Su Jeong Byun

Document Type

Thesis - University Access Only

Award Date


Degree Name

Master of Science (MS)

Department / School

Computer Science


In this thesis, I suggest a way of using voice to make up for the weak points in current security systems and downsize the load of users by affording natural and convenient speaker verification. This thesis contains the basic technology to build a practical speaker verification system using neural networks and research for implementing the system, which guarantees reliability. Firstly, this thesis will focus on making up for the weak points in current existing methods by applying the characteristics of neural networks like learning ability, parallel processing ability and adaptability. And this thesis will advance to characteristic presentation and extracting method, technique of dividing voice signal, research on new neural network model to solve speaker verification problem, adaptability for changing environment such as DTW using kernel-based PCA, spectral subtraction. To make improvements in system performance and adaptability, we study a recognition model using a module structured neural network and evaluate its appropriateness and usefulness experimentally by implementing a prototype.

Library of Congress Subject Headings

Automatic speech recognition Neural networks (Computer science) Electronic funds transfers -- Security measures



Number of Pages



South Dakota State University