Speech Emotion Recognition System for Human Interaction Applications

Speech emotion recognition systems are needed nowadays in many human interaction applications such as call-centers, e-learning, autonomous driver emotion detection, and physiological diseases analysis. The existing speech emotion recognition systems focus mainly on a single corpus. On the other hand, the speech emotion recognition performance on cross-corpus is still an ongoing challenge in the research domain. This paper presents a study for speech emotion recognition tested on 3 widely used languages (English, German and French) using RAVDESS, EmoDB and CaFE datasets. Four Machine learning classifiers: Multi-Layer Perceptron, Support Vector Machine, Random Forest and Logistic Regression have been used. A newly developed feature set was introduced consisting of main speech features as prosodic features, spectral features and energy. Very promising results were obtained using this feature set even when compared with the benchmarked feature set "Interspeech 2009". Furthermore, Feature importance techniques were used to study the feature importance per each classifier across each corpus. From our results, it was found that SVM is the best classifier from recognition rates and running performance point of view. The model achieved an accuracy of 70.56% on RADVESS, 85.97% on Emodb and 70.61% on CaFE.