Abstract: A novel digital audio watermarking technique based on SVM(support vector machine) is proposed in this paper. The main idea of the method is that the retrieval of embedded watermark can be considered as a two-class problem, and SVM can be used to learn the characteristics of the embedded watermark in audio. Because the SVM possesses the learning and adaptive capabilities, it almost exactly recovers the watermark from the watermarked audio. The extensive experimental results show that the embedded watermark is robust against audio signal processing, such as MPEG audio coding, cropping, filtering, resampling and requantizing. The watermarking method does not require the use of the original signal for watermarking detection.