Audio watermark encoding methods employ reversing polarity and pairwise embedding. A watermark signal is generated, mapped to pairs of embedding locations and inserted in the members of the pair with reverse polarity. The pairs of embedding locations correspond to adjacent regions or frames in time or frequency domains. A method for controlling an audio output device captures audio from the device, extracts its identity from a watermark, and modifies the operation of the device, e.g. varying watermark strength or audio loudness.