An Adaptive Audio Watermarking Method Based on Local Audio Feature and Support Vector Regression

Abstract

Based on local audio feature and support vector regression (SVR), an adaptive blind audio watermarking algorithm in wavelet domain is proposed in this paper. The audio signal is partitioned into audio frames, and the watermark is embedded in wavelet domain. For each audio frame, the energy and the maximal peaks of its all sub-bands are extracted as the local features, and SVR is used to model the relationship between the local features and the embedding strength of the audio frame in order to adaptively control the embedding strength of the audio frame. Due to the good learning ability of SVR, the watermark can be correctly extracted under several different attacks. The proposed watermarking method doesn't require the use of the original audio signal. The experimental results show the proposed algorithm is robust to signal processing, such as lossy compression (MP3), filtering, re-sampling and re-quantizing, etc.