摘要
针对当前音频哈希指纹方法不足以满足特定音频(如广告)的实时监测问题,提出一种快速的特定音频指纹提取方法,通过提取每帧感知最相关的频域信息,将其分为33个频带,并提取相邻2帧相邻频带的差异作为指纹值,每帧提32个0/1值。实验表明,与现有方法相比,该方法在保证音频检测准确性的同时,能实现指纹的快速提取。
In view of the current audio hash fingerprinting method is not sufficient to meet the real-time monitoring for specific audio（e.g. advertising）,this paper presents a fast fingerprint extraction method for specific audio. It extracts the information with the most relevant perception of each frame in frequency domain and divides the information into 33 bands. It extracts the differences between the equivalent bands from adjacent frames as the fingerprint. For each frame,32 0/1 values are computed. Compared with existing method,this method ensures the accuracy of detection and extracts fingerprint rapidly at the same time.
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第2期211-213,共3页
Computer Engineering
基金
湖南省教育厅优秀青年基金资助项目(06B095)
湖南省研究生科研创新基金资助项目(X2008yjscx19)
关键词
音频
指纹提取
音频哈希指纹法
audio
fingerprint extraction
audio hash fingerprinting method
