Speech Watermarking Technique Based on Singular Spectrum Analysis and Automatic Parameter Estimation using Differential Evolution for Tampering Detection

A singular spectrum analysis-based watermarking scheme is proposed to detect speech signal tampering. The watermark is embedded into the original signal by modifying a part of less-significant singular values of the original signal, and later the extracted watermark is compared with the original watermark to detect the tampering. Differential evolution is deployed to select a part of singular spectrum to be modified to balance between the robustness of the scheme and the sound quality of the watermarked signal. The experimental results show that the proposed method can detect the types and the position of the signal being altered. The performance of the scheme has been improved since the previous methods in terms of inaudibility resulting in the excellent sound quality of the watermarked signal. Because the robustness has to be traded off against inaudibility, the proposed method seems to be not robust. However, the robustness can be improved by revising the cost function.