Multimodal Emotion Recognition

Abstract

Recognizing different emotions of humans for system has been a burning issue since last decade. The association between individuals and PCs will be increasingly normal if PCs can see and react to human non-verbal correspondence, for example, feelings. Albeit a few methodologies have been proposed to perceive human feelings dependent on outward appearances or discourse or text, generally restricted work has been three models and other modalities to improve the capacities of the feeling acknowledgment framework. This paper describes the qualities and the restrictions of frameworks dependent on outward appearance or acoustic data or semantic and emotional word vector information. By the utilization of markers all over, nitty gritty facial movements were caught with movement catch, related to synchronous discourse chronicles and text inputs. The essential difficulties of feeling acknowledgment are picking the feeling acknowledgment corpora(speech database) distinguishing proof of various highlights identified with discourse and fitting decision of grouping. Feature Extraction utilized for feeling acknowledgment from video information are geometric and appearance-based while prosodic what more, phantom highlights are utilized for discourse information what more emotional and semantic word vector for text information. Later the given data is preprocessed as in called as Data Preprocessing. CNN is used to capture video and speech emotion-specific information. LSTM is used for text emotion-specific data. The basic aim of this models is to explore the capabilities of text, facial and speech features to provide emotion-specific information.