Robust Data Embedding Based Probabilistic Global Search in MDCT Domain

Abstract

The rapid growth of the information technology, data hiding in audio is an important role of digital media protection and secret communication. Thus, a technique of robust data embedding in the modified discrete cosine transform domain (MDCT) of cover speech signal is presented to solve the issues of digital audio steganography application. The major contribution of the proposed scheme is introduced Probabilistic Global Search Lausanne Algorithm (PGSL) as the optimization-based problem transformation method, to search the best positions in the first selected frames of MDCT Domain on cover speech samples. The covert data are embedded into the positions of lowest bit error rate value to guarantee perfect inaudibility and enhance imperceptible as well as robustness between cover and stego audio. Speech signals from TIMIT Dataset of 452 Male and Female speakers were served to the host cover audio data. The distortion between cover and stego audio is evaluated with Informal Listening Test and Spectrogram. Results of data hiding using TIMIT clean utterance showed that the distortions in the stego audio is inaudible and imperceptible from the original cover speech even with high payload in the Spectrogram and Informal Listening Test. The proposed system defined to apply the signal manipulation attacks as resampling and noise addition to increase in the robustness of data embedding scheme. Objective measurement of stego audio quality degradation by data embedding was performed by PSNR value.