Nonlinear dynamic neural network for text-independent speaker identification using information theoretic learning technology.

ABSTRACT

In this paper we present a novel design for a nonlinear dynamic neural network to implement text-independent speaker recognition without the benefit of exact voice signatures. The dynamic properties between the input neuron and the output neuron make use of a nonlinear high-order synaptic neural model with memory of previous input signals. The dynamic neural network is realized in the short-term-frequency long-term-temporal domain. Informatics metric is used to overcome the challenge of performing blind learning for the nonlinear network. The goal of this study is not only to improve the recognition performance but also to amplify the distinctiveness among different speakers.