Emotion Recognition for Romanian using DL_CNN Combined with Predictive Neural Network

This work studied how emotion can be detected using predictive neural networks in order to separate areas of interest for emotions represented by vowels, from regions of noise or pause between utterances. A Deep-Learning convolutional neural network (DL-CNN) provided slightly better results of about 80% when using MFCC cepstral coefficient images for training, compared to a lower recognition rate of about 77% when using MFSC mel scale spectrogram images for training. The recognition rate is better for MFCC images, although they appear visually of poorer quality than MFSC images.