基于最小平均复杂度的矢量量化音频分类方法

Audio Classification Based on Minimum Average Complexity Vector Quantization

首先提出了“平均复杂度”的概念,然后由信息熵公式给出了最小平均复杂度的计算方法,并以此为准则构造音频数据的矢量量化树,从而得到音频数据在特征空间的分布情况.根据不同种类的音频数据有不同分布这一事实,比较未知音频与已知音频种类的数据在特征空间中的分布情况的近似程度,就可完成音频分类.实验表明,该方法具有适应性强、计算效率高的特点.

The Concept of MAC(Minimum Average Complexity) is proposed first, and the calculation method is given according to the entropy formula. A VQ(Vector Quantization) tree is constructed via the MAC criterion, by which the distribution of audio feature vectors in the feature space can be obtained. In the fact that different kind of audio has different distribution, audio classification can be achieved by the degree of distribution similarity in the feature space between the unknown audio and the audio trained before. The algorithm is proven to be generalized and effective by the result of experiments.