Emotion Computation in Children Speech in a Robotic Environment Using Gaussian Support Vector Machine

摘要

This paper quotes a novel approach to deduce emotions from children in a machine-child interaction scenario. Emotion recognition can help the computer agents to adapt its interaction strategies to improve the efficiency of the application. In this work, four emotions are classified: confidence, hesitation, moody and puzzled. The emotions are deduced based on cue computations including lexical, prosody, spectral and syntactic. A gaussian support vector machine is deployed which takes the feature vector composed of the cues, and classifies each utterance of a speaker into one of the four classes. The approach has been implemented in PF-Star British English dataset. The methodology adopted is speaker-independent and yielded 75.98% accuracy in deducing the emotion classification.