Title:
An Audio Watermarking Scheme Based on Singular-Spectrum Analysis

Speaker:
KARNJANA, Jessada (JAIST)

Abstract:
In this work, we proposes a blind audio watermarking scheme based on singular-spectrum analysis (SSA) which relates to several techniques based on singular value decomposition (SVD). SSA is used to decompose a signal into several additive oscillatory components where each component represents a simple oscillatory mode. The proposed scheme embedded a watermark into a host signal by modifying scaling factors of certain components of the signal. Besides, a parameter set for the modification, i.e. the parameters specifying the modified components and their new value, is determined by differential evolution, which is a parallel direct search method that optimizes the parameters by iteratively improving candidate parameters with regard to an objective function and some constraints. We design our differential-evolution model to find the parameter set that minimizes bit error rate (BER) and sound-quality distortion. Test results show that, with predefined-parameter scheme, i.e. without employing the differential evolution, the proposed scheme satisfies imperceptibility criterion with the average ODG of 0.18. It is robust against many attacks, such as MP3 and MP4 compression, band-pass filtering, and re-sampling. Our average BER is better than that of the conventional SVD-based method. In addition, when the differential evolution is used to determine the parameter set, although there is a trade-off between inaudibility and robustness, the sound quality of watermarked signal could be improved considerably while the BER could be satisfied, i.e. less than 0.1. This work does not only propose a new watermarking scheme, it also discusses about the singular value and reveals its meaning, which has been deployed and played an important role in all SVD-based schemes. Furthermore, based on analyzing the first and the second derivatives of singular spectrum, it was found that our proposed scheme can be completely blind, i.e we can extract a watermark by inputting only a watermarked signal.