Noisy Speech Emotion Recognition in Romanian Language

Abstract

The ability to understand the emotions is essential in health development and harmonious integration into the environment for the human being. The anomalies of emotion perception can contribute to psychiatric disorders associated with social problems. In this paper, we studied the emotion recognition affected by noise using the SROL database which included 4 emotions (joy, anger, sadness and neutral tone), for the Romanian language. The additional noise introduced is between 5% - 25% from the signal energy. For the best features vectors $\\mathbf{MFCC} + \\mathbf{AC}$, we noticed a decrease in the recognition rate from about 90% for the signals without noise to a percent around 81% for signals with noise of 25% from the input signal amplitude.