Biometric Authentication: A Machine Learning Approach

9.1. Introduction

Chapter 3 outlined the basic principles of the EM algorithm. The key idea of expectation-maximization is to introduce a set of missing variables so that a density estimation problem without a closed-form solution can be decomposed into a number of iterative steps, from which closed-form solutions can be easily obtained. The EM algorithm has been shown to be a powerful tool for estimating the parameters of mixture densities.

This chapter demonstrates how the powerful EM technique can be applied to the modeling of speaker features, which is a key step in building speaker verification systems. Specifically, this chapter details and compares several kernel-based probabilistic neural networks that are particularly appropriate for ...