论文部分内容阅读
提出了一种基于级联投影的高斯混合模型算法.首先,针对不同的特征维度计算高斯混合模型的边缘概率,依据边缘概率模型构造出多个子分类器,每个子分类器包含不同的特征组合.采用级联结构的框架对子分类器进行动态融合,从而获得对样本的自适应能力.其次,在心电情感信号和语音情感信号上验证了算法的有效性,通过实验诱发手段,采集了烦躁、喜悦、悲伤等情感数据.最后,探讨了情感特征参数(心率变异性、心电混沌特征,语句级静态特征等)的提取方法.研究了情感特征的降维方法,包括主分量分析、顺序特征选择、Fisher区分度和最大信息系数等方法.实验结果显示,所提算法能够在2种不同的场景中有效地提高情感识别的准确率.
A Gaussian mixture model algorithm based on cascade projection is proposed.First, the marginal probability of Gaussian mixture model is calculated for different feature dimensions, and multiple sub-classifiers are constructed according to the edge probability model, each sub-classifier contains different feature combinations. The framework of the cascade structure is used to dynamically fuse the sub-classifiers so as to obtain the adaptive ability of the samples.Secondly, the validity of the algorithm is verified on the ECG emotion signals and the voice emotion signals, and the impatience, Joy, sadness and other emotional data.Finally, the method of extracting emotional parameters (heart rate variability, electrocardiogram chaos, sentence-level static features, etc.) was explored.The dimensionality reduction methods of affective features including principal component analysis, order features Selection, Fisher discrimination and maximum information coefficient.The experimental results show that the proposed algorithm can effectively improve the accuracy of emotion recognition in two different scenarios.