Context-rich Detection of User’s Emotions using A Smartphone

As an increasingly powerful computation platform, today’s off-theshelf smartphones are able to handle more sophisticated mobile sensing tasks, such as emotion sensing. Although numerous emotion detection technologies have been developed, the unpredictable variation of a person’s emotion raises the demand of sensing emotion in a mobile fashion. Also, detecting emotion is difficult, since people sometimes may hide their real emotion from the others. Thus, smartphones, as people’s everyday companion, can help detect people’s emotion in a more objective and unobtrusive way, and hence offers more accurate results. We propose a mobile emotion sensor on Windows Phone 7 called ’listen-n-feel’, based on the way people speak. Signal processing methods are used to extract speech features, and a machine learning algorithm called logistic regression is used for emotion prediction. We also discuss the possibilities of combining contextual parameters in emotion detection. The emotion prediction accuracy is around 71% when differentiating happy and sad emotions, evaluated by both cross-validation and a preliminary test on users. Mobile emotion sensing may have a broader appeal in fields of healthcare interventions, psychosocial studies, game design, etc.