Vidyalakshmi K Asharani R
Audio steganography remains a challenging area of information hiding due to the sensitivity of the Human Auditory System (HAS), where minor signal modifications may introduce perceptible distortion. Traditional Least Significant Bit (LSB)-based audio steganography methods offer high payload capacity but often suffer from fixed embedding strategies, limited adaptability to signal characteristics, and vulnerability to statistical detection. This paper presents an adaptive energy-based multi-bit audio steganography framework for securely embedding executable payloads into uncompressed WAV audio signals while maintaining perceptual transparency and computational efficiency. Unlike conventional fixed-depth LSB techniques, the proposed method dynamically adjusts the embedding depth (1, 2, or 3 bits) according to local audio sample amplitude to balance payload capacity and signal fidelity. To enhance confidentiality and data integrity, the executable payload is protected using Fernet encryption and validated through CRC32 and SHA-256 mechanisms. Furthermore, vectorized NumPy processing is employed to improve embedding and extraction efficiency for large payloads. Experimental evaluation is performed using multiple WAV audio files under varying payload sizes, and performance is assessed using Peak Signal-to-Noise Ratio (PSNR), Signal-to-Noise Ratio (SNR), embedding time, and extraction time. Results indicate that the proposed adaptive strategy achieves good payload embedding capability with minimal perceptual degradation under the evaluated conditions. The study also discusses reproducibility considerations, limitations, and future directions including robustness evaluation and steganalysis resistance