A robust watermark embedded into a Direct Stream Digital (DSD) audio signal including a flat frequency response in a specific frequency range which does not extend below 20 kHz or above 100 kHz. The watermark is therefore hidden in the noise spectrum of the DSD signal, such that the watermark is inaudible to a listener. Since the noise spectrum contains important information that helps provide the DSD signals with sharp transients and an accurate impulse response, the watermark cannot be removed from the DSD signal without causing significant degradation to the signal's audio quality.