Use Empirical Mode Decomposition and Ensemble Deep Learning to Improve the Performance of Emotional Voice Recognition

As the elderly population of Taiwan continues to rise, appearing the shortage of care measures and human resource. Emotional sounds can reflect physical and mental states of children, which are very important in child care. In this paper, we use emotional voice recognition technology to assist caregivers to understand the demands of the elderly and children, so that they can give appropriate care in time. Empirical mode decomposition (EMD) is applied to improve the detection and recognition of emotional sounds. Moreover, ensemble deep learning is further applied to alleviate the problem of overfitting. The experiments show that the classification accuracy of the proposed approach is 91.6%, which is much higher than that without EMD. We hope this technology will bring the benefit to the care of elderly and children.

I. Introduction

As the fast growth of machine learning and deep learning recently, researches on applying these technology for assisting elderly and children care become popular. This paper proposed an intelligent system for recognizing emotional voices such as laugh, cry, scream, wail or sigh, to assist caregivers to understand the demands of the elderly and children. Appropriate care can be given to them more quickly.