Genetic Programming for Robust Text Independent Speaker Verification

Genetic Programming for Robust Text Independent Speaker Verification

Peter Day (d-fine Ltd, UK) and Asoke K. Nandi (University of Liverpool, UK)
Copyright: © 2010 |Pages: 22
DOI: 10.4018/978-1-60566-705-8.ch011

Abstract

Robust Automatic Speaker Verification has become increasingly desirable in recent years with the growing trend toward remote security verification procedures for telephone banking, bio-metric security measures and similar applications. While many approaches have been applied to this problem, Genetic Programming offers inherent feature selection and solutions that can be meaningfully analyzed, making it well suited for this task. This chapter introduces a Genetic Programming system to evolve programs capable of speaker verification and evaluates its performance with the publicly available TIMIT corpora. Also presented are the effects of a simulated telephone network on classification results which highlight the principal advantage, namely robustness to both additive and convolutive noise.
Chapter Preview
Top

Background

The aim of ASV systems is to answer the question, “is the speaker who s/he claims to be”. More formally, the challenge of an ASV system is to establish the presence of a speaker within an unknown (open) set of speakers. This problem, together with the related Automatic Speaker Identification (ASI) problem (in which the aim is to establish which speaker is speaking from a closed set of speakers), has enjoyed sustained research interest over recent years.

Most proposed ASV systems share similar strategies. The typical process involves: some form of pre-processing of the data (silence removal) and feature extraction, followed by some form of speaker modelling to estimate class dependent feature distributions (see Figure 1). A comprehensive overview can be found in Campbell (1997). Adopting this strategy the ASV problem can be further divided into the two problem domains:

Figure 1.

A typical automatic speaker recognition (ASR) system (Figure originally published in Day and Nandi (2007) Ⓒ 2007 IEEE)

  • 1.

    Feature extraction and selection, and

  • 2.

    Speaker modelling and matching

We discuss each of these in turn below.

Purchase this chapter to continue reading all 22 pages >

Complete Chapter List

Search this Book:
Reset