Evaluation of feature extraction methods for speaker verification

Pao Yue-kong Library Electronic Theses Database

Evaluation of feature extraction methods for speaker verification


Author: Lam, Chin-lung
Title: Evaluation of feature extraction methods for speaker verification
Degree: M.Sc.
Year: 1998
Subject: Speech processing systems
Hong Kong Polytechnic University -- Dissertations
Department: Multi-disciplinary Studies
Dept. of Electronic Engineering
Pages: vi, 57, vi leaves : ill. ; 30 cm
Language: English
InnoPac Record: http://library.polyu.edu.hk/record=b1446533
URI: http://theses.lib.polyu.edu.hk/handle/200/3169
Abstract: A common problem in text-independent speaker verification systems is that a mismatch between the training and testing conditions sacrifices much performance. A common example of this mismatch is that training is done on clean speech while testing is performed on noisy or channel-corrupted speech. Robust speech processing techniques attempt to maintain the performance of a speech processing system under such diverse conditions. Two strategies of robust speech processing techniques have emerged to mitigate the problems that arise due to channel effects and noise. The first strategy is normally carried out in the front-end feature extractors. The second strategy aims at making the classifier more robust by compensating the distortions between the template patterns and the unknown patterns during the classification stage. The conventional linear predictive (LP) cepstrum and the delta cepstrum are the commonly used features for speaker recognition systems. The linear predictive (LP) cepstrum derived from an all-pole transfer function is able to approximate the spectral envelope of the speech signals. A newly proposed feature, namely the Adaptive Component Weighted (ACW) cepstrum, has been found to be robust against channel variations and noise. The ACW cepstrum is derived from a pole-zero transfer function whose denominator is a pth order LP polynomial. This dissertation compares the LP cepstrum and ACW cepstrum for speaker verification. Experiments were carried out based on the NTIMIT corpus where the feature vectors were classified by radial basis function neural networks, Experimental results show that the ACW cepstrum is better than LP cepstrum in extracting speaker features from telephone speech.

Files in this item

Files Size Format
b14465334.pdf 2.288Mb PDF
Copyright Undertaking
As a bona fide Library user, I declare that:
  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.


Quick Search


More Information