Emotional Arousal of Female Politicians: Distinguishing Enthusiasm and Anger Using Data Augmentation and Machine Learning

This study aims to distinguish between anger and enthusiasm, two high-arousal emotions with negative and positive valence, respectively, using prosodic features of speech. The dataset consists of plenary speeches from ten female members of the Israeli Parliament (Knesset), with a total of 482 utterances. These utterances were labeled as anger, enthusiasm, or “other” by 70 human labelers in an online perception test. We employed Decision Tree, Logistic Regression, eXtreme Gradient Boosting (XGBoost), Support Vector Machines (SVM), and Random Forest to conduct acoustic-prosodic analysis. Among the models, XGBoost achieved the highest accuracy. The results showed that the label “other” was easier to distinguish compared to anger or enthusiasm. However, when anger and enthusiasm were combined into a single “high-arousal” label, the distinction from “other” became more pronounced. Frequency and intensity parameters were found to be crucial across all categories. Additionally, significant differences in prosodic features were observed among different politicians, reflecting individual emotional styles in speech. A correlation was also found in the feature importance rankings across all three models, with fundamental frequency, rhythm, and intensity identified as the most important features.