Speaker recognition using neural network model

Other Title(s)

تمييز المتكلم باستخدام الشبكات العصبونية

Dissertant

Jaroun, Ayhim Rasem Fayiz

Thesis advisor

Samawi, Venus W.

University

Philadelphia University

Faculty

Faculty of Information Technology

Department

Department of Computer Science

University Country

Jordan

Degree

Master

Degree Date

2008

English Abstract

-The use of biometric information has been known widely for both person identification and security application.

It is common knowledge that each person can be identified by the unique characteristics of one or more of his/her biometrics.

One of the biometrics that a person can be identified by is the unique characteristics of his/her voice.

This work is concerned with the study of using voice as biometric information (i.e.

speaker recognition) for controlling access to the facility that need to be protected from the intrusion of unauthorized persons.

The main significance of this work is to study a number of experimental investigations of using neural networks to recognize speakers and suggest the best neural network architecture leading towards the goal of higher accuracy for speaker recognition.

Therewithal, attempt to draw a conclusion about the recommended feature-set (based on wavelet transformation) that could be used to discriminate speakers and improve the overall performance.

To do this, features are extracted from different levels of continues wavelet transformation (three maximum values with their locations and level number in addition to mean and standard division of each level), these features are used to train three of speaker recognition neural networks (Adaptive Neural Network, Feed-forward Backpropagation Neural Network, and Learning Vector Quantization Neural Network).

Experimental results that show the classification accuracy of each classifier were introduced.

Then these results were analyzed and compared to arrive at the best neural network (as speaker recognizer), and decide which of the feature (or combination of them) among the above set leads to the minimum classification error rate (i.e.

has the best discrimination ability).

A vocabulary of 240 speech samples is built for 4 speakers, where each authorized person is asked to utter every sample 10 times.

Two different modes are considered in identifying individuals according to their speech samples, in the text-dependent speaker identification, the system gives identification rate in range of 80% to 91%, while for textindependent identification mode (presented with 100 trials), where the best obtained identification rate is 71.87%.

Main Subjects

Information Technology and Computer Science

Topics

• Neural networks(Computer science)

No. of Pages

67

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-548944