Reconocimiento de emociones en el habla

Abstract

A methodology of feature extraction in emotional speech for emotion recognition is proposed. Four primary human emotions, including happiness, anger, surprise and sadness are investigated. In order to recognize emotional states, acoustic MFCC (Mel frequencycepstral coefficients) and time representation features are extracted from voice recordings. Experiments indicate that emotion recognition effectiveness comparable to human listeners can be achieved. Recognition accuracy of 94.00% for emotion detection was obtained from database SES (Spanish emotional speech).