In this paper, a spread spectrum based audio watermarking method with embedding positions adaption is proposed. A method of predominant local pulse (PLP) extraction is introduced to adaptively search for suitable embedded positions. Embedding watermark information at specific positions can reduce the amount of data modifications, thereby suppressing signal distortion caused by embedding. In addition, a embedding procedure based on spread spectrum, which provides robustness, is used in the DCT domain. The method discussed herein can embed watermark not only imperceivable but robust under common signal processing attacks and some of de-synchronization attacks.