中文摘要：

提出一种基于BP神经网络和信噪比（SNR）的自适应音频水印算法。通过修改音频分抽样后子音频帧小波变换近似分量均值的方法嵌入水印。水印嵌入强度由期望的SNR值确定，可避免反复实验或者由经验数据确定水印强度。水印的提取以子音频抽样帧小波变换近似分量均值为BP神经网络的输入，以神经网络的输出作为提取的水印。经过模板训练的BP神经网络能有效的恢复嵌入到音频中的水印数据，实现了水印的盲检测。实验结果表明，本文提出的水印算法具有较好的鲁棒性和不可听性。

英文摘要：

An adaptive audio watermarking algorithm based on BP neural networks and signal-to-noise ratio （SNR） is proposed. The watermark is embedded by modifying wavelet transform approximate coefficient mean of sub-sampling of audio frame. The watermarking strength is assured according to expected SNR and it can avoid using over and over experiments or experience data to confirm the watermarking strength. Mean of approximate coefficients of sub-sampling of audio frame is served as input of BP neural networks in process of watermark extracting. The output of neural networks is served as the extracted watermark. BP neural networks trained by template information can effectively restore the watermark embedded into audio, which realizes blind detection of watermark. The experimental results show that the proposed algorithm is robust and imperceptible.