基于信息增益的特征选择算法在语音识别系统中的应用

Information Gain Based Feature Selection Algorithm Application in the Speech Recognition System

在语音识别系统中,往往需要对输入的语音信息进行数据预处理操作,删除冗余的、不相关的特征值。针对传统应用于语音系统中特征选择算法中出现的效率低、错误率高的缺点,本文提出了基于信息增益的特征选择算法。该算法通过信息增益评价指标对属性进行排名及评价,选择最优的特征属性并删除无用的属性。通过大量的对比实验结果表明,本文提出的算法可以高效地完成特征选择语音数据预处理,并且提出的新算法与传统的特征选择算法选择出的特征属性应用在语音识别算法后能够更准确地识别和判断语音信息。

In the speech recognition system,we often pre-process the input speech information,and delete the redundant or irrelevant features.Focusing on disadvantages of traditional feature selection algorithms,such as low efficiency and high error rate,we propose a feature selection algorithm based on information gain in this paper.This algorithm could evaluate and rand the features according to the information gain metric,select the best features,and delete the useless attributes.The experimental results show that the algorithm proposed by this paper could complete the feature selection work in speech pro-process.Besides,the features selected by the new algorithm could recognize and judge speech information more accurately comparing with the traditional algorithms,when they apply into the speech recognition system.