基于深度学习的多模态情感识别综述

A survey of multimodal emotion recognition based on deep learning

简要介绍了文本、语音和人脸等3种单模态情感识别方法,总结了常用的多模态情感数据集。通过分析基于深度学习的多模态情感识别的研究现状,按照融合方式将基于深度学习的多模态情感识别分为基于早期融合、晚期融合、混合融合以及多核融合等4种情感识别方法,并进行了对比分析。最后,指出了情感识别技术研究进展存在的问题及未来发展趋势。

Three unimodal emotion recognition methods,namely text,speech,and face,are briefly introduced,and common multimodal emotion datasets are summarized.By analyzing current research status,multimodal emotion recognition by deep learning is classified into four emotion recognition methods by early fusion,late fusion,hybrid fusion and multi-kernel fusion according to the fusion method,and a comparative analysis is made.Finally,the existing problems and future development trend of emotion recognition technology are pointed out.