A Non-recurrent Low-Complexity Neural Network for the Speech Prediction

Abstract: In this paper we describe a neural network for the nonlinear adaptive prediction of non-stationary signals and demonstrate its application to a speech signal. The network is a multi-layer perceptron trained with the backpropagation algorithm. The detail of this structure is that it is no recursive, it permits a reduction in the days of training and a big gain at the time of the implementation. Furthermore, we formulate a comparative survey of the complexity of this architecture with similar works published. We present experimental results to validate the calculation method, this leads to a significant improvement in the total computational efficiency of such a predictor. The simplicity of this neural network makes for it a useful and practices for this application.