Sie sind auf Seite 1von 13

i

BILINGUAL SPECCH RECOGNITION FOR TAMIL AND INDIAN ENGLISH

By J. JEFFIN GRACEWELL Reg. No. 1042242

A PROJECT REPORT Submitted to the

FACULTY OF INFORMATION AND COMMUNICATION ENGINEERING In partial fulfillment of the requirements for the Phase-I project of MASTER OF ENGINEERING IN COMMUNICATION SYSTEMS

AUTC
ANNA UNIVERSITY OF TECHNOLOGY CHENNAI CHENNAI 600 113 December 2011

ii

BONAFIDE CERTIFICATE
Certified that this project report titled Bilingual speech recognition system for Tamil and Indian English is the bonafide work of Mr.JEFFIN GRACEWELL. J (Reg. No: 1042242) who carried out the project under my supervision. Certified further, that to the best of my knowledge the work reported herein does not form part of any other project report or dissertation on the basis of which a degree or award was conferred on an earlier occasion on this or any other candidate.

SUPERVISORS Ms.B.RAMANI, Assistant Professor, Department of ECE, SSN College of Engineering, Kalavakkam 603 110

HEAD OF THE DEPARTMENT Dr. S. RADHA, Professor & Head Department of ECE, SSN College of Engineering, Kalavakkam 603 110

Dr. P. VIJAYALAKSHMI Professor, Department of ECE,


SSN College of Engineering, Kalavakkam 603 110. Submitted for the Phase-I Project Viva- Voce examination held on .

Internal Examiner

External Examiner

iii

ABSRTACT
Speech recognition systems are speech to text conversion systems. In India there are more than 30 languages spoken across the country. Developing a multilingual recognition system for all the languages need enormous resources in terms of data collection from the entire locality and processing it. The way the words are pronounced also differs for a language in different regions. Hence developing a recognition system based on adaptation to their utterances is also a tedious one. In this project a bilingual speech recognition system is to be developed for two languages namely Tamil and Indian English along with language identification. The system first identifies the language followed by the monolingual speech recognition. i.e., from the given unknown utterance (speech signal), the system has to recognize the language of utterance, incorporate the required acoustic & language models, and recognize the utterance. The language identification is based on Gaussian mixture modelling and the monolingual speech recognition system is based on hidden Markov model. The system that is being developed is speaker-independent and domain specific on the

kanipori(computer). In this project, so far Tamil speech recognition system has been developed. The data collection is done and preprocessed for three male speakers and three female speakers. The performance of the developed recognition system showed the correctness of 80.11% and accuracy is 74.78%.

iv

ACKNOWLEDGEMENT
I thank ALMIGHTY GOD who gave me the wisdom to complete my first phase of my project. My sincere thanks to our beloved founder Mr. Shiv Nadar, Chairman, HCL Technologies. I also express my sincere thanks to our Principal Dr. S. Salivahanan, for all the help he has render during this course of study. My heartfelt gratitude goes to Dr. S. Radha, Professor and Head of the Department, ECE for her words of advice and encouragement and I also express hearty gratitude to project Co-ordinators Dr. P. Vijayalakshmi, Professor, Dept of ECE and Dr. E.Janardhanan, Professor, Dept of ECE and professors of the department for their scholarly guidance. I am deeply obliged and indebted to the timeless help and guidance provided by Ms.B.Ramani, Assistant Professor, ECE and Dr. P.Vijayalakshmi, Professor, ECE, and also express my heartfelt thanks for making this project a great success. I also thank all the faculty members of the ECE department for their kind advice, support and encouragement and last but not the least I thank my parents and my friends for their moral support and valuable help.

TABLE OF CONTENTS CHAPTER NO. TITLE PAGE NO. ABSTRACT (ENGLISH) ACKNOWLEDGEMENT TABLE OF CONTENTS LIST OF FIGURES LIST OF TABLES LIST OF PHONEMES LIST OF ABBREVIATIONS 1 INTRODUCTION 1.1 Overview 2 LITERATURE SURVEY 2.1 Introduction 3 THEORY AND IMPLEMENTATION OF HIDDEN 3.1 Introduction 3.2 Discrete time markov processes 3.3 Extension to hidden markov models 3.3.1 The urn and ball model 3.3.2 Elements of an model 3.4 The three basic problems for HMMs iii iv v viii ix x xi 1 1 3 3 6 6 6 7 7 8 9

vi

3.4.1 Problem 1 3.4.1.1 Probability evaluation 3.4.1.1.1 The forward procedure 3.4.2 Problem 2 3.4.2.1 Viterbi algorithm 3.4.3 Problem 3 3.4.3.1 Baum Welch reestimation 3.5 Types of HMMs 3.6 Continuous observation densities in HMMs 4 TRAINING PHASE 4.1 Introduction 4.2 Experimental setup 4.3 Preperation of speech data 4.3.1 Recording of the speech data 4.3.2 Feature Extraction 4.3.3 Manual Segmentation of the data 4.3.4 Forced Viterbi Alignment 4.4 Generation of monophone HMM

9 9 10 11 12 12 13 14 15 16 16 16 17 18 19 21 23 24

vii

4.4.1 Proto generation 4.4.2 Generating monophone Model 4.4.3 Realigning the training data 5 TESTING PHASE 5.1Preperation of dictionary 5.2 Recognition of the speech data 5.3 Running the recogniser live 6 RESULTS AND CONCLUSION REFERENCES

25 28 28 32 32 33 35 36 40

viii

LIST OF FIGURES

FIGURE NO. Figure- 3.1 Figure- 3.2 Figure- 4.1 Figure- 4.2 Figure- 4.3 Figure- 4.4

TITLE Ergodic HMM Left-to-right HMM Speech waveform for different speakers Parameters of the configuration file Map Table Segmented data with the spectrogram of the waveform

PAGE NO 14 15 19 20 21 22

Figure- 4.5

Lab file generated for the segmented waveform

22

ix

LIST OF TABLES
TABLE NO. Table- 4.1 TITLE Speakers and the duration of recorded data Table- 4.2 Table- 4.3 An example of an MLF file. Proto generated for 3 states and 5 mixture components Table- 4.4 Table- 5.1 Model generated for a phoneme Recognized transcription for speaker Jeffin Table- 5.2 Recognized transcription for the speaker Aasha Table- 6.1 Table- 6.2 Table- 6.3 Result for the speaker Jeffin Result for the speaker Aasha Result for all the speakers 37 37 38 35 30 34 23 26 PAGE NO 18

LIST OF PHONEMES (for Tamil)

A ai b ch dh e E g h i I id ing j M N nn ll N o

k L O P r R Sil th tr u U v y

xi

LIST OF ABBREVIATIONS

HMM Hidden Markov Model HTK HMM tool kit

xii

xiii

Das könnte Ihnen auch gefallen