A Novel Phase-Domain Speech Watermarking Method using Perceptual Orthogonal Matching Pursuit

Abstract

This paper proposes a new blind method for speech watermarking utilizing the perceptual orthogonal matching pursuit (POMP). The employed POMP algorithm, which has been recently introduced to achieve a high-quality sparse representation of speech signals, is combined into a watermarking method in this work. The proposed watermarking approach is carried out as the following. First, an orthogonal dictionary is constituted of gammatone basis functions to be used in the POMP method. Thereafter, the POMP algorithm is executed on short-time speech frames to find the most relevant gammatone kernels and the corresponding coefficients. Finally, the watermark bitstream is embedded into the phase of extracted coefficients and the reconstruction step is performed. By utilizing POMP representation with an orthogonal gammatone dictionary, a high secure speech watermarking scheme has been attained with a high capacity and an adequate robustness against various types of attacks.