论文部分内容阅读
语音特征集通常具有较高的维度,高维度的特征集不仅包含噪声数据和冗余数据,影响情感识别精度,而且在分类识别的过程中将花费大量的计算开销.如何从较高维度的特征集中选择出规模更小,性能较优的特征子集对语音情感识别系统具有重要作用.本文融合过滤式和封装式两种筛选策略,提出信息增益与和声搜索算法相结合的方法进行语音情感特征选择.试验结果表明,采用过滤和封装相结合的两步策略进行特征选择,综合了过滤策略的低时间开销和封装策略的高识别率的优点,而且可以选择出较原始数据集维度更低且分类性能较好的特征子集.
Speech feature sets usually have higher dimensions, and high-dimensional feature sets contain not only noise data and redundant data, but also affect the accuracy of emotion recognition, and will take a large amount of computational overhead in classification and recognition. Centralized selection of smaller and better feature subset plays an important role in speech emotion recognition system.This paper combines filtering and encapsulation two kinds of screening strategies and proposes a combination of information gain and harmony search algorithm for speech emotion Feature selection.Experimental results show that the feature selection is based on the combination of filtering and encapsulation, which combines the advantages of low time overhead of filtering strategy and high recognition rate of encapsulation strategy, and can choose a lower dimension than the original data set And the classification of a better feature subset.