Study the Effect of Gender on Speech Emotion Recognition Using Algerian, German, and Persian Databases

Speech emotion recognition (SER) is an attractive and challenging task in human-computer interaction and artificial intelligence technologies. SER is the process of recognizing emotions from speech utterances. This work is based on emotion recognition from speech, and our purpose is to study the effect of gender classes on the SER. Three emotional databases with different languages: ADED, EMO-DB and ShEMO in Algerian dialect, Garman and Persian languages respectively, are used to evaluate the performance of the SER system. Extracting features from speech is a required step in the creation of the SER system. Combinations of prosodic and MFCC features are exploited as speech features in the system of recognition. This last is based on Linear Discriminate Analysis (LDA), Deep Neural Networks (DNN) and Support Vectors Machine (SVM) as methods of classification. The results obtained show us that the system of SER is influenced by gender classes. The recognition rates of SER systems with gender distinction are higher than the recognition rates of SER systems without gender distinction.