An Augmented Feature Space for Speech Emotion Recognition

With rapid developments in artificial intelligence applications, the need for better human-machine interaction is more tangible and subjects such as emotion recognition have become controversial research areas. In this paper, we develop a speech emotion recognition model by introducing novel features, which results in higher accuracies, yet less computational complexities compared to the literature. The evaluation of the proposed framework is carried out using three classifiers including support vector machine (SVM) with linear kernel, radial basis function (RBF) SVM, and extreme gradient boosting(XGBoost). The best performance is reported for linear-kernel SVM with accuracies of ۸۶.۲۸% and ۱۰۰% on RAVDESS (Ryerson Audio-Visual Database of Emotional Speech and Song) and TESS (Toronto emotional speech set) datasets, respectively. Furthermore, the performance achieved is compared with that of neural network-based methods. It is demonstrated that our results are highly comparable with the literature, although the method holds lower complexity.

کلیدواژه ها:

emotion recognition ، SVM ، XGBoost ، RAVDESS ، TESS