Nowadays, most prevailing approaches to advanced audio coding (AAC) steganography are content non-adaptive which have low embedding capacity and poor security. In this paper, we construct a new distortion by integrating the uniform embedding distortion with the masking threshold, and propose an adaptive steganographic scheme for AAC audio by modifying quantized modified discrete cosine transform (QMDCT) coefficients. To hold the statistical distribution of QMDCT coefficients we introduce uniform embedding idea. Then the psychoacoustic model is used to avoid the declining of hearing quality. Finally, we minimize the overall embedding distortion by utilizing syndrome-trellis codes (STCs) technique and the defined distortion function. Comprehensive experimental results show that the proposed scheme achieves better performance than related works. The detection error of 128-kbps-AAC dataset is higher than 24.78% when the embedding payload reaches 3.70 kbps, which is significantly lower than the state-of-the-art AAC steganographic methods.

如今，最先进的高级音频编码（AAC）隐写方法是不自适应内容的，其嵌入能力低且安全性差。在本文中，我们通过将均匀嵌入失真与掩蔽阈值相集成来构造新的失真，并通过修改量化的改进的离散余弦变换（QMDCT）系数，提出了针对AAC音频的自适应隐写方案。为了保持QMDCT系数的统计分布，我们引入了统一的嵌入思想。然后，使用心理声学模型来避免听力质量下降。最后，我们利用校正子格码（STC）技术和定义的失真函数将整体嵌入失真降至最低。综合实验结果表明，该方案比相关方案具有更好的性能。