Noise Cancelling for Robust Speaker Identification Using Least Mean Square

(1)

ｾＬ＠｣ｯｳｔ･｣ｨ

ｾ＠

PROCEEDING

1 st

International Conference on Science and

Technology for Sustainability

Volume I. October

2014

UIN !>lJ'>l\A KIAlJ

Organized by

Faculty of Science and Technology

Universitas Islam Negeri Sultan Syarif Kasim Riau

In Cooperation with

IEEE Indonesia Section

I Page:

275 + x hllp://w\1 w.icostechs.org

The Commictee of ICoSTechS is not responsible for the content of articles that published in this proceeding.

(2)

1Sl International Conference on Science and Technology for Sustainability

Proceeding, Volume i, October 2014

ISSN 2356-542X

PREFACE

·i

cosTechs)

The I st International Conference on Science and ｔ･｣ｨｮｯｬｯｧｾ＠ for Sustainability 210.J (ICoSTechs201.J) is an international event hosted b) Facult) of Sciences and Tcchnolog). Uni' ersity of Islam Negcri Sultan S) arif Kasim Riau (UIN SUSKA Riau). The purpose of this conference is to pro' idc a forum for researchers. scientists and engineers to exchange nc'' ideas and interact in-depth through discussion with peers from all over the world in the fields of elcctric:il and ckctronics engineering. informatics. mathematics and industrial engin..:cring. The maio goal of this C\ ..:nt is to facilitat.: communications among researchers and practitioners. not only concerning the core areas but also invoh ing multi -<lisciplinary and interdisciplinary \\Ori...

We are gratd'ul to all those "ho have contributed to the success of ICoSTechs2014. There are a number of plll'ties that have assisted us in organizing this conference become a reality. We would like to thank all authors, participants. faculty members for their participation and support. IEEE Indonesia sections. lllT (The lntern:itional Institute of Islamic Though) nnd BKS PTN Barat (State Uni\ersities Cooperation Agency of Western Region). Last but not least. we greatly appreciate the committees and external re' icwcr's precious and timely revie\\S. Their c:\pcnise is very vital in ensuring Lhe success of this event.

We really hope that all participants benefit tremendously from the conference. Finally. "e would like to "ish the participants success in the presentations and social networking.

Dr. Alex Wenda

General Chair

The 1st International Conference on Science and Technology for Sustainability 210.J. (ICoSTechs201.J) Batan1, Indonesia

(3)

111 _{lntematlonal Conference on Science and Technology for Sustainability}

Proceeding, Volume l, October 20 14

·i

cosTechSJ

ISSN 2356-542)(

WELCOME

First. blessing and mercies so "e can be here together in this room 10 healthy condition. And we also convey our

syallawat to Prophet Muhammad.

ｎｯＬｾ｡､｡ｹｳＮ＠ sustainabilit} has been an emerging major issue of thc \\Orld in order to create a Ckuer Life for the Current and Future Generations. Achieving a Sustainable Oevelopm..:n1 will require changes in many fields. including Sciences and Technologies.

More than one hundred definitions of sustainable development exist, but the most '' idcly used one is the definition from the Brundtland Report of the World Commission on Environment and Development. presented in 1987. It states that sustainabk development is the "Development that meets the needs of the present without compromising the ability of future generations to meet their own needs.'' Sustainable development promoh:s the idea that social. environmental. and economic progresses arc all attainable within the limits of our earth's natural resources. Sustainable developmenl approaches everything in the world as being connected through space. time and quality of life.

Sustainable development constantly seeks 10 achieve the social and economic progresses in ''il)'S that will not exhaust the earth's finite natural resources. The needs of the world today are real and immediate. yet it is necessary lo develop ways to meet these needs that do not disregard the future. The capacity of our ecosystem is not limitless. meaning that future generations may not be able to meet their needs the way we arc able 10 now.

The world's resources are finite. and the growth that is not well-managed nor unsustained would lead to an increased poverty and declined condition of the environment. We O\\e it to the future generations to explore lifestrles and paths of the development that would effectively balance the progress with an awareness of its environmental impacts. In order to preserve the future, we must appreciate the interconnectedness bet\\een humans and nature at all levels. Sustainable Science and Technology practices may help us do this. and through education. research and building awareness, preserving the future is within evel')one's reach.

I

http:/ /www.icostechs.org

(4)

1" International Conference on Stlence and Technology for Sustalnability

Proceeding, Volume i. October 2014

ISSN 2356-542X

·i

cosTechs1

Th.: I SI lnh:rnational Conf.:rcncc on Scicm:c and l'cdmolog) for Susiainabilit) ( ICoS I cch5201 4) 20 14 otkrs a place and opportunitic..-s for n::scarchcr.; and profossionals from academic. business. industries. Go\l:mments. '!GO:.. and other s<:ctor.> to c\;change their scientific aml technological infonnation. I CoS I'cchs201 4 is also pro' ided for students to pr¢Sc:nt their ｲ･ｾ｡ｲ｣ｨ＠ paper...

<\II submissions "ill Ix pet:r re\

ji:,,

c..'ti. :\ccepteJ papers "ill be publish.:d in th.: cont<:rence procceJing:.. Sckcted Pap.:r with a tew corrections "ill lx propose to published in IEEE Xplort Digital Library. ｔｅｌｋｏｾｉｎｉｋａ＠ J ournal (lndu by SCOPUS and ISi). and IAES Journals (Institute of Advanctd Engineering and Srience Jou ma ls).

On this occasion I \\ish to thanJ.. th.: Rt:ctor of UIN SUSKA Riau \\hO had agn:t:d to auend on this occasion and be

apridt: for us on his presenc.:. and \\I! also appeal to the Rector in order to open a in1cmmional seminar \\JS offo:iall).

Thank )OU \Cr) much to Prof Aluander Jakob Boris Zthnder O\cr a gi,en time. hopct'ull) on the other occasion \\e

ma) ｩｮｾｩｨＡ＠ you bacl .. Thanks also 10 IEEE Indonesia Stction. BKS-PTN Barat - Enginttring. lottrnational lnstitutt of Islamic Thought (lllT).

Thanks 10 1hc imitations has the pleasurc to pres.:nt. thanks to th.: chairman and thc entire comm inc.: "hich has prepared this activities. and also to thc all of participants in this scminar. ｨｯｰ｣ｴＧｵｬｬｾ＠ be bendicial to all of us. Congratulations and hop< that with the implementation of this seminar mal...:s our rok in 1h.: \\Orld of science and technology more cf earl)' 'isible. so our pn:scncc is ｴｾｬｴ＠ by the community. So any rcmarks from mc, linally I apologiLc

if there an! words 1hat are no1 in place. Congratulations.

Ora . Hj. Yenita Morena, M.Si

Dean

faculty of' Science and Tcchnology

Uni\ersitas Islam Ncgcri Sultan S)arifKasim Riau

Cop)Tight ©October 2014, Faculty of Science and Technology, UIN Sultan Syarif Kasim Riau. lCoSTechS

I

http://www.icostechs.org

(5)

1• International Conference on Science and Technology for Sustalnability Proceeding, Volume l, October 2014

ISSN 2356-542X

Committee

Honorary Executive Committee

•1

cosTech'SJ

• ProC Dr. H. Mundzir Hitami, M.A (Rector oflJni\el"iitas blam Negcri Sultan ｓＩ｡ｲｩｦｬＭＺｾＱｭ＠ Riau) • Prof. Dr. Ir. Djoko Santoso. M.Sc (Director of DGllE. the Ministry of National Education. Republic of

Indonesia)

• Prof. Dr. Ir. Satf)O Sol!mamri Brodjonegoro (Visiting Proti:ssor. TUT Japan and ITB lndont.-sia)

• Prof. Dr. dr. S}ahril Pasaribu. DTM&l I, M.Sc (CTM) Sp. A(K). {Rector of University of Sumatera Utara) • Prof. Dr. Ashaluddin Jalil. MS (Rector o f Uni\ersit) ofRinu)

• Prof. Akhmaloka, Ph.D (Rector of Institute ofTechnology Bandung) • Prof. Dr. Ir. Djoko Suhano (Institute of Technology Bandung) • Prof. Dr. Samsul Rizal. M.Eng (Rector ofUni\ersit) ofS) iah Kuala) • Prof. Dr. Badia Pcri.tadc. MBA (Rector of Uni\crsit) ofSriwijaya) • Dr. H. Werry Darta Tai fur. SE. MA (Rcctor of Unin:rsit} of Andalas)

International Advisory Board

• Prof. Emeritus Hiromi I lomma (Professor Emeritus Toyohashi, Uni\ersit) ofTechnolog) Japan) • Prof. Dr. Magnus Jaeger (Ostbayerische I echnischc I lochschule (OTI I). Weidcn)

• Prof. Masashi Daimaruya (Muroran Institute of Technology. I lokkaido. Japan) • Prof. Masanori Kikuchi (Science Unhcrsity ofTok)o. Japan)

• Prof. Dr.Ing. Peter Dietz ( lnstitut fur Mascincnwescn der Tcchnischen Universitet. Germany} • Prof. Dr. Bustami Syarn (University ofSurnatera Utara)

• Prof. Yasuhiro Kanto (lbaraki University, Japan)

• Prof. Ahmad Kamal Ari fin (University Kcbangsaan Malaysia. Malaysia) • Prof. lchsan S Putra (Institute of Technology Bandung)

• Prof. Jamasri (Uni,·ersity ofGajah Mada) • Prof. Dr. Samsul Rizal (Uni\'ersity of Syah Kuala)

• Prof. Or. Aini Hussain (Universiti Kebangsaan Malaysia. Malaysia)

• Ora. 1 lj. Yenita Morena. M.Si (Facult) Sains and Technology. UIN SUSKA Riau, Pekanbaru, Indonesia)

Conference Chair

• Dr. Alex Wenda (Faculty Sains and Technology. UIN Suska Riau, Pekanbaru. Indonesia)

Conference Co Chair

• Dr. Ladda Suanmali (Suan Dusit Rajabhat University. Thailand) • Dr. Fahrni Arif(Universiti Teknikal Malaka. Malaysia) • Suherman, Ph.O (University ofSumatera Utara)

(6)

1" International Conference on Science and Technology for Sustainability

Proceeding, Volume 1, October 2014 ISSN 2356-542X

Technical Program Commi tt ee

·i

cosTechs)

• Dr. Okfalisa (Fawlt)' Sains und Tcchnolog). UIN Suska Riau. Pekanbaru. Indonesia) • Muhammad Ar) Mun). ST. MT ( T di.. om ｬｮｾｴｩｴｵｴＮＺ＠ of T cchnolog). Indonesia) • Dr. Noritm1ati (Uni,crsiti Tcknologi Mara (UiTM). Malaysia)

International Program Committee

• Assistant Proti:ssor Dr. Ali Ahmed A.Ab<lclrahim (Karat) Uni\ ersit). ｏｭ､ｵｲｭ｡ｮＮＱＲＳＰｾＮ＠ Sudan) • Dr. Mohammc.:d Abdulatef Ali Al-Shargabi (Najran Uni\'crsit). Kingdom of Saudi Arabia) • Dr.Dcwi Nasicn (Uni\'crsiti Tcknologi Mala} sia. Johor. Mala)sia)

• Dr. Mohammed Abdulatd Ali Al-Shargabi (Najran Univcrsit). Kingdom of Saudi Arabia) • Assistant Proffcssor Dr Ahmcd Ham.c.a Osmun r\hmcd (International Uni,crsit) of Africa) • Pro[ Ali Sc lam at (Uni' crsiti T cknologi ｾＡ｡ｬ｡Ｉ＠ sia. Johor. ｾＡ｡ｬ｡Ｉ＠ｾｩ｡Ｉ＠

• Prof.Zaki Abu Bakar (Univc:rsiti Teknologi Mala)sia. Johor. ｾｦ｡ｬ｡ｹｳｩ｡Ｉ＠

• Asst Prot:Dr. Sakesun Suthummanon (Prince ofSongkhla Uni\crsit). lladyai. Songkhla. Thailand) • Dr.Kanya Auckaraarcc (Prince ofSongkhla Universit}. lladyai. Songlhla. Thailand)

• Dr.lndris Sulaiman (lnstitute-Chulalongkom Univ.-Tll & Adjunct Fellow. ANU·CECS & Encrg) Change Institute. Australia)

• Assoc. Prot: M. Zamcri Mat Saman. M.Eng (Univcn.iti Teknologi Mala)sia)

• Prof. Dr. Abdul Hakim Halim. M. Eng (Institute ofT echnolog) Bandung (ITl3). Indonesia) • Dr. Rosniwati(Univcrsiti Sains Malaysia (USM))

• Dr. Mohd As)raf Zulkilley (Universiti Kcbangsaan Mala)sia (UKM )) • Dr. Teddy Pumamirza (UIN Suska Riau. Pekanbaru, Indonesia) • Dr. Rado Ycndra (UIN Suska Riau. Pckanbaru. Indonesia) • Dr. Eng. Listiani Nurul I luda. MT (Univ..:rsily ofSumatcra Utara) • Prof: Dr. Ir. Rahim Matondang. MT (University ofSumatera Utara) • Dr. Ing. lkh" ansyah lsranuri (Uni\ ersity of Sumatera Utara) • Dr. Eng. lndrn Nasution. MSc (Univcrsit)' ofSumatcra Utara) • Dr. I limsar Ambarita (Uni\'crsity of Sumatera Utara) • Suherman, Ph.D (Unhersity ofSumatcra Utara) • Dr. Tnslim. MT (University ofSumatera Utara)

• Ir. Bauni Hamid, M.DcsS. Ph.D (Uni\ersity ofSumatera Utara) • Ir. Dwira Nirfalini Aulia. MSc .. PhD (University ofSumatera Utara) • Dr. Benny 0. Marpaung (Uni\ersit) ofSumatera Utara)

• Prof. Johanes Tarigan (University ofSumatera Utara) • Prof. Dr. Ir. Roesyanto. MSC (Unhersity ofSumatera Utara)

(7)

111 _{lntematlonal Conference on Science and Technology for Sustalnablllty} Proceeding, Volume 1, October 2014

ISSN 2356-542.X

ICoSTechs 2014 Schedule

Wed, October 22nd 2014

Conference

Day

07.30- 10.10 OJXning Ceremony

CotTec Break

10.10-12.20 Session I Presentation from Keynote Speaker

Prof. Zehnder Alexander Jacob Boris

Moderator: Kunail1. S.T. M.Sc

13.30-15.30 Session 11 Paralel Presentation

Moderator Room A: lsmu Kusumanto, M.T

Moderator Room B : Rika Susanti, S.T. M.Sc

Coffee Break

15.30 - 16.00 Closing Ceremony

·1

cosTech°S)

(8)

1• lntematlonal Conference on Science and Technology for Sustainability Proceeding, Volume i, October 2014

ISSN 2356-542X

·i

cosTech'S>

Contents

Preface ... ...

11

Welcome .... ...

11 1

Committee ... .. . ... . . .. .. ... . . .. ... . .. ... ... .... . . .. . . . ... ... . . . .. . .. .. . . ..

v

ICoSTechS 2014 Schedule ... ...

v1 1

Zaina Norhallis Zainol, Masiot Mad Tap, ｾＱ｡､＠ Rtbi Abdul Rani Pilot Study Thermal Ergonomic for Fir..: Fighters In Mala)sia

2 Muhammad S ufi, Jalil Ali, Sa ktiolo 8

Chemical Kinetic Modd in Equilibrium Stat..: Thermal Plasma of Carbon Ion SJ)\!dcs

3 Fahrul Agus, Rahmat Sholth, lltliza Rahmania Hatta, Tarbi)atul l\lunawwarah 15

Fu1.zy Anal) tical Hierarchy Proco.:ss for Land Suitability Analysis Compared to Analytical Hierarch) Process

4 Imam Tahyudin, Brrlilana Ｒｾ＠

The Effectiveness Analysis of Application of Analytical Hierarchy Process (AHP) Method in Decision Support Sys1em for Emplo)ee Selection

5 Andang S unarto, J . Sulaiman, A. Sa udi 35

ａｰｰｲｯｸｩｭｾｬｬ｣＠ Solution For The Time-Fractional Diffusion Equations Using Quartcr-S110.:cp Gauss-So.:idcl l\kthod

6 ustari Handayani, lwan lskanda r, Wtli Andrian

n

Analysis and Implementation of the Kohoncn Neural Net\\'Ork for Arabic Character Recognition

7 Rah mad Kurniawan, Okfalisa, Mohd Zakrtt Ahmad Natri SI

A Comparative Study Of Decision Trc.:. Artificial Neural Nell\ork and Rough So.:t Theory On Mulli1ariate Data Set Characteristics

8 Abdul Rahman Htmdi, M. 1\1. i\lahadzir, l\luhamad, Salwa Mahmood, l\luhamad Zamtri i\lat 58

Saman, Safian Sharif, Mtlfa Yola

Fuzzy Based Sustainability Assessment: Case Study of Passenger Car

9 Azriyenni, M.W. Mustafa 68

Identification Fault In Transmission System Utilize Neuro-Fuzzy Systems

10 Imam Tahyudin, Moha mmad lmron, Sill Alvi Solikha tin

Decision Support System Using Data Mining Method for A Cross Selling Strategy in Retail Stores

I I Mohamad Satori, Rtni Amaranti, Endang Prasttyaningsih

Development of Gasification - Pyrolysis Technology using Residual Household Waste 10 Produce Alternative Energy

12 Aji Prasetya Wibawa, Andrew Nafalski

Javanese Classifier and Language Selector for Statistical Machine Translation (SMT)

I

http://\\'ww.icostechs.org

73

79

87

(9)

1• 1ntematlonal Conference on Science and Tethnology for Sustainability

Proceeding, Volume 1, October 2014

ﾷｩ｣ｯｳ

ｔ･｣ｨｾ＠

ISSN 2356-542X

13 Dian Mursyitab 9.t

Design PIO Sliding Surface For Sliding Mo<lc Controller

14 Zulratri Aini 100

Optimal Reacti\'e Po,, er Flo" through Regulation of Voltage Control Vuriabk

15 Otdi lrawan, lsmu Kusumanto, Saktioto, Jalil Ali 106

Theoretical Optimization of Coupling Coenicient for Weakly Coupled Fib.:r Coupler

16 Angraini, Yozita ｏ･ｾｩ＠ 112

Audit of Hospital Management System (HMS) at RSIA Andini of Pekanbaru Using COBIT Frame\\Ork .t. I

17 Nanda Putri Miefthawati, Mohammad Arir 118

Autonomous Pick and Place Mobile Robot in a I.EGO NX

r

18 Adhatus Solichah A., Owi Sunaryono, Riyanarto Sarno, and Andre Vittorio Alloruung 125 Customer Relationship Management Application on Enterprise Resource Planning System

19 Elin Haerani, Samsinar 13 I

Interactive Map To Detenninc The Location Transmigration With Fuzzy AHP

20 Oewi Fitria, Miklas Scholz and Gareth Swift 135

Impact of Rapid Mixing Velocity and Rapid Mixing Time on Sludge Dcwaterabilit;.

21 Rice Novita, Fadli Gunawan l.t5

Web Service System Data Push And Pull PDPT (Higher Education Data Oases) (Case Study: UIN SUSKA Riau)

22 Fanny Camelia, T. L J. Ferris, Merry Sisk2 155

The Need for Introducing System Engineering to Industrial Engineering Students

23 Oifana Meilani, Merry Sisk.a, Feli Elysa 165

Accelerate Out Patient Health Services with Lean Sen ice

24 Rado Yendra, Ari Pani Oesvina, Rahmadeni and Abdul Aziz Jemain 175

fatreme Strom Duration Modelling at Peninsular Mala) sia

25 Wayan Firdaus Mahmudy 181

Solving Flexible Job-Shop Scheduling Problem Using lmpro\'ed Real Coded Genetic Algorithms

26 Meeawati, Kridanto Surendro 189

Design Model Incident and Problem Management Based Integration Of ITIL VJ and COBIT 4.1

27 Teddie Darmizal, Benny IUnti 197

Analysis of Online Training System Using Ranti's Generic ISllT Business Value And Economic Value Added (Case Study: Bank Rakyat Indonesia)

28 Tengku Nurainun 206

A Multi Due Date Batch Scheduling Model on Dynamic Flow Shop to Minimize Total Production Cost

29 ldria Maita 214

Spatial Data Analysis For Development of Web Based Geographic Information S)'stem (GIS) (Case Study : Universiti Teknologi Malaysia (UTM)

(10)

111 _{Intern ational Conference on Science and Technology for Sustainability} Proceeding, Volume i, October 2014

ISSN 2356-542X

·i

cos Tech$)

JO Haviluddin, IUyntr Alfn d 22-t

Comparison of ANN Bad, Propagation T ..:..:hniqu..:s in Moudling Nct\\ork frnflic ActivitiC!)

JI \'ola J\lelfa, Ra hma ya oti Dina 232

Production Planning for PO\\ Cr Pole in PT. fa} a Scntrikon Padang Pariaman with Goal Programming t.kthod

32 Alu \hnd a 238

Utilization of Web-based Power Qualit) Anal) sis

33 Febi Ya oto, R. Joko J\tus r idb o 2-t3

P..:rfom1U11cc Anal) sis of line Maze Sol\ ing Algorithm on Cuncd and Zig-zag rra<.:k from Linc Maze Robot

34 lngg ih Ptrma na, A gus Buono, Bib Pa ruhum Sila la hi 247

Noise Cancelling for Robust Six:akcr luentilication Using Least ｾｬ･｡ｮ＠ Square

35 Ntsdi Evril) an R ozan da, Faisal Am ir 253

Designing an Application of Customer Relationship Managem..:nt

36 Sutoyo, Liliana 258

Frcquenc) Channel Management of I IF Rudio In Initial Implementation of ALE Stations Network Riau

37 T ed dy P umam irza 266

The Realization Study on the Reconfigurable Functions orRadial Linc Slot Arra) (RLSA) Antennas

(11)

1" lntematlonal Conference on Science and Technology for Sustainability

·i

cosTech'S>

ISSN 2356-542X

Noise Cancelling for Robust Speaker

Identification Using Least Mean Square

lnggih Permana

Depilrtmc:m of Information S)stc:m. Universitas Islam Nc:geri Sultan S)arifKasim Riau.

e-mail: inggihpcrmana'!?uin-suska.ac. id. inggihja) a Ci'gmail.com

Agus Buono

Department of Computer Science. Dogor Agricultural University e-mail: pudcsha1₁₁_yahoo.co.id

Bib Paruhum Sila lahi Dc:panmcm of Mathc:matics. Oogor Agricultural Univc:rsit). c:-rnail: bibparuhum I

fl?

yahoo.com

Abstract - This study focu sed on noise cancelling

using Least Muns Square (Ll\IS) for robust speaker identifiration. Noise ranrrlling ran be used to oHrcome sptaker idrntification diffirulties in recognizing noisy voicrs. This study used the LMS on data preprocessing. Based on experimental results, it can be concluded that speaker identification using LMS at data preprocessing produces higher arcuracy than without using LMS. The highest accurary of data that using the LMS is 84.64•/e whereas the data without using the LMS only 2. 11%.

Keywords - least means square, noise cancelling,

speaktr identification

I. INTRODUCTION

Speaker identification is part of the speaker recognition that aims to find out who is speaking with an existing voice (I ). Speaker identification can be done because no voice among people same. This is because the shape ohhe vocal tract, rhythm, intonation, pronunciation patterns, choice of vocabulaI) and so forth in humans are different (2). This studies focused on the speaker identification of the noisy voice. II is interesting 10

study because most of the research on speaker

identification has obtained good accuracy on the voice that ｩｮｴ･ｮｴｩｯｮ｡ｬＡｾ＠ is made '' ithout noise:. but not on the "oice that full of noise (2).

One of the main parts of speaker identification is feature extraction. The extraction of the features is needed to reduce unnecessary information on the speech signal captured by the sensor and con\erting important information from the speech signal into a format that is more simple and clear for further processing. One of the feature extraction method use

frequently is MFCC (Mel Frequency Cepstral Coefficient). The working of MFCC is based on the frequency difTc:rence can be captured by the human ear so that it can represent how people recei\C voice signals (3

J.

MFCC is often used because it is considered a better perfonnance than other methods. such as in the case of a reduction in the error rate. But

ｾＡｆｃｃ＠ has shortcomings. that is not ｾ･ｲＩ＠ resistant to noise (4). Therefore. in this study added ANC (Active Noise Canceling) in the preprocessing or data in order to resolve the issue. This will help MFCC to produce a better feature extraction. This is because the speech signal produced by the ANC is the speech signal that has undergone a reduction in noise. So the purpose of this study to create a model of speaker identification that robust to noise can be achieved.

ANC is an approach to noise reduction using a

(12)

111 _{lntematlonal Conference on Science and Technology for Sustainability}

ISSN 2356-542X

sccom.lut) noise source: that dc:structh cl) remo' cs Ufl\\antcd noise: (5). ANC method usc:d in thb stud) is the Lea.st Mcan Square (LMS). This method

''a:;

tirst proposed b) \\'idro" and Hoff (6). The LMS method becomi.:s ,·cry popular because of the low complc\it) and capabilities that promis.:.

Another important part in the :.peal.er identitication is the making of thc codebook. Codehook is 'oicc prints produced through a !raining of data (7 j. In this stud). Self Organizing ｾｬ｡ｰ＠ (SOM) is used as a tmining algorithm. SOM sum:ssfull) applied to high-dimensional data (8], "hich is the traditional m.:thod may not be able to do so. The results of i\ 1FCC might produc.: a high-dimensional data. dcpcnding on ho'' man) col!llicicnts arc determin.:d in the process of MFCC. This is the rcason for choosing the: SOM as a method to make codebook.

II. RESEARCH METIIOD

Broad!) speaking. this StUd) consistcd or t\\O puns. Illustration of these t\10 pans can be s.:en in Figure 1. The first part is the making of thi: cod.:book. Making codebook performed on training 'oicc data for each spcakcr. The voice da1a arc not given the noisi: before:. After that, the feature i:xtraction process doni: b) using MFCC. Vectors result from the MFCC process clustered by using SOM. Vectors of the SOM cluster results arc: used

as

a codebook.

' - - - - . . . - ! ｾｴｵｲＹＧｦｔＧｊｩｗＱ＠ "'"""'9

Ｈｾ｟ＮＬＬＮＬＮ＠

I - O I _{_....} _I

1_ - - - ｾｾ＠ ... ｾＭ J

Fig. I llUSlration of research method

The second part b the: similarity measurement. In this pan the data u.>ed is 11.-sting \Oicc:s. The 1oice data rue gi' en noise. Noise usc:d in this study is \\hite noisi: "ith 'arivus 'alues b.!1011 6.5 dD. A Her that. the: data prcproccssing is Jon.: b) eliminating noise using the

LMS mc:thod. Thc:n MFCC proce:.s p<rformc:d to product: a foature !!\traction. Not like: thc process of creating codebook. \l!ctors of MFCC ure not proci:ssi:d

b) using SOM. but dircctl)' be measured the distance to the codcbook b) using Euclidean distance. The: speaki:r th;u thc:ir codebook \CCtors gel the smallest distance to the vectors of test \Oices detcrmined as speakers represcnting 1·oil.:i: input.

Ill. VOICE DATA

Voice dutu used is cvc:r usc:d b> Reda [9J in their study. The voice data consists of 83 speakcrs. 11 hich are di' idcd into 35 ti:male speakers and .t8 male: speakers. The speakers an: Indian citi1ens of diffcrcnt backgrounds. Each speakcr has 5 'oice files in "av format. The voice fih: length is I to 39 seconds. The \1ords that speak by the: speaker is a random combination of numbc:rs. Ri:cording is <loni: on the phone using an IVR system {Interactive Voice Rc:sponse). Sampling rate used is 8000 I lz.

faperimc:nts in this study \\ill be performed in several combinations of parami:ters. At each combination of parameters oni: voice files that owned by each speaker will be used to create the codebook. At1er that. testing performed using all \Oice data. All voice data usi:d has been removed silent time. This is done 5 times

so

that all voice tiles for each speaker l!Vl!r be the data to create the codebook. For each experiment arc calculated the resulting accuracy. Aller all the experiments carried out in a combination of particular parameters. then computed the average of accuracy. This accurae} is used as the level of speaker identification ability.

(13)

• 1" lntemaUonal Conference on Science and Technology for Sustainability

•1

cosTech°S)

ISSN 2356-542X

IV.

MEL FREQUENCY CEPSTRAL COEFFICIENT

(MFCC)

This study uses MFCC designed ｢ｾ＠ Slaney [ 10). This MFCC is chosen because ha'e bet:n ha,ing ERRed (Equal Error Rate) and DCFopt (Decision Cost Function) are lower than other types of MFCC (11 ). This MFCC implements -10 tilters that handle frequencies from 133 Hz to 6854 Hz. The tilter is di\ ided into two parts. 13 pieces or filter of linear space (200-1000 Hz) and 27 pieces of logarithmic space (107 1-6400 Hz). The collection of the filter is called filter bank. Illustration of filter bank can be seen in Figure 2.

t+

...

.

,

__

..,,.

-....

_,

.

' -

--

,,,

Fig 3. lllust11111on ofMFCC stages

Based on Figure 3. in general MFCC has 5 steps. The first stage should be done are dividing the incoming signal into several frames. This stage is called the frame blocking. The second step is smoothing each frame by using hamming windows. This is done to eliminate signals that are not continuous on the frame. The third step is changing all the frames from the time domain to the frequency domain. This is done by using FFT (Fast Fourier Transform). The fourth step transforms the frame into a met scale. This is conducted by calculating log10 result multiplication o f

frames and filter bank. The last step returns the voice frames from the frequency domain to the time domain using OCT (Discrete Cosine Transform).

v.

SOM (SELF ORGANIZING MAP)

SOM (Self Organizing Map) was first offered b)

Kohonen ( 12). SOM or also known as Kohonen. is one

ｴｾｰ･＠ of ANN (Artificial NeurJI Network). which is unsupen ised lc-Jrning ｳｾ＠ !>h:ms. SOM assumes a 1opulllgical ｾＮｴｲｵ｣ｴｵｲ｣＠ bct\\een the clll)tcr unib. it i!I run

｢ｾ＠ a human brain but is not present in any other ANN.

SOM net\\ork onl) consists of t\\o la)ers, namel} the input layer and output layer. Each unit in the input layer is connected to all units in the output tarer.

TABLE I EXJ>Ut.\t[ST AL PAA .\METERS

Parameter Value

MFCC

frame length 11>••

overl3p 04

20

SOM

number of cluS1ers 9, 16. 25. 36. 49. 6-i. 81 and 100

radius of neighborhood 3

topology ｨ･ｾ｡ｧｯｮ｡ｬ＠

LMS

IC3mmg r.ue 0 l.O 3. 0 5. 0 7 and 0 9

In SOM training should be determined how many clusters to be produced, the type of topology, and the radius of neighborhoods. Afterwards. make units in the output la}ers as many as clusters desired and collate based on the type of topology. The unit is a vector that has the same dimensions with the dimensions of the data. Initialize weights on each unit with random values. For each input data. find the smallest distance to the unit as BMU (Best Matching Unit). Distance measuring method used in this study is the Euclidean distance. Update the BMU and its units are still neighbors. Updating the \\eights is done to make uni t closer to the data. Neighbors are updated depending on the t}pe of topology and adjacency radius that previous!} selected. The following formula for the update unit:

wlJ(new)

=

w"(old)+a[x, -w,₁(old)]

( I)

(14)

111 _{International Conference on Science and Technology}_for_{Sustalnabllity}

ﾷＱ｣ｯｳ

ｔ･｣ｨ

ｾ＠

ISSN 2356-542X

'' hc:rc: w is the: weight of the: unit in the output la) er. r is the: input data and a is the: learning rate:.

VI. LMS {LEAS r M EAN SQUARE)

ANC mc:tho<l usc:d in this stud) is the: LMS (Lcast Mean Square). LMS is appl) ing the: gradient descent. This method was ｬｩｲｾｴ＠ proposcd by \\'idro11 and I !off

(6). Stec:pest desccnt, \1hich is one method that impkments gradic:nt dc:sccnt actuallr 1>¢::n 1 c:r: good to gcncratc: optimal 11c:ights. but this mcthod rc:quirc:s a true: gradient at each stcp. LMS can 01 c:rcomc: thc:sc: shoncomings bc:causc LMS can instant!) estimate the gradient at each stc:p

16J.

L.MS nlgorithm is bricll) as follows:

IJ Create a filter and initialization the: \1c:ight (w) of the filtc:r. Sp<cify n 1 alue oft he: learning rate (a). l) Perfonn steps 3 to S if then: is an incoming 10icc

signal (d).

3) Calculate the anti-noise using thc follo11 ing c:quation:

M

Y,

=

L

w,111

ＱｾＱ＠

(2)

11hcrc y is anti-noise and u is reference noise. -4) Calculate the residual signal using the following

equation:

e,=d,-y,

(3)

where e is residual signal and dis a primary voice signal. Residual signal is the result of noise cancelling.

5) Change the weights using the following equation:

w _I(new)

=

w _I+ 2ae 11

1 I (4)

VII. EXPERl"IENTS <\ND RESULTS

Results and analysis prcsc:nted in this section arc a summllr) of C:\p<rimcnts IJ)ing 1 arious combinations of parameter values. The parameters used are listed in ·1 able I.

A. £ffec1s of {,'sing L.\/S

This section about comparing the accurac) of speaker idcntitication on nois)· 1oicc b<:tween uses and did not use the LMS on the preprocessing of data. E\perimc:nts \1ere chosen to represent this part is C\periment 11ith learning rate of LMS is 0.7. Graph of C\perimental results can b<: seen in Figure 4.

Accuracy of sp<aker identification did not use thc LMS in preprocessing of data is very low. v.hich only reached 2. 11 %. Accuracy becomes greatly increased aller using the LMS in data preprocessing. which reached 84.64%. This result shows that the noise removal by the ANC docs not eliminate: the human 1·oicc characteristics. Despite that. the aceur..ic) of speaker identification is still IO\\a than mice '' ithout

ｮｯｩｾ･Ｎ＠ the difference is about I 0%.

Figure 4 shows that the data processed b.>- LMS that the greater the number of cluster SOM tends to increase the accuracy of speaker identification. Despite that. the resulting improvement is not significant. DifTcrcnccs between the highest accuracy (I 00 clusters) and the lowest accuracy (9 clusters) is only 2. 77%. Increasing the number of clusters will increase the dimension of the data which will increase the complexity. This will make the calculation time of speaker idcntilication becomes increasingly longer. Due to the increase of the

number of clusters did not produce a significani

improvement of accuracy, so it is suggested to choose a relatively low cluster number.

(15)

1>l International Conference on Science and Technology for Sustainability

•1

cosTechS'>

ISSN 2356-542X

t

I

t

I

Compari son o f Speaker ldentln catlon Accuracy M FCC Coefficient 20, L M S Learning Rate 0.7

100 00 90 00 60 00 70 00 6000

50 00 4 0 00

30 00 2000 1000 000

9 18 25 38 49 8 4 81 100

• Wllt'I no••• I WI In Ou I l MS 2, 1 1 87 , 27 0118 I 02 1 1'4 0 116 0 84

• Wllt'I no••• I Wllt'I L M S 8 1 87 8::1811 84?:> 8381S 84 10 84 4 0 8 4 50 04 04 94 88 94 88 9 4 70 94 94

The number o f SOM clu••,..

O Wl-.Oul no1so I Without LMS 93 37 9 4 10 9494 !M 82

10000 0000 llOOO 7000 eooo 6000 4000 3000 2000 1000 000 ae a HI

025 0:16

a 49

aM

• 111 c 100

Fig 4 Companson of spc:iker ｩ､ｾＧｬｬｴＱｦｩ｣｡ｴＱｯｮ＠ accuracy

Effect of L M S Leam i ng Rate to Speaker ldentJflcatlon A ccuracy MFCC Coefficient 20

0 I 03 0 5

'"07 ｾｉｍｓ＠ 7705 11117

33 43 118 61 71176 113 IMI

33111 69 28 llO $4 84 22

33 13 118 43 llO 24 113 IMI

3265 6928 _llO" 6410

32115 ea ee eo. 1• 11440

33 31 118 :J7 eo eo 84 5e

3229 61170 llO 78 114 ...

L e unlng rate of L M S

F11t. S EITccr of LMS leamin1t rate

77113 llO 24 6030 ao eo

Ill),,

6024 llO 30 llO 411

B. Effects of LMS uarning Rate

This experiment to see the effect of learning rate (LR} LMS to speaker identification accuracy. The experimental results presented here arc the result of experiments on MFCC coefficients 20. The effects of LR can be seen in Figure 5.

Figure 5 shows. using LR 0. 1 the accuracy of speaker identification is low. only 33.61%. At LR

increased to O.J. Lhe accuracy has increased but is still below 70%. At LR increased to 0.5. the accuracy has reached greater than 70%. At LR increased to 0.7. all the speaker identification accuracy has been above 80%. If LR is increased again. the accuracy is decreasing. Increased LR did not increase the data dimension. Therefore advisable to choose the LR that produces the best accuracy. From the experiments

[image:15.622.28.542.35.542.2]

(16)

1o1 International Conference on Science and Technology for Sustainability

Proceeding, Volume

t,

October

2014

•1

cusTechS?

ISSN 2356-542X

ixrformed ｢｣Ｎｾｴ＠ LR 'aluc: obtained b 0. 7.

VIII.

CONCLUSION

13ased on the ｣Ｚｸｾｲｩｭ｣Ｚｮｴｳ＠ conducted it c:an be conduded that sixaka idc:ntilic:ution for the: ｮｯｩｳｾ＠

\Oic.: that u:.ing LMS at data prepr<Xessing producL""> u

higher ｡｣｣ｵｲ｡｣ｾ＠ than "ithout using LMS. The: difference: is very signi ti cant. the highest accuracy using LMS is 84.6.to/o "hile the: highest accurnc) 1ha1 "ithout using ｌｾＱｓ＠ onl) 2. 11 ｾｯ＠

lncrc:asing the number

or

ｓｏｾ＠ I clusti:r:. :.uccc:s:.full) increases ｡｣Ｚ｣ｵｲ｡｣Ｚｾ＠ but doc::. Mt proportional with incrl!aS<! of comple\it). Increasing. LR on the: LMS un til limib certain 'aluc: c:an im:rc:ase the accurac)

or

speaker idcntilic:a<ion but the accurac) tell buck alkr passing the: limit 1aluc. I rom the: C:\pcrimental results obtained best LR is 0.7.

In the future: studies are advised to trying the noise that comes from the: rc:ul \1orld to kno,1 the effect on

ｳｾ｡ｫ｣Ｚｲ＠ identification. So robustness h:chniqucs olli.:red in real applications can be kno\\ n.

ACKNOWLEDGMENT

Acknowledgemc:nts to Facult) of Science and Technology of State: Islamic Unh crsit) of Sultan Syarif Kasim Riau which had supported this stud). Ackno\\ ledgcmcnts to Aziz Rah mad. Li gar Sckar Wangi and Dhieka AHillia Llintana 10 imprO\C English \1riting of this paper.

RffFRF1'CES

( l I R Tognen '1111.1 D PulldlJ, "An O'c" tel\ of Spcalcr 11kn11fic1111on ａＧ｣ｵｲｯ｣ｾ＠ 11ml Robustness Issues:· IJ:'FI

('m11111untl .\>"rm> \fax11:1M.\ol 11, no 2, pp ＲｊｾｬＮＲＰＱＱ＠

(:!I T ｾｭｮｵｮ｣ｮ＠ anJ II l1 ... An O\cn1c" ot Tc\1-lnJcpnnlmt

Spcakcr Rc.:ognmon from Features to Supcrwcrors ·· Ｎ｜ｲ＾＼Ｇｾ＼ｨ＠

C ·umm1m1w111m. 101 S::! no I. pp 12-10, WIO

131 I. Mucb. M l3cg3m. 3/ld I l·lam,3/uth1 "Voice R(( ognouon ａｬｩ［ｯｮｴｨｭｾ＠ Us1ni; Mel ｾｲｃ｜ｪｵ｣ｮ｣ｾ＠ Ccpwal Cocffic:1ent ｜ｍｾｌＧｌ＠ 1

and O)nam11: Trmc W.1rping 10n\' 1 T<ehno4ucs:· J1111mttf oJ

C'ump111111x. 1 ol ::!. nu 3, Pl> IJ8-l-IJ, 2010

ｴｾｊ＠ v l)ai;• .inJ c Wclkl.:ns. ··on Dcs.:ns111tini; lhc Mel·

Ccpsuum to Spurious Spa:u.11 Components for ｒｯ｢ｵｾｴ＠ Spo:«h K...:ogn111on ," on !W 5 / •,.,, ｉｴＭＮｾＮＧ＾Ｎ｜ＯＧＬ＠ pp S2v-5J.?

( 51 1\ V Oppo!rthe1m. E Wcinsrcm E. K C Zani:•. M Feder. and () Gauger, "Smgk·Scnsor t\rn1c Norse C..uKcllatton, //;'/:/.

/T•llt•

s,.. ...

h """Au.Im /•n ... nunx 101 ::!. no 2. pp ::!85-290,

19'U

1()1 0 Wtdrow anJ M E Holl. "AJ.1ptl\C ｓＱＱＱｴ｣ｨｭｾ＠ｃＱｲ｣ｵＱＱｾ＠ ... in

/'I(,() 11'/:.\( '(},\ ( 'um.,•11111111. H.·wrd /'art II'. pp 96-104

Pl E W1snuJ1sa.stra .ind A Ruono. "Pcni;cnJllll ChorJ p:kLi Al3t Musil Grtill Meni;gunalon L'od.:Bool dcni;:in Tel.nil. H su;ils1 Cm MFCC," )11rna/ llm111/i llm11 K.,mp11tcr, 101 14, no I. pp ＱＶﾷｾＱＮ＠ 2010

(!II J Y:m. Y lhu. 11 lk. anJ ｾ＠ :-iun · Ｇ｜ｬｵｬＱＱﾷｃｯｮＱＱｮｧｾｮ｣Ｑ＠ <.'.is.::idmg ａｮＺｩｬｾｳＱｳ＠ ofSm.1rt (jnd O;i,,cd on Sell-Oli;anttmg

Mlp" //:'/:'/:' / ni111. /11/t1rm111u111 /.'"""''' • cttkl .\<TJll'll.1', \Ol 8,

no 4. pp 1>-16·"· ｾｯ＠ 13

(91 A RcJ:i. S l'ani11:im . .ind F Cuudl 11)1..: A Lo11-l'o't Remote 1\ncnd.1nc.: Tr3clini; S)Stcm for rx,clop1ni; lki;1ons· tn !UI I /'rt•: . .51/o AC'.\/ 11'11rk•lrt>p 011 ,\',•11wrk."I SJ,,, m.• /or 1>.. .. ···l11r1nx u,·i:mm. pp 15·20

1101 M Slane) . "Aud1tol') Toolbox." lntmal Research Corporauon.

T cch Rep, 1998

I 11) T G:1nchc1, N Fakotal1s. and G K0Ll1nal1s. ··comp;ir:111'e E1alua11on of Vanous MrCC Implementations on The Sp.:aker Vcrttic:iuon T;uL;· in !fill.I l'ro1.· /,117/:HA fAC i, pp 191-194 ( 111 r Kohonen. "Sclf-Org:in11cd Form:111on ofTopoloi;1c:1ll>

Com-ct Feature Maps:· Hwliii:1wl C)/t.:m.:1t<'.<. \'OI ·U. no I. pp 59-09. 1981

I

http://www.icostechs.org

Noise Cancelling for Robust Speaker Identification Using Least Mean Square

ｾＬ＠ ｣ｯｳ ｔ･｣ｨ

Sponsored by

I Page:

ISSN 2356-542X

PREFACE

Dr. Alex Wenda

cosTechSJ

ISSN 2356-542X

cosTechs1

Ora . Hj. Yenita Morena, M.Si

ISSN 2356-542X

Committee

International Advisory Board

Conference Chair

ISSN 2356-542.X

ICoSTechs 2014 Schedule

cosTech°S)

ISSN 2356-542X

cosTech'S>

ﾷｩ ｣ｯｳ

ISSN 2356-542X

ISSN 2356-542X

cos Tech$)

cosTech'S>

ISSN 2356-542X

Noise Cancelling for Robust Speaker

I. INTRODUCTION

ISSN 2356-542X

Ｈｾ｟ＮＬＬＮＬＮ＠

cosTech°S)

ISSN 2356-542X

ﾷＱ ｣ｯｳ

ISSN 2356-542X

cosTechS'>

ISSN 2356-542X

Proceeding, Volume

ISSN 2356-542X

ACKNOWLEDGMENT

ｾＬ＠｣ｯｳｔ･｣ｨ

ﾷｩ｣ｯｳ

ﾷＱ｣ｯｳ