一种基于快速KNFL的模式分类法及其在寂声/语声段识别中的应用

Pattern Classification Based on Fast KNFL Method and Application in Quiet Speech Identification

本文提出了一种基于快速K最近特征线(K Nearest Feature Line,KNFL)的模式分类法,这种方法可看作是对K最近邻法(KNN)的推广,它以最近特征线(Nearest Feature Line,NFL)为基础,先找出距离待识别点最近的K条特征线,再找出这K条特征线中属于某一类别数目最多的那一类作为识别结果。而在对NFL的搜索中,提出了一种快速的计算方法,使得KNFL计算量大大减少。此外,本文可以把这种算法应用于语音信号的寂声/语声段的识别中。实验结果表明,将KNFL应用于语声段/寂声段的识别会得到良好的效果。

A new pattern classification method based on fast KNFL is presented in this paper. This method based on Nearest Feature Line (NFL) is essentially an extension of the K Nearest Neighbor (KNN) method. Moreover, an algorithm is also proposed in this paper to simplify the computation of the NFL and simplify the computation of KNFL accordingly. Experimental results demonstrate that better performance in quiet speech identification can be achieved by means of fast KNFL method.