Achieving imperceptibility in all-pass watermarking of speech signals

A watermark is an ideally imperceptible signal embedded into another signal for carrying extra information. In this paper, we investigate the imperceptibility problem of a watermarking system developed by Yardimci, et. al. [1] and suggest two approaches in order to eliminate it. The system in question uses all-pass filters to embed data into a speech signal and introduces some artifacts in doing so. The two approaches we suggest in this paper remove these artifacts while keeping the watermark detectable and all the advantages of the method.