摘要： 在互联网中重复上传他人已经分享的歌曲会消耗网络带宽，浪费存储空间，但目前的重复数据删除方法主要基于文件的二进制特征，无法识别经过信号处理或压缩后的歌曲。针对该问题，提出一种基于声学指纹的海量MP3文件近似去重方法。结合文件消息摘要的确定性与声学指纹的鲁棒性，在采用布隆过滤器对文件消息摘要一次去重的基础上，根据降维后的声学指纹值进行二次近似去重，保证高效的同时提高去重率。实验结果表明，与可变分块检测方法相比，该方法的去重率可提高1倍以上，扩展性较好。

关键词: 声学指纹, 重复数据删除, 近似去重, 布隆过滤器, 海量数据

Abstract: Song re-uploading had been shared wastes network bandwidth and storage, which needs to use data de-duplication technology. However, the current approach to de-duplication based file bit-feature does not recognize the same song after signal processing or compression. Aiming at this problem, this paper proposes a near de-duplication method of massive MP3 files based on acoustic fingerprint. It combines the certainty of message digest with the robustness of the acoustic fingerprint, after Bloom Filter(BF) de-duplicate data based on the message digest, then reduces acoustic fingerprint for the secondary near de-duplication based on the dimensionality. It ensures efficient at the same time, greatly improves the de-duplication ratio. Experimental results show that this method can improve the de-duplication rate by one time than Content-defined Chunking(CDC) method, and has good extensibility.

Key words: acoustic fingerprint, data de-duplication, near de-duplication, Bloom Filter(BF), massive data

中图分类号: 

• TP311