Emotion Recognition from Speech using Convolutional Neural Network and Combination of Four Speech Features

Along with technological developments, interaction based on speech does not only occur between humans but also occurs between humans and computers, known as Human-Computer Interaction (HCI). Most HCI system applications focus on contextual information. Whereas in a conversation, emotions are also important information in communication. This study combined the differential MFCC with other features, namely MFCC, chroma, and mel frequency spectrogram. The highest accuracy results in the SER data test were obtained from the test when only using one feature, namely the MFCC of 95.24%. The second-best accuracy is when using delta MFCC with an accuracy of 90.18%. And the use of 4 features (MFCC, Delta MFCC, Chromagram, and Mel Spectrogram) has the lowest accuracy compared to the others, 87.58%. In addition, the use of four features also has a longer computing time.