中文摘要：

为了实现对音频作品的完全级认证和对篡改操作的定位,提出了一种新的基于局部保持映射LPP（locality preserving projections）和小波包线性预测技术的音频数字签名算法。为降低音频特征向量的维数,同时保持其内在的低维结构,方便构造更有效的数字签名算法,在经过小波包变换和线性预测形成音频特征后,利用LPP算法得到音频特征集的低维流形,实现对音频数据高维特征的降维。通过混沌算法对低维流形进行置乱来得到最终的签名。实验结果表明,签名对音频数据的篡改具有较高的敏感性,可以实现对音频作品内容的完全级认证。

英文摘要：

A new audio digital signature based on LPP（locality preserving projections） and liner prediction on wavelet packet was presented to complete the authentication of audio data in the level of entireness and locate the operation of the tamper of audio data. In order to reduce the dimensionality of the eigenvector of audio data, to keep its inherent low dimensional structure and to make it easy to construct more effective digital signature algorithms, LPP method was introduced to get the low dimensional manifold of the audio features and realize the dimension reduction of the audio＇s high dimension features after the audio features generated by wallet packet transform and liner prediction. The chaotic encryption method was used to scramble the low dimension manifold to get the final signature. The experiments confirm that the signature is sensitive to the tamper of audio data, so the algorithm can be used in the authentication of audio data in the level of entireness.