Reversible watermarking technique enables to extract embedded information without any loss of the host signal.For the reduction of the embedding distortion, a desirable reversible wa termarking approach should exploit efficient prediction way to generate prediction errors with a smaller magnitude for expansion embedding.In this paper, we present a reversible audio water marking scheme based on a new non-causal prediction method and embedding strategy.The proposed non-causal prediction method provides non-integer prediction errors and the proposed expansion embedding strategy can proceed them for a lower embedding dis tortion.Experimental results have shown that the proposed re versible technique has a lower embedding distortion for the same embedding payload in comparison with the existing state-of-the art works.