Toward robustness of audio watermarking systems to acoustic channels

This article deals with blind audio watermarking systems dedicated to data transmission applications, where high embedded information bitrates are prospected. We present an original way to improve the robustness of such systems to perturbations yielded by acoustic convolutive channels. The proposed method, using the analogy between watermarking and digital communication, relies on 1) a channel estimation stage based on an original adaptation of the trained RICE algorithm to watermark inaudibility constraint and 2) a dedicated equalizer, built to invert the convolutive channel effects before data extraction. The efficiency of the proposed method is evaluated through simulations conducted for various real acoustic channels and audio signals. It is shown that the system bit error rate can be decreased from 0.2 to 9.10^-4 thanks to our contribution when the bitrate transmission is 100 bps and the channel is the acoustic one.