基于VPRSM的音频特征选择

Audio feature selection based on VPRSM

在音频索引中保持音频特征非常重要,但是在很多情况下特征数量又很庞大,直接处理这些海量数据是非常耗时的。特征选择作为数据挖掘的一个处理步骤,在特征维数的减少和非相关数据的约简方面已经有很成功的使用。提出了一种基于变精度粗糙集模型(variableprecisionroughsetmodel,VPRSM)的音频特征选择算法。实验结果表明,该算法能够得到最小约简,并且最大程度地保持了音频数据的特征,提高检索效率。

Audio features are contained in sets of audio frames, but huge number of frames causes the audio indexing with heavy computation cost, hence how to extract those most concerned audio features is very significant. Feature selection is an important step in data mining which is successfully used to reduct the dimension of the feature set and in-elated data. A new approach of audio feature selection based on variable precision rough set is presented, Experimental evaluations indicate that the proposed algorithms produce selection sets faith to the original data and improve the efficiency of audio indexing.