面向非特定人语音情感识别的PCA特征选择方法

在语音情感识别中，如何选取有效的情感特征是识别过程的重要环节。迄今为止，一些常用的特征选择算法虽然能够帮助提高识别性能，但也存在理论性不强、随机性高、计算量大的缺点。因此提出了一种基于主成分分析(PCA)的特征选择方法，亦即对原始特征集合先进行PCA变换，再利用变换矩阵分析出原始特征进行变换时各自的权重，最后根据权重的大小对原始特征进行选择。实验结果表明，选择出的特征对识别率具有较大的贡献，属于重要特征。

PCA Based Feature Selection Algorithm on Speaker-independent Speech Emotion Recognition

A very important part of emotion recognition is how to select effective emotional features. Until now, some feature selection algorithms, which are usually used, can help boost recognition accuracy. But some defects, such as less robustness in theory, a higher randomness, more computation, still exist. For these reasons, a new feature selection algorithm based on PCA (principal component analysis) was proposed. First the original feature set was transformed by PCA, then analyzing the weights of these features using the transforming matrix and finally, choosing the important features according to their weights. hhe experiment result shows that features, which arc selected by this method, make a high contribution to the recognition accuracy and they are important.