摘要： 利用音频特征检测电视广告具有处理速度快、准确率高等优点，为此提出一种快速鲁棒的广告音频指纹提取方法。该方法将优化后的频域分为33个频带，分别计算当前帧相邻2个子带频域幅值的差异值与前一帧对应的相邻2个子带频域幅值的差异值，并将2帧的差异值之和相应设为0或1，每帧提取32个0/1值作为指纹。实验结果表明，该方法在保证快速提取指纹的同时，对常见的音频信号处理具有较强鲁棒性。

关键词: 广告检测, 音频指纹, 指纹提取, 鲁棒性, 实时性

Abstract: TV commercial detection using audio features has many advantages such as high speed and accuracy. This paper presents a fast and robust audio fingerprint extraction algorithm. The optimized frequency domain is divided into 33 bands, and the energy differences of the neighbor bands in current frame and in the previous frame are extracted separately. And the sum of the energy differences is set as corresponding value of 0 or 1. The audio fingerprint can be generated by 32 values of 0 or 1 in each frame. Experimental results show that the method proposed ensures high extraction speed and robustness against common audio signal operations.

Key words: advertising detection, audio fingerprint, fingerprint extraction, robustness, real-time performance

中图分类号: 

• TN911.73