Title :
Nonlinear adaptive prediction of speech with a pipelined recurrent neural network
Author :
Baltersee, Jens ; Chambers, Jonathan A.
Author_Institution :
Signal Process. Lab., Tech. Hochschule Aachen, Germany
fDate :
8/1/1998 12:00:00 AM
Abstract :
New learning algorithms for an adaptive nonlinear forward predictor that is based on a pipelined recurrent neural network (PRNN) are presented. A computationally efficient gradient descent (GD) learning algorithm, together with a novel extended recursive least squares (ERLS) learning algorithm, are proposed. Simulation studies based on three speech signals that have been made public and are available on the World Wide Web (WWW) are used to test the nonlinear predictor. The gradient descent algorithm is shown to yield poor performance in terms of prediction error gain, whereas consistently improved results are achieved with the ERLS algorithm. The merit of the nonlinear predictor structure is confirmed by yielding approximately 2 dB higher prediction gain than a linear structure predictor that employs the conventional recursive least squares (RLS) algorithm
Keywords :
adaptive Kalman filters; adaptive signal processing; filtering theory; learning (artificial intelligence); least squares approximations; nonlinear filters; pipeline processing; prediction theory; recurrent neural nets; recursive estimation; speech processing; ERLS algorithm; RLS algorithm; WWW; World Wide Web; adaptive nonlinear forward predictor; computationally efficient learning algorithm; extended Kalman filter; extended recursive least squares; gradient descent learning algorithm; linear structure predictor; performance; pipelined recurrent neural network; prediction error gain; recursive least squares algorithm; simulation studies; speech signals; Computational modeling; Least squares approximation; Least squares methods; Pipeline processing; Predictive models; Recurrent neural networks; Speech; Testing; Web sites; World Wide Web;
Journal_Title :
Signal Processing, IEEE Transactions on
