Adaptive Selection of Embedding Locations for Spread Spectrum Watermarking of Compressed Audio

Abstract

The main stream in audio watermarking, namely spread spectrum (SS) based methods, embeds the watermark into a fixed range in the low-frequency band of the audio in order to correctly match the spectrum coefficients of the tested signal with the watermark pattern during the correlation operation at the detector. However, a watermark that is always inserted to low frequencies could be lost during the coding of the audio files with high frequency content. In this paper, in contrary to the fixed embedding locations, we first propose to adaptively select the watermark embedding locations with respect to the maximum energy region in the frequency spectrum of the cover signal. The proposed method achieves to find the same watermark embedding and extraction locations with over 95 % detection rates down to the coding bitrates of 32 kbps. Then, the proposed adaptive selection is compared to the fixed selection of embedding locations for spread spectrum watermarking of coded audio. The experimental results on watermark bit error rates (BER) have indicated the superiority of the adaptive approach over fixed embedding against the audio coding (32-128 kbps), additive noise (10 dB-45 dB) and low pass filtering (1.4 kHz-12.4kHz). In particular, very high performance of the proposed method for high frequency audio content has demonstrated the better applicability of the scheme for large databases of audio files with various characteristics in exchange networks.