kbps Adaptive Differential Pulse Code Modulation of Speech
S. Haykin. L. Li · Psychology Press eBooks · 2013
The conventional form of ADPCM relies on the use of a linear predictor, yet it is widely recognized that the biological mechanism responsible for the generation of speech signals is in fact nonlinear. This points to a shortcoming of the conventional form of ADPCM. In this paper we address this issue by investigating the use of a neural network as the basis for the design of a nonlinear predictor. However, the choice of a suitable neural network architecture for such an application is constrained by the fact that the network has to perform its learning in an on-line fashion as the speech signal is being continuously processed . In this paper, we describe a novel neural network that accomplishes this requirement .