A study about MFCC relevance in emotion classification for SRoL database

The focus of this paper is to establish the relevance of MFCC coefficients in the emotion recognition for Romanian language, comparing with prosodic features: F0 fundamental frequency, F1-F4 formants, jitter and shimmer. We noted that the accuracy recognition rate is improved by using MFCC feature vectors around 90%. In our previous works we obtained only 65% percent in emotion classification with feature vectors which contain F0, F1-F4 formats, jitter and shimmer. We also studied the relevance of the derivative ΔMFCC and ΔΔMFCC. The obtained results are remarkable considering that the SRoL database contains only “normal” voices. In literature, similar performance is reported usually on the databases with professional voices.