Genetically Optimised Feedforward Neural Networks for Speaker Identification

摘要

The problem of establishing the identity of a speaker from a given utterance has been conventionally addressed using techniques such as Gaussian Mixture Models (GMMs) that model the characteristics of a known speaker via means and covariances. In this paper we pose the task as a binary classification problem, and whilst in principle any one of a number of classifiers could be applied, this work compares the performance of genetically optimized neural networks versus the conventional approach of GMMs. The test data used in the experiments was the data used for the 1996 National Institute for Standards Technology (MST) evaluation of speaker identification systems.