Pitch and Energy Contribution in Emotion and Speaking styles Recognition Enhancement

This paper highlights the influence of some prosodic features in enhancing the recognition accuracies of emotions and speaking styles in speech. In this work, we seek to recognize 10 emotions and speaking styles based on real speech. After having extracted a large amount of cues, we use the hidden Markov models classifier. Results are given on text-independent emotion recognition using SUSAS database. The aim of this work is to test the influence of energy and pitch in emotion recognition. Throughout this study, MFCC, log energy and pitch frequency, are used as the base features. The obtained recognition accuracy for 10 different emotions and speaking styles exceeds 85% reaching 89.35% for the slow style using the best combination of spectral and prosodic features