关键词: 音频指纹, 小波包分解, 奇异值熵, 样本熵, 特征提取

Abstract:

Digital audio fingerprinting technology plays an important role in the audio analysis and processing.Aiming at the problem that the traditional audio fingerprinting is extracted based on time frequency analysis using the energy of signal as a single feature parameter that can not fully characterize the complexity and irregularity,the paper proposes a method for audio fingerprinting extraction based on wavelet packet decomposition and reconstruction and combining the sample entropy of wavelet packet coefficients and the entropy singular value as a signal characteristic parameters to extract audio fingerprinting.Experimental results show that the proposed algorithm is accurate in audio recognition,robust in common audio signal operations,and capable of distinguishing different audio and locate tampered position.

Key words: audio fingerprinting;wavelet packet decomposes;entropy of singular values;sample entropy;feature extraction