Mandarin Emotion Recognition Based on Weighted Fusion Strategegy

In emotion recognition, different emotion features play different roles in emotion recognition. In order to improve the correct rate of emotion recognition, we modify the equivalent emotion feature weight in the traditional emotion recognition algorithm. Firstly, the weight value is set according to the contribution of each emotional feature to emotion recognition, then the weighting matrix of each modal feature is created, and finally a multimodal emotion recognition model based on feature level fusion and decision level weighted fusion is constructed and verified in CHEAVD data set. The results show that the multimodal emotion recognition model constructed in this paper improves the recognition rate by 5.43% compared with the emotion recognition model with only feature level fusion, and improves the recognition rate by 4.52%, 7.56% and 4.915% respectively compared with other mainstream emotion recognition models.