Speech emotion recognition using a deep autoencoder

NERI E. CIBAU; ENRIQUE M. ALBORNOZ; HUGO L. RUFINER

San Carlos de Bariloche, Río Negro

Congreso; XV Reunión de Trabajo en Procesamiento de la Información y Control (RPIC); 2013

Universidad Nacional de Río Negro

The main objective of the emotion recognition systems is to improve the human-machine interaction, giving them a more natural behavior to attend different situations and user requirement. Several works on this domain use the prosodic features and the spectrum characteristics of speech signal with classifiers based on neural networks, Gaussian mixtures and other standards classifiers. In this paper a deep autoencoder based on a Multilayer perceptron was used as a classifier. Following the deep learning paradigm, an autoencoder training strategy layer by layer was implemented, which results in a stack of perceptrons with a set of ?good? weights. A final fine-tune training was applied to the whole classifier. Many configurations were evaluated in order to predict six different emotions and neutral emotional state. Performance of the classifier was over 70%, promoting better results for this novel approach.