http://www.ece.rochester.edu/courses/ECE472/resources/Papers/Lipshitz_1992.pdf for a very influential paper by Lipshitz, Wannamaker and Vanderkooy, 1992. The dither code in Shairport Sync was inherited from Shairport and does not conform to the recommendations in the paper -- specifically the implementation would give one bit of dither where the paper recommends two bits peak-to-peak. The other thing is that the inherited dither code was actually broken in Shairport Sync. So, the new dither code gives a two bit peak-to-peak dither based on a Triangular Probability Distributing Function (TPDF). It sounds like a very low-level white noise, unmodulated by the audio material. It would be nice if it was even lower, but it's better than listening to the artifacts present when dithering is disabled.