Emotion recognition from semi natural speech using artificial neural networks and excitation source features

Abstract

This paper proposes Linear Prediction (LP) residual of speech signal for characterizing the basic emotions. LP residual is extracted from speech signal by LP analysis, by inverse filtering of the speech signal. LP residual basically contains higher order relations among the samples. Instant of glottal closure in a speech signal is known as an epoch. The significant excitation of vocal tract usually takes place at the instant of glottal closure. For analysing speech emotions, the LP residual samples chosen around glottal closure instants are used. A semi-natural database GEU-SNESC (Graphic Era University Semi Natural Emotion Speech Corpus) is used for modeling the emotions. This database is collected by recording dialogs of film actors from Hindi movies. In the study four emotions namely anger, happy, neutral and sadness are used. Auto-associative neural network models are used for characterizing the basic emotions present in the speech. Average emotion recognition of 66% and 59% is observed respectively for the epoch based and entire LP residual samples. © 2012 Springer-Verlag.