Exploring Recurrence Properties of Vowels for Analysis of Emotions in Speech

Speech Emotion Recognition (SER) is a recent field of research that aims at identifying the emotional state of a speaker through a collection of machine learning and pattern recognition techniques. Features based on linear source-filter models have so far characterized emotional content in speech. However, the presence of nonlinear and chaotic phenomena in speech generation have been widely proven in literature. In this work, recurrence properties of vowels are used to describe nonlinear dynamics of speech with different emotional contents. An automatic vowel extraction module has been developed to extract vowel segments from a set of spoken sentences of the publicly available German Berlin Emotional Speech Database (EmoDB). Recurrence Plots (RPs) and Recurrence Quantitative Analysis (RQA) have been used to explore the dynamic behavior of six basic emotions (anger, boredom, fear, happiness, neutral, sadness). Statistical tests have been performed to compare the six groups and check possible differences between them. The results are promising since some RQA measures are able to capture the key aspects of each emotion