A Survey of Speech Emotion Recognition Through Learning on Limited Data

Speech emotion recognition endows machines with the ability to recognize emotions, enhancing the naturalness of Human Computer Interaction experiences. Recent investigations in Speech Emotion Recognition (SER) have been focused on machine learning methods. However, suffering from timeconsuming and privacy-related issues, the SER tasks through Learning on limited data emerge as the times require. Starting from the concept and the historical process of limited-datarelated speech emotion recognition. Then, systematically summarizes the speech emotional databases and features extraction in speech. Subsequently, throuth listing different meta-learning speech emotion recognition methods, we analyze the problems in limited-data speech emotion recognition tasks and provide guidance for further research in terms of datasets, features and emotional recognition models. Significant recent researches effort that few-shot and zero-shot learning is a promising and valuable directions in SER.