?
An analysis of automatic techniques for recognizing human's affective states by speech and multimodal data
In this paper, we present an analytical survey of state of the art models, methods and techniques for automatic recognition of affective states by analyzing human’s natural speech and multimodal data such as acoustic and visual signals, as well as textually transcribed conversational speech. We consider computer based processing means for such human’s affective states as natural emotions, sentiment, aggression, depression (and some other human’s characteristics that may be indicators of a possible mental disease or communication disorder) suitable for analyzing both adults and young people. We also review existing speech and multimodal electronic resources, challenges and datasets available for creation and machine learning of computational models of various individual affective states. Additionally, we propose a novel methodological approach for a complex simultaneous analysis and multimodal recognition of multiple human’s affective states in a parallel manner.