A Hybrid Audio Watermarking Scheme Using LU Decomposition on Linear Prediction Residuals with Lightweight Chaotic Encryption

Abstract

The expansion of digital audio communication has made data security and copyright protection a critical challenge in modern multimedia systems. This paper presents a hybrid audio watermarking scheme that combines several complementary techniques to achieve imperceptibility, robustness, and security. In the proposed method, the host audio signal is divided into fixed-length frames, and both Linear Predictive Coding (LPC) and Discrete Cosine Transform (DCT) coefficients are extracted. An Ant Colony Optimization (ACO) algorithm is then employed to select the most suitable coefficients for watermark embedding. Prior to embedding, the watermark is encrypted using lightweight chaotic maps to provide resistance against unauthorized access and intentional attacks. The encrypted watermark is embedded using Quantization Index Modulation (QIM), and the modified coefficients are organized into a square matrix where LU decomposition is applied. The watermark data are stored in the upper triangular matrix (U), which improves numerical stability and enhances robustness against noise and signal processing operations. Extraction of the watermark is carried out in a blind manner, without requiring the original signal. Experimental evaluations demonstrate that the proposed scheme achieves Signal-to-Noise Ratio (SNR) of 47.135 dB, and embedding capacity of 716.2 bps. The method offers reliable recovery with a low bit error rate and demonstrates resilience against common attacks such as MP3 compression, noise addition, filtering, and re-recording. By integrating chaotic encryption with the LPC–DCT–ACO–LU framework, the method provides a secure and resilient solution suitable for copyright protection and authentication in digital multimedia applications.