Prosodic Parameterisation and Ensemble Machine Learning Techniques for Speech Emotion Detection

Abstract

This paper presents a robust approach to speech emotion recognition using an auto-correlated dynamic classical algorithm selection method. It employs FFT-based autocorrelation to extract fundamental frequency power spectral density for intensity, and inter-stress intervals for rhythm. Machine learning models (SVM. Decision Tree Naive Bayes, KNN, and Random Forest) were trained on RAVDESS, Emo-DB, TESS. IEMOCAP. and CREMA-D datasets, with ensemble learning achieving a state-of-the- art accuracy of 99.98%. The results highlight the effectiveness of the proposed approach in accurately detecting emotions from speech.