Answers are grounded in this paper and cite their sources.
Humans communicate with other humans and machines in speech at different times with different emotions. Analyzing and classification of emotions play a crucial role for effective communication. This work mainly deals with, deriving of novel features from speech signals to classify the emotional speech with better accuracy. Emotional speech is pre-processed and dual-tree complex wavelet transform is performed on each frame, to a three-level decomposition. Teager energy operator is applied and its coefficients, statistical features from time- and frequency domains along with Mel Cepstral Coefficients, center of gravity, pitch are extracted and considered as feature vector. Three classifiers, decision tree (DT), support vector machines (SVM), and linear discriminant analysis (LDA), are used to classify emotional speech signals. The proposed method classifies four universal emotions Angry, Sad, Neutral, and Happy with a classification accuracy of 87.21% with an increase of 2.21% compared to previous work proposed by Anusha et al. As the novel features are increased without overfitting the model, the accuracy in classification of emotions also increases.
No takes yet. Share an insight, caveat, or question.
Dendukuri et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: