Increasing the capacity of LSB-based audio steganography

Conventionally, a perceptual limit of three bits per sample is imposed to the basic LSB audio steganography method. In this paper, we present a novel modification to standard LSB algorithm that is able to embed four bits per sample, thus improving the capacity of data hiding channel by 33%. The proposed algorithm makes use of minimum error replacement method for LSB adjustment and modified error diffusion method for decreasing SNR value. Objective test showed the algorithm succeeds in this task, while keeping SNR value close to the level of SNR obtained by standard LSB embedding with three bits per sample capacity. Subjective listening test proved that high perceptual transparency is accomplished even if four LSBs of host audio signal are used for data hiding.