PRoADS: Provably Secure and Robust Audio Diffusion Steganography with Latent Optimization and Backward Euler Inversion

This paper proposes PRoADS, a provably secure and robust audio steganographic framework based on audio diffusion models. As a generative steganography scheme, PRoADS embeds secret messages into the initial noise of diffusion models via orthogonal matrix projection. To address the reconstruction...