Pre-processing modules are configured to compensate for time and pitch scaling and shifting and provide compensated audio frames to a watermark detector. Audio frames are adjusted for time stretching and shrinking and for pitch shifting. Detection metrics are evaluated to identify candidates to a watermark detector. Various schemes are also detailed for tracking modifications made to audio stems mixed into audio tracks, and for accessing a history of modifications for facilitating identification of audio stems and audio tracks comprised of stems. Various approaches address interference from audio overlays added to channels of audio after embedding. One approach applies informed embedding based on phase differences between corresponding components of the channels. A detector extracts the watermark payload effectively from either additive or subtractive combination of the channels because the informed embedding ensures that the watermark survives both types of processing. Other approaches applies different polarity patterns, watermark mappings, or protocol keys to the channels. These techniques enable the watermark to survive ambient mixing, conversion to mono, as well as channel differencing to reduce interference from voice-overs and other audio overlays.