Discriminating coding applied to the Automatic Speaker Identification

In this paper we focus on the speech signal encoding applied to Automatic Speaker Identification system. We present the extension to the nonlinear field of the linear predictive coding (LPC) method usually used in ASI system. This extension is based on a neural network multilayer perceptron (MLP) in the context of prediction, and it is called Neural Predictive Coding (NPC). We present an experimental study from the Numenta Speakers database. A comparative study with the other traditional coding methods LPC and MFCC are explored. Advantages and disadvantages of each method are discussed, the effects introduced by the speech coding and the speakers number were taken into account. The Results indicate that an improvement in recognition rate and the ASI system complexity by minimizing the necessary feature number by using the NPC feature extraction.