Pilot Study to Enhance Cover-Selection-Based Audio Steganography (CAS) Using Feed-Forward Neural Network

Steganography is a method of concealing a hidden message inside another medium ranging from image to video. The specification of the cover audio used for message embedding plays a role in the whole steganography performance. The Cover-Selection-Based Audio Steganography (CAS) technique addressed cover selection in audio steganography. However, finding the optimal cover audio using the CAS technique currently takes a significant amount of time. Therefore, the CAS technique is improved by utilising a machine learning technique called Feed-Forward Neural Network (FFNN). Similarly to CAS, Least Significant Bit (LSB) encoding is utilised for data embedding. The proposed technique’s effectiveness is assessed by comparing it with CAS regarding time performance, precision, and the stego audio quality, using a dataset of 95 inputs. The pilot study demonstrated that the FFNN model achieved 60% precision over the CAS technique in machine learning evaluation. For the audio stego evaluation, the finding shows that the proposed technique performed slightly lower than the CAS technique in the imperceptibility aspect while performing better than the CAS technique in the robustness and capacity aspects. The proposed technique achieved faster cover selection with a 5,126.89% speed reduction in performance evaluation. This study offers a valuable reference for future research on audio steganography, particularly in enhancing the performance of cover selection using machine learning