Incorporating Spectral Shaping Filtering into DWT-Based Vector Modulation to Improve Blind Audio Watermarking

Abstract

A spectral shaping technique emerging from autoregressive modeling is incorporated into vector modulation to achieve efficient blind audio watermarking. This technique allows the watermarking process to be performed in a broader frequency band with the embedding strength adapting to auditory masking thresholds. To ensure accurate watermark retrieval, we slacken the condition for binary embedding and develop an iterative algorithm to carry out energy-balanced vector modulation. As a result, the proposed scheme reaches a capacity as high as 818.26 bits per second but still possesses sufficient robustness and transparency. The effectiveness of the proposed scheme has been demonstrated using the perceptual evaluation of audio quality (PEAQ) and bit error rates of recovered watermarks. The PEAQ confirms that the watermarked audio signal is perceptually indistinguishable from the original one. Compared with other recently developed DWT-based methods with less payload capacities, the proposed scheme can achieve comparable, if not better, robustness for attacks such as resampling, requantization, amplitude scaling, noise corruption, lowpass filtering, DA/AD conversion, echo addition, jittering and MPEG-3 compression.