Abstract: Doublets of wavelet coefficients are defined and a watermarking scheme for digital audio based on doublet of wavelet coefficients is proposed. Original audio signal is decomposed by wavelet transform and doublets of the wavelet coefficients are selected and classified into two classes to embed and extract the watermark according to the relationship of the two wavelet coefficients. A shuffling algorithm is adopted to randomize the watermark sequence for the purpose of security. Experimental results show that the degradation caused by the proposed algorithm is very slight, and the watermarking scheme is robust to those audio processing and malicious attacks in common use such as re-sampling, re-quantization, low-pass filtering, Mpeg coding, and Gaussian noise, etc.