基于稀疏核主成分分析的语音情感识别研究

Speech Emotion Recognition Based on Sparse Kernel Principal Component Analysis

为提升语音情感识别的能力,本研究提出一种基于稀疏核主成分分析(Sparse Kernel Principal Component Analysis,SKPCA)的方法。该方法结合核主成分分析以及稀疏表示的方法,能够同时满足特征降维和样本稀疏,起到降维和降噪的作用。本研究首先利用openSMILE工具包提取情感语音样本的声学特征及其统计特征用于情感识别,然后介绍SKPCA的算法原理及推导过程,最后使用多种分类器在柏林库做了大量的实验,实验结果表明,使用SKPCA方法可取得较好的识别结果。

In order to enhance the ability of speech emotion recognition, this paper proposes a method based on sparse kernel principal component analysis（SKPCA）. This method combines the kernel principal component analysis and sparse representation, which can satisfy both dimension reduction and sample sparse to reduce the feature dimension and noise. First, the acoustic features and their statistical features of emotional speech samples are extracted by the openS- MILE toolkit for emotion recognition. Then the algorithm and the derivation process of SKI^A are introduced. Finally a lot of experiments are carried out in the Berlin database using multiple classifiers. The experimental results showed that the use of SKPCA method achieved better recognition results.