Salient cross-lingual acoustic and prosodic features for English and German emotion recognition

Abstract

While approaches on automatic recognition of human emotion from speech have already achieved reasonable results, a lot of room for improvement still remains there. In our research, we select the most essential features by applying a self-adaptive multi-objective genetic algorithm. The proposed approach is evaluated using data from different languages (English and German) with two different feature sets consisting of 37 and 384 dimensions, respectively. The obtained results of the developed technique have increased the emotion recognition performance by up to 49.8% relative improvement in accuracy. Furthermore, in order to identify salient features across speech data from different languages, we analysed the selection count of the features to generate a feature ranking. Based on this, a feature set for speechbased emotion recognition based on the most salient features has been created. By applying this feature set, we achieve a relative improvement of up to 37.3% without the need of time-consuming feature selection using a genetic algorithm.