Recurrent fuzzy neural networks for speech detection

This paper proposes a recurrent fuzzy neural network (RFNN) for speech detection. The underlying notion of the proposed RFNN is to consider minimum classification error (MCE) and minimum training error (MTE). The weights of RFNN are updated by maximizing the discrimination among different classes in MCE. Besides, the parameter learning adopts the gradient descent method to reduce the cost function in MTE. Therefore, the novelty of this paper is to minimize the cost function and maximize the discriminative capability. Finally, the experiment of speech detection is applied to test the proposed RFNN, the results show that the proposed RFNN exhibits excellent classification performance.