Skip to Main Content
A novel data hiding technique for robust watermarking is proposed. The technique is given the name "Segment-Difference Classification" (SDC), and is applied on speech signals. Each block of speech is divided into segments, and the weighted sum of each segment is calculated, where a secret set of random weights is used. The segments' weighted-sums are sorted by amplitude. The sign of the difference between the highest and the lowest segment-sums is used to classify the watermark binary value. Selective embedding is done to enhance the robustness of the watermarking by increasing the difference positively for +1 watermark bit value and negatively for -1 watermark bit value. The SDC is compared with spread-spectrum (SS) and scalar quantization-index modulation (QIM) embedding techniques over additive WGN attacks, shuffling attacks and filtering attacks. The SDC shows significant improvement in robustness over the other two techniques for the investigated attacks. The paper also proposes an over-sampling, delta-modulation technique to generate binary watermark sequences from analog watermark signals.