Digital audio watermarking using psycho-acoustic frequency masking and Gaussian mixture model.

Abstract

Most of papers related to the watermarking techniques are currently focused on the development of a blind watermarking detection algorithm and a watermark embedding algorithm based on the psychoacoustics. This paper proposes a blind watermark scheme based on psycho-acoustic frequency masking (PAFM) and Gaussian mixture model (GMM). PAFM is used for watermark embedding into each frame and GMM is used for detecting watermarks. This paper compares robustness and sound quality between the proposed method and informed watermark scheme. Informed method is implemented by using errors-in-variables (EIV) and linear predictive coding (LPC) and is also based on spread spectrum. In this method, target frames are chosen by comparing the average energy for whole signals with the energy of every frame and watermarks are embedded into principal peaks around estimated formants computed by EIV and LPC. In order to evaluate robustness of proposed method, we attacked the sound by using LPF, cropping, and additive noise. [This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea Government (MEST) (No. R01-2008-000-20493-0).]