An audio watermarking scheme is designed to resist synchronization attack.Firstly,it is poved that the audio mean is a stabl e feature parameter for the time scale modification (TSM) operation,and then audio mean is applied to embed the watermarking in the statistical mean of the low frequency decomposition coefficients 4-level discre te cosine transform (DCT);In order to avoiding t he continuous shifting of watermarking synchronization structure caused by means of synchronization attack s,the variable frame length dividing frame method is used in the process of extr acting watermark;In order to realizing watermarking blind detection,watermarking is embedded and extracted to audio by quantization method,and signal-to-noise ratio (SNR) is used to control the watermarking strength for watermarking inaudibilit y.The experimental results show that the proposed metho d is robust to the operation of conventional signal processing,and even can resist 20% TSM and 1/10dithering operation.