A new statistical attack resilient steganography scheme for hiding messages in audio files

We present a novel approach for audio steganography that preserves first-order statistical properties of cover audio after embedding a secret message. This approach can avoid detection by histogram-based or similar statistical attacks. This approach partitions the audio samples in the cover audio, which is followed by reordering of the samples in each partition through an encoding process for embedding the secret message. Partitioning of samples is governed by a specified error limit on individual samples, and the error limit is determined from signal-to-noise ratio that needs to be maintained in the stego audio to avoid detection by an automated system or human auditory system. Experimental results on effectiveness as well as on capacity are presented using 8-bit and 16-bit audio as covers. It is shown that the proposed approach can achieve high capacity while maintaining its effectiveness against attacks.