Advances Research in Speech Emotion Recognition Based on Multi-task Learning

In recent years, the powerful feature learning capability of deep learning has boosted the development of speech emotion recognition. However, issues like model overfitting or unbalanced feature distribution across corpora still challenge this area. Therefore, multi-task learning methods have been applied to speech emotion recognition, which shares valuable information among multiple related tasks to improve model performance. Given its sufficiently learning effect, this paper outlines and summarizes the recent literature on speech emotion recognition based on multi-task learning. Our contributions can be organized as follows: (1) we briefly review the current status of speech emotion recognition research and provide an overview of multi-task learning, (2) highlight the current multi-task speech emotion recognition algorithms, and (3) finally summarize the challenges and opportunities by analyzing the performance of different methods. We believe our review can summarize the related field's development and inspire discussion of future research.