Audio prewhitening based on polynomial filtering for optimal watermark detection

Watermark detection algorithms in the spread-spectrum audio watermarking systems would have optimal performance if host audio had properties of additive white Gaussian noise. However, statistical properties of the host audio signal are generally different from properties of AWGN. Therefore, there is a need for decorrelation of watermarked audio in order to achieve optimal detection based on correlation. In this paper, we describe whitening procedure for audio using Savitzky-Golay FIR filter, based on polynomial fitting of data. Residual signal has more Gaussian-like distribution and significantly smaller variance compared to the case of unprocessed watermarked audio, which was verified using two hypothesis tests for data distribution normality. The procedure considerably improves detection results and has higher resistance to various watermark attacks in comparison with standard correlation detection.

[1]  N. Cvejic,et al.  IMPROVING AUDIO WATERMARKING SCHEME USING PSYCHOACOUSTIC WATERMARK FILTERING , 2022 .

[2]  Darryl J. Downing,et al.  Univariate Tests for Time Series Models , 1993 .

[3]  Hongseok Kim.  Stochastic model based audio watermark and whitening filter for improved detection , 2000, 2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100).

[4]  Darko Kirovski,et al.  Robust Covert Communication over a Public Audio Channel Using Spread Spectrum , 2001, Information Hiding.

[5]  Ahmed H. Tewfik,et al.  Robust audio watermarking using perceptual masking , 1998, Signal Process..

[6]  David J. Groggel,et al.  Practical Nonparametric Statistics , 2000, Technometrics.

[7]  Ioannis Pitas,et al.  Robust audio watermarking in the time domain , 1998, 9th European Signal Processing Conference (EUSIPCO 1998).