Speech Emotion Recognition Using Label Smoothing Based on Neutral and Anger Characteristics

Speech emotion recognition (SER) is a technique of estimating emotions from speech. Recently, deep neural networks (DNNs) have improved the performance of SER, although the further improvement of SER performance is still a challenge. One of the reasons is that most SER models are trained with one-hot labels which do not represent emotions well. In this paper, we propose novel label smoothing for S...