Speech watermarking robust against recapturing and de-synchronization attacks

Abstract

Watermarking is a solution for forensics tracking for digital speech signal, while recapturing and de-synchronization attacks may remove the watermark effectively. It brings threat to watermarking technology used for forensics tracking. In this regards, we defined the discrete cosine transform coefficients logarithm mean (DCT-CLM) feature of speech signal and concluded that the interference caused by recapturing attack has similar impact on the DCT-CLM feature of two adjacent segments. Based on the feature, we proposed a speech watermarking method robust against recapturing and de-synchronization attacks. For the mute parts of speech signal have no meaning and are not the major component for forensics tracking, we focused on non-silence (NS) parts in this paper and embed watermark into DCT-CLM feature belonging to these parts. Compared with the state-of-the-art works, the BER value of the proposed method is decreased by 9% for recapturing attack. Besides, the method outperforms the recent robust watermarking methods notably for recapturing and post-processed with signal processing operations and de-synchronization attacks.