中文摘要：

音频认证可以分为硬认证、基于质量和基于内容的软认证等三种。硬认证只允许格式转换和无损压缩,基于质量和内容的软认证则分别允许一些保持听觉质量或语义的音频处理。在绝大多数应用环境下,需要对音频进行与人类听觉感知系统特性相符合的基于内容（语义）的软认证。音频认证可采用数字水印或数字签名,提出一种基于小波包最优基分解的数字签名算法,利用与音频内容密切相关的小波包系数作为特征进行语义级的认证。实验结果表明该算法对常见的信号处理MP3、WMA、RM等中等强度的有损压缩、添加噪声、重采样等保持内容操作具有很强的鲁棒性,而对局部替换、修改、删除、复制音频等恶意操作脆弱,并能准确定位被篡改的位置。

英文摘要：

Audio authentication can be classified into hard, quality-based and content-based soft authentication. Hard authentication allows only for format conversion and lossless compression,while soft authentication based on quality and content permits certain degree of audio frequency quality or content preserving manipulations respectively. In most application scenarios, what is needed most is the content-based soft authentication that can maintain the audio semantic meaning, which also meets the characteristics of human auditory system. Besides watermarking or digital signature is another important technique adopted for audio authentications. In this paper,we present a digital signature algorithm based on wavelet packet （WP） best basis decomposition, it selects the WP coefficients highly related to audio content as the feature to carry out authentication at semantic level. Experimental results show that the proposed scheme is robust to admissible content preserving operations of lossy compression, noise addition and re-sampling, etc. in medium intensity linked with MP3, WMA and RM, and fragile to malicious operations including malicious local substituting, modifying, deleting, copying, etc. , with the ability of precisely locating the tempered positions.