Arabic Voice Recognition
Abstract
Speech recognition is being the interest of scientists for more than four decades. Arabic speech recognition was an interest of Arabic scientists but they couldn’t reach a good performance until now. This is because they don’t have an existing speech dataset that they can build there systems on it.
In this project, we are going to build a Speech Recognition system using the HTK tool. We are also going to build a speaker-independent speech dataset that can be used in our project to train our system and to validate it. This speech dataset can be helpful in other Speech Recognition researches. We will also run our recognizer interactively so that we can make use of it in information retrieval applications
Table of Contents
Part1: Introduction 
Part1
1.1.1 Vocabulary Size
1.1.2 Speaker Dependency
1.1.3 Continuity of Speech
1.1.4 Background Noise
1.2 Problem Statement
1.3 Contributions
1.4 Software Development Life Cycle Model
Part2: Speech DataSet
(Database) Part2
2.1 Previous Work
2.2 Data Collection
2.3Dataset Statistics
2.4 Support Tools
Part3: Hidden Marcov Model
(HMM) Part3
3.1 Discrete Markov Process
3.2 Extension of Hidden Marcov Models
3.3 Elements of HMM
3.4 The Three Basic Problems of HMM
Problem 1:
Solution :
Problem 2:
Solution:
Problem 3:
Solution:
3.5 Types of HMM
3.6 Optimization Criteria
4. Hidden Marcov Model Toolkit (HTK)
4.1 Simple HTK Example
4.2 Difficulties in HTK
Part4: Modeling and Labeling Techniques Part4
5.1 Example on Word Level Labeling Using HTK
6. Speech Recognition Different Labeling Schemes Experiments
6.2 Phonetic-Based Labeling
6.3 Literal-Based Labeling (Old Version)
6.4 Literal-Based Labeling (New Version)
6.5 Comparison Between The Four Methods
Part5:Results of Word-Level Labeling Part5
8. Results of Literal-Based Labeling
9. Results of phonetic-Based Labeling
10. Conclusion
10.1 Future Work