Table Of Contents

Table Of Contents
List Of Illustrations
List Of Tables
List Of Abbreviations
Acknowledgements
Chapter 1. Introduction
	Section 1.1 Introduction
	Section 1.2 Overview Of Research In Speech Recognition
	Section 1.3 Recognition Of The English Alphabet
	Section 1.4 Structure Of Thesis
Chapter 2. Neural Networks For Speech Recognition
	Section 2.1 Introduction
	Section 2.2 The Perceptron
	Section 2.3 The Multi-Layer Perceptron
	Section 2.3.1 The Error Back Propagation Algorithm
	Section 2.3.2 The Fully Connected Network
	Section 2.3.3 The Scaly Neural Network Architecture
	Section 2.4 Summary
Chapter 3. Preprocessing Of The Speech Data
	Section 3.1 Introduction
	Section 3.2 Time Alignment Algorithms
	Section 3.2.1 Linear Time Alignment
	Section 3.2.2 Dynamic Time Warping
	Section 3.2.3 Trace Segmentation
	Section 3.3 Summary
Chapter 4. Investigation Of Time Alignment Algorithms
	Section 4.1 Introduction
	Section 4.2 The English Alphabet Database
	Section 4.3 Mel Frequency Cepstral Coefficients
	Section 4.4 The Neural Network Simulator
	Section 4.5 Experimental Procedure
	Section 4.6 Simulation Results
	Section 4.6.1 Training On 'A' and 'B'
	Section 4.6.2 Training On 'A', 'B' and 'C'
	Section 4.6.3 Training On The Whole Alphabet
	Section 4.7 Discussion
	Section 4.7.1 Effect Of Architecture On Performance Rates
	Section 4.7.2 Effect Of The Activation Function
	Section 4.7.3 Effect Of Learning Rate
	Section 4.7.4 The Time Alignment Method
	Section 4.8 Conclusions
Chapter 5. Investigation Of The Number Of Inputs To The Scaly Neural
	Section 5.1 Introduction
	Section 5.2 Experimental Procedure
	Section 5.3 Simulation Results For The E-Set
	Section 5.4 Simulation Results For The Whole Alphabet
	Section 5.5 Discussion
	Section 5.5.1 Effect Of The Architecture
	Section 5.5.2 Effect Of Activation Function
	Section 5.5.3 Effect Of Node Target Values
	Section 5.5.4 Use Of The E-Set
	Section 5.6 Conclusions
Chapter 6. Further Investigation Of The Scaly Neural Network Architecture
	Section 6.1 Introduction
	Section 6.2 Experimental Procedure
	Section 6.3 Simulation Results
	Section 6.3.1 Size Of Input Zone
	Section 6.3.2 Size Of Input Zone Overlap
	Section 6.4 Discussion
	Section 6.4.1 Size Of Input Zone
	Section 6.4.2 Size Of Input Zone Overlap
	Section 6.4.3 Size Of Hidden Layer
	Section 6.4.4 Output Node Target Values
	Section 6.5 Conclusions
Chapter 7. Conclusions And Future Work
	Section 7.1 Time Alignment Algorithm
	Section 7.2 Neural Network Architecture
	Section 7.3 Other Features Of The Neural Network
	Section 7.4 Output Node Target Values
	Section 7.5 An Optimal Speaker Independent Speech Recognition System
	Section 7.6 Future Work
Appendix A
References