Speech Emotion Recognition - Acoustic Measurement of User States in Human-Computer Interaction (HCI)

摘要

Many efforts have been reported in the literature for measuring biosignal based emotional states. These systems mainly focus on (a) oculomotoric data (eye blinking, eyelid movement, and saccade eye movement), (b) electrophysiological data (EEG, EMG, EDA), and (c) behavioral expression data (gross body movement, head movement, mannerism, and facial expression) in order to characterize emotional states. Apart from these promising advances made in these fields, there has been recently renewed interest in vocal expression and speech analysis. This fact is mainly promoted by the progress in speech science and the gaining presence of speech in voice guided HCI. Using voice communication as an indicator of emotion would have the following advantages: obtaining speech data is non obtrusive, free from sensor application, calibration efforts, and robust against climatic environmental conditions. This paper reviews the state-of-the-art in speech emotion recognition by describing cognitive-physiological mediator mechanisms of emotional states and speech changes, explaining the most relevant acoustic features and providing a pattern recognition based speech emotion recognition framework. Several validation studies analysing emotional states are presented reaching recognition rates for 2-class problems of about 80-90%. Finally, current limitations and future demands for an efficient progress in speech emotions recognition are discussed.