Abstract: Watermarking has a broad range of applications for multimedia copyright protection.This paper discusses the traditional approach of spread-spectrum audio watermarking and the correlation detection method.By exploring the deviation of mean and variance of original audio signals in practice from the correlation detection model,we present a novel mechanism with adaptive encoding intensity and using the psycho-acoustic model.The purpose is to improve detection convergence and robustness and the water-mark imperceptivity.The proposed mechanism is tested on a benchmark suite including several attacks.