Robust Audio Watermarking Using Perceptual Masking

We present a watermarking procedure to embed copyright protection into digital audio by directly modifying the audio samples. Our audio-dependent watermarking procedure directly exploits temporal and frequency perceptual masking to guarantee that the embedded watermark is inaudible and robust. The watermark is constructed by breaking each audio clip into smaller segments and adding a perceptually shaped pseudo-random sequence. The noise-like watermark is statistically undetectable to prevent unauthorized removal. Furthermore, the author representation we introduce resolves the deadlock problem. We also introduce the notion of a dual watermark: one which uses the original signal during detection and one which does not. We show that the dual watermarking approach together with the procedure that we use to derive the watermarks effectively solves the deadlock problem. We also demonstrate the robustness of that watermarking procedure to audio degradations and distortions, e.g., those that result from colored noise, MPEG coding, multiple watermarks, and temporal resampling.