Вы находитесь на странице: 1из 11

SPEECH RECOGNITION IN TAMIL LANGUAGE

A PROJECT REPORT

Submitted by

ISHWARYA.P (110815104027)
KARTHIKA LAKSHMI.A (110815104034)

in partial fulfillment for the award of the degree

of

BACHELOR OF ENGINEERING
in

COMPUTER SCIENCE AND ENGINEERING

JAYA ENGINEERING COLLEGE, THIRUNINRAVUR

ANNA UNIVERSITY : CHENNAI 600 025

APRIL 2019
ii

ANNA UNIVERSITY : CHENNAI 600 025

BONAFIDE CERTIFICATE

Certified that this project report ​“SPEECH RECOGNITION IN TAMIL

LANGUAGE”, ​is the bonafide work of ​“ISHWARYA P (110815104027)​ and

KARTHIKA LAKSHMI.A (110815104034)”, ​who carried out the project

work under my supervision.

Prof. M. KUMARAN ​Prof.G.BHARATHI MOHAN


Head of the Department, Supervisor & Professor,
Department of CSE, Department of CSE,
Jaya Engineering College, Jaya Engineering College,
Thiruninravur – 602024. Thiruninravur – 602024

\
​ iii

ANNA UNIVERSITY : CHENNAI 600 025

VIVA VOCE EXAMINATION

The viva-voce examination of the project work titled ​“SPEECH

RECOGNITION IN TAMIL LANGUAGE”, ​submitted by “​ISHWARYA .P

(110815104027)​ and ​KARTHIKA LAKSHMI .A (110815104034)”, ​held on

INTERNAL EXAMINER EXTERNAL EXAMINER


​iv

ABSTRACT

The project intention to create a way to speech description for the query in

search process in system. Analyze the query after user selection and then sent

outcomes to the recipient in Tamil language. There has not been much

consideration for Tamil language to be used in voice process. Automatic speech

would provide more flexibility in smartphone experience. Also people who have

only been used to their native language, it will feel easier to use ASR technique.

Automatic speech recognition have achieved a great success in many

application. Among them, template matching techniques like dynamic Time

Warping (DTW), statistical pattern matching technique such as hidden Markov

Model (HMM), Fourier series technique is used for better accuracy. For this

system, highest word recognition accuracy is achieved with HMM technique. It

offered 100% accuracy during training process and approximately 98% for

testing process.
​v

TABLE OF CONTENTS

PAGE NO
ABSTRACT iv
LIST OF TABLES vii
LIST OF FIGURES viii
LIST OF SYMBOLS, ABBREVIATIONS & DEFINITIONS x

1.INTRODUCTION 1
1.1 System Overview 2
1.2 Objective 2
1.3 System Study 2
1.4Organization of the Project 4

2.SOFTWARE REQUIREMENTS SPECIFICATION 6


​2.1 Over all Description 7
2.2 External Interface Requirements 8
2.3 System Features 9
2.3.1 Voice Deliver 9
2.3.2 Fast Fourier Transform 10
2.4 Other Non-Functional Requirements 11

3.SOFTWARE DESIGN 12
3.1 System Architecture 14
3.2 Decomposition Description 19
vi
3.2.1 Query as Speech 19
3.2.2 Speech Rate Category 19
3.2.3 Rate of Speech Manipulation 20
3.2.4 Time Normalization 20
3.2.5 Syllable Based Speech Process 20
3.2.6 Outcome as Speech 22
3.3 Data Design 21
3.4 Component Design 22
3.5 Human Interface Design 26

4. IMPLEMENTATION 29

5. TEST PLAN AND TESTING 38

6.RESULTS AND OBSERVATIONS 44

7.CONCLUSION AND FUTURE WORK 46

APPENDIX 48

REFERENCES 51
vii

LIST OF TABLES

S. No Table No Description Page No


1 2.1 Search Process 10
2 2.2 Frequency Range 11
3 3.1 Data Dictionary 21
4 5.1 Unit testing for MATLab IDE 40
5 5.2 Unit Testing for Search process 40
6 5.3 Unit Testing for Query Match 41
7 5.4 Integration Testing 42
8 5.5 Validation Testing 43
viii

LIST OF FIGURES

S.No Figure No Figure Names Page No


1 3.1 System Architecture 14
2 3.2 Use Case Diagram 15
3 3.3 Class Diagram 16
4 3.4 Activity Diagram 17
5 3.5 Sequence Diagram 18
6 3.6 Collaboration Diagram 19
7 3.7 Component Design 22
8 3.8 Screen Image - Frequency Match 26
9 3.9 Screen Image - Noise 27
Classification
10 A1 Frequency Match 49
11 A2 Noise Classification 50
​ ix
LIST OF SYMBOLS, ABBREVIATIONS AND DEFINITIONS
SYMBOLS

: External Entity

: Decision

: Flow of Data

: User/ Actor

: Process in use case

: State

: Initial State

: Final State
x

ABBREVIATIONS

HMM Hidden Markov Model

DTW Dynamic Time Warping

ASR Automatic Speech Recognition

NN Neural Network

MFCC Mel Frequency Cepstral Coefficient

SVM Support Vector Machine

DEFINITIONS

HMM

Hidden Markov Model is a statistical pattern matching technique. The

state is not directly visible , but the output ( in the form of data or “token” in the

following), dependent on the state, is visible.

DTW

Dynamic Time Warping is one of the algorithms for measuring similarity

between two temporal sequences, which may vary in speed.

Вам также может понравиться