An Audio Watermark Designed for Efficient and Robust Resynchronization After Analog Playback

摘要

We propose a spread spectrum (SS) audio watermark designed to withstand analog playback, including desynchronization caused by small differences between playback and recording rates. Desynchronization robustness relies on detecting short blocks of a magnitude-only watermark embedded in the frequency domain where the resolution of the SS chips can be reduced. Lost spreading gain due to the lower number of SS chips is compensated using blind dynamic time warping (DTW) detection (does not access original signal). DTW aligns sequences of blocks to improve robustness to interference while mitigating the vulnerabilities of long SS sequences to desynchronization. Results demonstrate that the proposed watermark survives analog playback and warping up to ±2%. Additionally, compared with a recent baseline scheme that uses brute force resampling to search for resynchronization, the proposed watermark is 300 times more computationally efficient, and does not compromise robustness to either desynchronization (e.g., jitter, resampling, time warping, frequency scaling) or non-desynchronizing modifications (e.g., AAC compression and additive noise).