| Title |
Optimal feature vector selection for emotion classification of similar emotional speech |
| Authors |
노경민(Kyung-min Roh) ; 채다은(Da-eun Chae) ; 이석필(Seok-pil Lee) |
| DOI |
https://doi.org/10.5370/KIEE.2023.72.10.1231 |
| Keywords |
Emotion classification; Voice data; Feature vectors; Similar emotion groups; Machine learning |
| Abstract |
This study focuses on exploring features that classification emotions that show similar patterns and are difficult to differentiate when judged based on Russell's circumplex model. To achieve this, we utilized the KEMDy19 Dataset and the EmoDB dataset to evaluate the validity of feature vectors necessary for classifying similar emotion groups. Through experiments, this research identified the most effective measure of separability among various methods for classifying similar emotions and subsequently applied these findings to an SVM classifier to explore the optimal feature combination. As a result, we confirmed that the choice of feature vectors is crucial in emotion classification, and the optimal combination can vary depending on the database and emotion group.
This study proposes an effective feature selection strategy for improving the accuracy of emotion recognition, expecting it to contribute significantly to the advancement of emotion recognition research. |