No abstract

Audio steganography is implemented based on three main features: capacity, robustness, and imperceptibility, but simultaneously implementing them is still a challenge. Embedding data at the Least Significant Bit (LSB) of the audio sample is one of the most implemented audio steganography methods because the method will give high capacity and imperceptibility. However, LSB has the lowest robustness among all common methods in audio steganography. To cater to this problem, researchers increased the depth of the embedding level from fourth to sixth and eighth LSB level to improve its robustness feature. However, consequently, the imperceptibility feature, which is commonly measured by Peak Signal to Noise Ratio (PSNR), is reduced due to the trade-off between imperceptibility and robustness. Currently, the lack of study on the estimation of the PSNR for audio steganography has caused the early assessment of the imperceptibility-robustness trade-off difficult. Therefore, a method to estimate PSNR, known as PSNR Estimator (PE), is introduced to enable early evaluation of imperceptibility feature for each stego-file produced by the audio steganography, which is important for the utilisation of embedding. The proposed PE estimates the PSNR based on the pattern collected from the embedment at different levels. From the evaluation, the proposed method has 99.9% of accuracy in estimating PSNR values at different levels. In comparison with the Mazdak Method, the proposed method performs better in all situations. In conclusion, the proposed PE can be used as a reference for embedding and further reducing the calculation complexity in finding the feasible value to minimise the trade-off between robustness and imperceptibility.

Existing embedding techniques depend on cover audio selected by users. Unknowingly, users may make a poor cover audio selectionthat is not optimised in its capacity or imperceptibility features, which could reduce the effectiveness of any embedding technique. As a trade-off exists between capacity and imperceptibility, producing a method focused on optimising both features is crucial. One ofthe search methods commonly used to find solutions for the trade-off problem in various fields is the Multi-Objective Evolutionary Algorithm (MOEA). Therefore, this research proposed a new method for optimising cover audio selection for audio steganography using the Non-dominated Sorting Genetic Algorithm-II (NSGA-II), which falls under the MOEA Pareto dominance paradigm. The proposed method provided suggestions for cover audio to users based on imperceptibility and capacity features. The sample difference calculation was initially formulated to determine the maximum capacity for each cover audio defined in the cover audio database. Next, NSGA-II was implemented to determine the optimised solutions based on the parameters provided by each chromosome. The experimental results demonstrated the effectiveness of the proposed method as it managed to dominate thesolutions from the previous method selected based on one criterion only. In addition, the proposed method considered that the trade-off managed to select the solution as the highest priority compared to the previous method, which put the same solution as low as 71 in the priority ranking. In conclusion, the method optimised the cover audio selected, thus, improving the effectiveness of the audio steganography used. It can be a response to help people whose computers and mobile devices continue to be unfamiliar with audio steganography in an age where information security is crucial.