Using learning automata in brain emotional learning for speech emotion recognition

We propose an improved version of brain emotional learning (BEL) model trained via learning automata (LA) for speech emotion recognition. Inspiring from the limbic system in mammalian brain, the original BEL model is composed of two neural network components, namely amygdala and orbitofrontal cortex. In this modified BEL model, named brain emotional learning based on learning automata (BELBLA), we have employed the theory of the stochastic LA in error back-propagation to train the BEL model in decreasing the high computational complexity of the traditional gradient method. Hence, the performance of the model can be enhanced. For a speech emotion recognition task, we extract the usual features, such as energy, pitch, formants, amplitude, zero crossing rate and MFCC, from average short-term signals of the emotional Berlin dataset. The experimental results show that the BELBLA outperforms some opponents, like hidden Markov model, Gaussian mixture model, k-nearest neighbor, support vector machines and artificial neural networks, for this application. © 2017, Springer Science+Business Media New York.