Text-Dependent Speaker Identification Based on The Modular Tree: An Empirical Study

摘要

Recently, a novel self-architecture modular neural network model called Modular Tree was proposed based upon the principle of divide-and-conquer (1)(2). In this paper, we apply the modular neural network architecture to text-dependent speaker identification. Based upon the experimental results, we demonstrate that the performance of the system using the Modular Tree is satisfactory. Moreover, the use of the Modular Tree yields fast training and updating in the system.