10:
90
V
OICE SEARCH
FOR QURANIC
VERSES
RECITATION
AmmarMohamm ed MaGIC-X UTM-IRDA Universiti TeknologlMalaysia
JohorBahru,Malaysia ammar@magicx.my
Mohd Shahrizal
Bin SURar
MaGIC-X UTM -IRDA UmversitiTeknologiMalays iaJohor Bahru,Malaysia shahrizal@utm.my
Md.Sah Hj.Salam . Faculty of Computing UniversitiTeknologi Malaysia
JohorBahru, Malaysia sah@ulrn.com
,Abs" act•This pap er describ es cha llengesandsolutionsfor buildingII.suecessful voice sel rc hsystem as applied to Quranlc vers es. The paper descnbestbe techniques usedto deal with an finite\'ocabularJ how modelling completely in the voice domain (or langu age mod el and dictionary can avoid some system cornplealry,andhowwebuiltdiction aries,language andacoustic modelsin thefram ework, The Holy Qura n Iswri tt en In Arabi c language, and the Ara bicisoneofthe oldest languages Inthe world thatpresents itsown features andchallenges while searching for Arable-based content. The most search systems for Ihe Holy Qura nIs organized around text words (contained In thetarget verses)butnosystemorganized aroundvoice"hile there is need tosear chIn quranlered tation .Thespeech recognitionapproaches areapplled to buildthe dictionary and test thesystem, "'hilethe MFCCandstemming techniqueswill be willbeappliedto find the stem of theword.Thedevelopmentofvetcesearch for Quranlc redtationledtoa significa ntsimplification of the originalprocess to hulld a system to retrievingaudio information related to the Qur'an , it also helpsto buildasystemtoserve peoplewith spedal needs such as blind.paralyzed,and othe" who can not usethe keybeerd.
X~words- vetcesear ch; Quranicrecitation ;Quranic aClllstlc;
Spe« h recognition
I. lNTRODUCIlON
voice search the ability to perfonn web searches simply by speaking, first appearedaround 2008 on iPhoneand Android phone in US English. Soon after a several of researchers focusing on developing voice search systems for other languages.[ I]
Arabic language is not similar to US English, it is a semantic language with a composite morphology. Arabic words are categorizedasparticles,nouns, orverbs.Unlike most western languages. Arabic script writingorientatio n is from right to left. Thereare28characters in Arabic. The charactersareconnected and do not start withcapital letter as in English. Most of me characters differ in shape based intheirposition inme sentence and adjunctleners.[2,3)
The Quran is the holy book of Islam ,origi nally written in ClassicalArabiclanguage,consists of 6236 versesdividedinto 114 chapters called suras. Each surah also differs from one another interms of thenumber of verses(ayat ) [3, 4}.
Orthographicvariationsand the use ofdiacritics andglyphsin the representat ion of the lang..uage of Classical Arabic increase thedifficulty of stemming[5].Many versesare similarand even identical.
Searchingforsimilar words (e.gverses)could return thousands ofverses.that whendisplayedcompletelyor partly as list would make analysis and understanding difficult and confusing. Moreover it would be visually impossibleto instantlyfigureout the overalldistribution of the Identified or retrievedversesin the Quran.(4J
II. LITERATUREREVIEW
Using our voiceto accessinforma tionhas beena part of science fiction ever.Today, with powerful smartphones,cloud-based compu ting,and speech recognitiontechniquesscience fiction is becomingreali ty.
One of the important voice applications is is a speech recognition technology that allowsusersto search by saying terms aloud rather than typing them into a search field. The information normally existsin a largedatabase, and the query has to be compared with a fieldin the database to obtainthe relevant informationle,7].
The proliferation ofsmartphonesandother small,Web-e nabled mobile deviceshas spurred interestin voicesearch.
The voice app lications are available in the markets whic h recitesthe holy Quran.One of themost popular andcommonly used isQuran Auto Reciter(QAR) [8],andsomeofapplications depend on speec h recogniton techni ques like Hafas[9] and Ehafiz[ I 0,II].
This studypropo ses the systemthe abilityto search inrecitation by voice, And the expected benefit of Quranic voicesearch system is using in retrieving audio info rmatio nrelate d to the HolyQuran,as wellas a systemsto servepeo ple with special needswhocannot use thekeyboard.
ARELATED SnIDIES
In2008.anew versionofGOOG-41 1 has been deployedwhich allowe d(and encouraged)the user tostate theirneedin a single utteranc e rather than in sequentialutterances thatsplitapart the location andthe business.This wasmotivated by our desireto
accommodate faster interactions as ...ell as allow the user greater flexibilityin how they describe their needs(6].
With regard to
Quranic
search systems. most of researchershave interested on the developmentof search techniques for the
Quranic text, but the principle can be applied in the
developmentof voice search systems.forexample.
Kadri et at (3] proposed a new stemming method that triesto
determinethe core of a word according to inguistic rules. The new method shows the best retrieval effectiveness. The Iinguistic-based method can better determinethe semantic core ofa word.
Naglaa Thabet [5J proposed a new light stemming approach thai gives better results.when applied to a rich vocalized text as the
Quran. The stemmer is basically a light stemmer to remove
prefixes and suffixes and is appliedtoa version of the
Quran
transliterated into western script.Riyad Alshalabi [3] provided a technique for extracting the
trilateral Arabic mol for an unvccahzed Arabic corpus. It
provides an efficient way to remove suffixes and prefixes from the inflected words. Then it matches the resulting word with the available patterns to find the suitable one and men extracts the three letters of the root by removing all infixes in that pattern.
Regard ing researches in the developmentofspeech recognition techniques to use in Quran recitation applications, there are several resear chinthe fieldofQw'anic voicerecognition. of the
most famous of such research is an automated delimiter
introduced by Hassan Tabbalet al.[S] Which extracts verses from an audiofile andcovertsQuran verses inaudio file,using
speech recognitio n tec hnique. The Sphinx. IV framewo rk is usedtodevelop this system.
The core recogniti on process is provided automatically bythe sphinx engine using the appropriate language and acoustic
models. The sphinx framework mustbe configure d using an
xrnl based configuration tile.
The recognitio n ratio in thecase ofTarteelis slightly betterthan
inthecase ofTAJWE ED . One possiblereason for thiscould be that themajorityof the Tarteelrecitat ionsavailable now follows the same monotony and theduration (in time) ofeach phoneme
differs slightly from one reciterto another.There is also the extranoisethat is causedbythecompressi onof theaudio files and the low quality oftherecordi ngs.
Although we had anticipat ed this by using noisy aud io files during the tra ining. but the differen ces incompressionratios between the files add alot ofven ation of the added noiseand thus causing extra errors. When unskilled persons tested the system (we even tested it on children),it behaved astonishi ngly
welleven ...ben the reciter was aw"Oma n.IIcase thatcannot be
encounteredinreal life becauseit'snot usualtohavea woman
reciting theHoly Quean.There is alsoaninterestin g observation dra...l1from these tests: Itis elways reco mmend ed [12]totrain the systemwithmorethan 500 different voicesin ordertoreach
speaker independence. But we didn' ttrain our system withthis
relatively large number and still we were able to have
remarkablespeaker indepe ndenceresults.
The system introduced by Hassan Tabbal[8J an automated
delimiter, which extracts verses from the audio tiles, using
MFCe
feature extraction. This systemis useful for peoplewho are well versed with Tajweedrules. However.USCTS'whoare not Arabiespeakers don't benefit from thissystem. In addition,it may also not help reciters to improve recitation abilities. The system is useful 10 those people who already know the correctrecitation of the holy Qw'an and the subsequent rolesand not suitable for none Arabic speakers.A system that will be able 10 help users to know recitation rules (TAJWEED), pointing out mistakes made during recitation is a necessity and a task achieved by the E-Hafiz system.
The other system introduced by Bushra Abro [13] Arabic
language which included isolated words and sentences. The dataset comprised of few Arabic sentences and words. The
system was buill on Al-Alaoui algorithm lhat trains Neural
Networks (NN) and has been able to achieve8~;'accuracy on
sentence recognition.But the drawback of this technique is that the system was trained on distinct sentences.
Similaritiesin sentences separateNNs and therefore are needed tobetrained whichis computationally very expensive.for the
purpose of Qur'an memori zat ion,
MFee
was used to extractfeatures. The dataset consists of few small Quranic verses and pattern matchingwas doneby Vector Quantization.The system gained good recognitionrate but the approach is statisticaland therefore difficult to scaleup for larger systemto be built on completeQur'an.Itis also knownto takemuchtime andspace complexity whi ch is a sbortfa llof a real time system.
Zaidi Razzak [14] presents different recognition techniques
used for the recitation ofthe
Quran
verses in Arabic verse pointingoutthe adva ntages andthe drawba cks. Themost useful method for theproject"Quranic verseRecitation Recognitionfor suppo rt j-QAF leaming" is explained therein.J·QAFis a
pilot program. thataims toencouragelearnersto learn Qurao read ingskills,understandingTejweedand Islamicobligations.
The method ofteachin g j-QAF (teacher and student) is still
handled manuall y.Onebasic goalof this paperisto automate
the learning process.
B.HIDDEN MARKOVMODELS
Markov cha in is a series of cases and final set of random
functionsin times of intermittentand eontinuous[ 15]. Where
assume some conditions. and notices that are generated
randomlyas initialvalues for the model. And is moving from case to case againby the probability matrix.Cansee the output
of the random function and its relationshipwith the case, but can't figure out stages of the transitionfrom oneslatetoanother.
nor knowledge of the sequence of observations and time. so called hidden Markov models. Hidden Markov models are
classified on the basis of time into two classes: Notes in
emerge out, like fromthe roots of aplant ; thestem and many branchesemerge.
Forexample.the root",
J
e.~ gives thestemwords:",.M;-"(ilm) and...(alam). Furth ennore, fromthe stem word:-
,Joo.-(ilm) comes the branching words :"~. (ya'tamu),-+.
(a1eem),"~-(salim)etc; andfrom-,Ji:."(alam} comes the branching words-~"(aalameen}etc.In this stage,also remove of noise fromdigitalaudio signals in orderto extractthedistinct ive features of the voice clear ly. Also the time and frequency must be consolidated, thai can be
applied inseveralways such as (MA Analysis )Inorder tobe a specific time and frequency( EX. lOs, 22.0S0hz) [10]. This study willfocu s ontherootsofQuranicwords(representing its keywordsindatabas efile).
Mel-Freq uencyCeps tral Coeffi cients (MFCC) are extensively inASR.MFCC isbased onsignal decompositionwith thehelp Several differentfea ture extraction algorithmsexist, namely Linear Predictive Cepstral Coefficients (LPCC): computes Spectral envelop before conven ingit intoCepstraI coefficient.
PerceptualLinearPrediction (PLP)Cepstra: it isbasedonthe Nonlinear Bark scale. The PLP is designed for speech recognition withremovingof speakerdependent characteristics.
III. METHODOLOGY
B.PREPROCESSING&FEATURE EXTRACTION
This stage, to remove of noise from digital audio signals in order toextractthe distinctive featuresof thevoiceclearly. Also the time and frequency must beconsolidated, that can be
applied inseveral ways suchas (MA Anal ysis)Inorder tobea specific time and frequency( EX.lOs,22.050hz ).
Alsotheprocess offindingthe roots willbe doneatthis stage, Stemming is a pre-processing stepinany InfonnationRetri eval
B
·.
~-
[:ata
(IR)system.Itistheprocessofremovingallaffixes (prefi xes,B:I::i'
M~~
.&Fearu: 11 ' suffIxes andaffixes) froma wordto extract itsroot.Extra<:Uoa eo ecun Stemmi ng has been widely used in several fields of natural L ---l g language processingsuchas data mining,informationretrieval, Firarc1 Rcwan:bItalts and multi variate analys is. h is a method for iIllJroving the
perfonnaoce of information retrieval systems.
Thetwo approachesof stemming arc light stemming andheavy stemming,Thefanneris-theprocessof strippingoffprefixes andsuffixestoprodu ce stemofthcword";and heavy stemmi ng is ..the process ofstripingoffprefixes, suffixes and affixes to producetherootof theword".
Aswell in thisstage the audiosigna lisprocesseduntil we get a clearparameters which enables us torecognize the Speech to analyzeitschara cteristics beforerecognitionprocess.
Features extraction using MFCCmethodconsistsofsevensteps [14].Pre-emphasizeis thefirststep in the process followedby framing,windowing,filtering,transfonningusingFastFourier Transform, lo g calculationsand invertingthe Discrete Fourier Transfonn.
Thegoal offeatureextraction istofindaset of propertiesof an utterancethat haveacous tic correlationsin the speec hsignali.e. parametersthat can somewhatbe estimatedthrough proce ssin g of the signal wavefonn.Suchparameters aretermedasfeatures . There are four major stages inthis resea rch .The phasesare
sequentia lasdescribedinFigure 1.The stagesare asfollo ws; C.STEMMING TECHNIQUES
Stemmingis a pre-processingstep in most ofsearchsystem.It is the process ofremovingall affixes (prefixes,suffixes and affixes) fromaword to extract itsroot.
Stemming has been widely used in several fields of natural language processingsuch asdata mining.informationretrieval,
and multivariate analysis. It is a method for improving the
performanceofinformation retrieval systems.
Thetwoapproac hesofstemm ing arc light stemm ingandheavy stemm ing.Theformer is ..the proce ss of stripping offprefixes andsuffixestoproduce stemoftheword";and heavy stemming is "theprocessof striping offprefixes, suffixes and affixesto producetherootof the word"[3].
A.DATACOLUCTlON:
This stage,Acoustic datais necessaryto buildinitial acoustic modelsandtestsets to measureperformance .
TheHolyQurancontain77439wordsand theroots of quranic wordsis1739roots[16}.
Root letters areletters that form themain part of a word.Most of theArabic wordsare constructedfrom rootletters .
Theserootlettersprovide the basiclexicalmeaningof theword. Theyare joined alo ngwithsome otherletters to fonn
different words which have related meanings.Justas theroot determinesthetype of planI,thegenesdeterminethe
characterist ics of a person; similarly roots determine the meanin g of'theword.
Examp le: theroot - -fAj-joins together to fonn the original word"H"'1(Rah im): womb.Mercy is derived from thewomb. If weunderstand themercyofamother,then wecanunderstand the concept ofmercythat is related tothisroot, Thus,theroot givesthe characteristicofmercy.
Most Arabicwords havethreeletter rootsandrarelyfour or five letter roots. and they always maintain their order specific arrangement,Example:
-
rk-"
-
knowledgeand",J
e.
..
"action ( changing the order gives a completely different meaning!).Thisrootscome Fromone setofroot letters, come multiple stem words .. ~- and from them many wordsAcousti c modeling
..
""
-+-...
Mlddl
FIn.
e
I
j
l
•
"
Inttl
.
1
l
j
t
•
Jl
l
.\1;
<11]
.Uo
.
..
...
.
"
."
Th.'
l. 1
H'a'
sa.
Rai
Tta'
0.1 B.'
T.'
Dha'
Kha' Th.'
Chad
Alif ...&1\ ,
Seen
Jeem
Sheen
TableI.The variouster ms
or
Ara bicletterCharact
NamIsolate
er e
d
1,/a:1,Iu:1)and thecorrespondingsemivowe1s (/ylandIwl),if applicable. A letter can have two to four different shapes: Isolated,beginningofa (sub)word,middle ofa (sub)word and endofa (sub)word. Letters aremostly connecte dand there is no capitalization.The letter is represented as below at table,in their variousforms.
Windowing: Tn order to minimize andeliminate discontinuity
fromthestart andendof each frameof the signal, Hammi ng window process is applied.It is the most commonly used in
MFCC to minimize thediscontinuities of the signalby tapering
the beginningandendof eachframe to zero.
Mel Filter-bank:Lowfrequency componentinspeechcontains
useful informa tionas comparedto highfrequency.Itrepresents
the relationship between the frequency in Hz and Mel scale frequency. In order to perform Mel-scaling, a number of
triangular filter-bankisused andtherefore,abankof triangular filters is createdduring MFCC calculation.collectingenergy
fromeachfrequencyband[1 7].
MFFC is themostwidely usedtechniquesin the Arabspeech recognition because it is effective in noisy, vocal tract, and provide higher result oflow bandwidth.The MFCC Processes are; frame blocking, windowing,OFT,MelScale Filter,Inverse
Discrete Fourier Transform (lD FT) block. In the frame, blockingthe speechwaveformis cropped to removesilenceor acoustical interference that may bepresent at thebeginning or end of thesound file. As anoutco meof this processFourier transformationprocess is enabled[17].
of a filterbank,which uses theMel scale.The MFCC results
on Discrete Cosine Transform (OCT)of a real logarithmofa
short-timeenergy expressed in the Mel frequency scale.
DiscreteFourier Transform (OFf):Due to the speech signal form isaset ofN discretenumbe r of samples(windo wed signal VI [k] .. .VI [mn, eachframe sampleis convertedfromtime
domainintothefrequency domain (acomplexnumber Y2 [kJ).
OFT is normally computed through Fast Fourier
Transformation(FFT) algorithm.
..
...
.J.+
...
J
..I
..
J
J
e
...
••
."
v.
'
H.'
Ka
'
Lam
F.'
A'in
Q.,
w.w
NoonGhaln
Meem
O.DICTIONARYBUILD ING
Afterfeature extractio n process executed,tobuildthe Quranic
recitation dictionary, therecognition process willcomparethe
C.LAN GUAGEMODEU tNG
Inverse DiscreteFourierTransformation (IDFT): The finalstep
of MFCC feature extract ion isto takeinverse ofOFT.As an output of thisstep we get features ofspeechinvectorformat called feature vectors. The feature vector is obtained . This
feature vectorisused asinput ofthenext phase.MFCC feature
isconsidered forspeaker-independentspeechrecognition and
for the speakerrecognitiontasksas well.
Asit is known thatthe Quran is writtenintheoriginal classical
Arabic. And Arabic is one ofthe languages that are often
described as morphologically complex and the problem of language modeling for Quranicrecitationaremultipart by the methods and speed of recitation.It is also there are many
difficulties begin when dealing with the specialties of the
Arabic language in AI-Qura n,due tothedifferencesbetween written and recite AI-Quran (8, 18, 19].The Quranic Arabic
alphabetsconsistof28 letters, knownas hijaiyahletters (from alif( l)...until ya(<i)) (I2,19,20). Those letters includes 25
extracted features with its reference model. This reference model isdeveloped, after the enrolment ortraining phase had been successfully implemented. In this case, the reference model (stored model in database) used consist of 2 types of
models, which are Word based Model and Phoneme based
Model.
The reference model for phoneme[I2 , 21) based model is
totally differs from word basedmodel, where speech features that have been extracted are directly compared to the word templates.
The Quranic recitation has 60[22J phoneme that led more difficulties to the dictionary building. Here, each of word templates in direct matching model will store as a vector of featuresparameters. Wordbased modelwill be used as a first
model, while phoneme based model been the second model
used as templatematching at testing/recogni tionpan.
E.ACOUSTIC MODELING&TEST
The acoustic modelling will be doneby HMMs,Nevertheless
there are three methods for: Hidden Markov Models HMM,
Vector Quantization (VQ), Artificial Neural Network
(ANN )[12Jand Hybird Model[23].
HMM or VQ can be apply for training and testing. HMM is
usedwhen Arabic language recognitionhas to perform and VQ for English language[24] HMM had introduced the Viterbi
algorithm for decoding HMMs, and the Bawn-Welch or
Forward-Backward algorithm for training HMMs. All the
algorithmofHMM playacrucial roleinASR.Itinvolved with
states, transitions, and observations map into the speech recognition task.
The extensions tothe Beum-Welchalgorithms needed to deal
with spoken language. These methods had been implemented
by D. Jurafsky and 1. H. Martin (2007) in theirresearch. Here, speech recognition systemstraineach phone HMM embedded in an entire sentence. Hence, the segmentation and phone alignmenl are performed automatically as parts of the training procedure{l 4]. It consists of two interrelated stochastic processes commonto describe the statisticalcharacteristicsof the signal. One of which is hidden (unobserved) finite-state Markov chain,and the other is the observationvectorassociated
with each state of the Markov chain stochastic process
(observable)[l7).
ArtificialNeural Network (ANN)is a computational model or mathematicalmodelbased onbiological neural networks . The procedure depends on the way a personappliesintelligence in visualizing, analyzingand characterizedthe speechbased on a set of measured acoustic features[14], but the basic neural networksare not well equipped to address these problems as
compared to HMM's.
Vector Quantization (VQ) Quantization is the process of
approximati ng continuous amplitude signals by discrete
symbols. It can be quantized on a single signal value or
128
parameter known as scalar quantization,vector quantization or others. VQ is divided into 2 parts,known as features training and matching features.Features training is mainlyconcerned with randomlyselecting feature vectors and perform training for the codebook using vector quantization(VQ) algorithm[14].
IV. EXPECTED RESULT
The results canbesummarized as expected:
t. Create a new system thatableto search inHolyQuran by vioce.
2. Help the special needs people in the Quran search and access to the selectedSura easly.
3. Assist in the development of audio informationretrieval systems regardingthe Quran .
V. CONCLUSION
This paper has covered many aspects of speech recognition system and Quranic recognition,and focusing on The voice searchinHoly Quran. And this research funding will behighly beneficialwith specialneeds and whocannot use the keyboard, as it will be useful in informationretrieval researches in the Holy Quran ..
By the endof this study it is expectedthat the existing models willbe enhancedand improve dfor more suitablefor specificity theHoly Quran.
ACKNOWLEDGMENT
This research is supported by UTM Magicx at Departmentof Computer Graphics and Multimedia, Faculty of Computing, UniversitiTeknolog iMalaysia.
REFERENCES
I Schuster, M., and Nakajima, K.: 'Japanese and korean voice search', in Editor (Ed.)"(Eds.): 'Book Japanese and korean voice search'(IEEE, 2012,edn.), pp. 5149·5152 2 Al-Shammari,E.T.: 'A Novel Algorithmfor Normalizing Noisy ArabicText', in Editor (Ed.)"(Eds.): 'Book A Novel
Algorithm for Normalizing Noisy ArabicText' (IEEE,2009,
edn.), pp.477-482
3 Al-Taani, A.T., and Al-Gharaibeh, A.M.: 'Searching
ConceptsandKeywords inthe HolyQuran', 2010
5 Thabet,N.: 'Stemming the Qur'an' ,in Editor(Ed.)"(Eds.) : 'Book Stemming theQur'an' (Association for Computational Linguistics,2004,eOO.),pp. 85-88
6 Schalkwyk,
J
.,
Beefennan, D., Beaufays, F., Byrne, R,Che1ba, C., Cohen, M.•Kamvar, M., and Strope, 8.: "Your Word is my Command": Google Search by Voice: A Case Study' : 'Advances in Speech Recognition' (Springer, 2010), pp.6 1-90
7 Wang, Y.-Y., Yu, D., Ju, Y.-C., and Acero, A.: 'An
introduction to voice search', Signal Processing Magazine, IEEE,2008,25, (3), pp. 28-38
8 Hassan,T.,Wassim,A.-F.,andBassem,M.:'Analysisand Implementation of an Automated Delimiter of' Quranic" Versesin AudioFilesusingSpeechRecognitionTechniques' , Robust SpeechRecognitionand Understanding,2007,pp. 35 1
9
10 Elhadj, Y.a.M.: 'E-Halagat: An Eel.earningSystem for
Teaching the Holy Quran' , Turkish Online Journal of
EducationalTechnology, 2010,9, (1)
II Muhammad,A.,u1Qayyum,Z., Tanveer,W.M.M.s.,and Syed, A.Z.: 'EiHafiz: Intelligent System to HelpMuslimsin
Recitationand MemorizationofQuran' ,LifeScienceJournal, 2012,9,(I),pp.534-541
12 Ibrahim. N.J., Idris, M.Y.I., Razak, Z., and Rahman, N.N.A.:'Automated tajweedchecking rules enginefor Quranic learning' , Multicultural Education & Technology Journal, 2013,7, (4),PP.275-287
13 Abrc, B., Naqvi, A.B., and Hussain, A.: 'Qur'an recognition for the purpose of mernorisarion using Speech Recognition technique',in Editor (Ed.)"{Eds.): 'Book QUI'an recognition for the purpose of memorisation using Speech Recognition technique'(IEEE,2012,edn.),pp.30-34
14 Zaidi Razakf-, N.JJ., Mohd Yamani Idna Idrist , Emran Mohd Tamilt, Mohd Yakub@Zulkifli Mohd Yusofftt and Noor Naemah Abdul Rabmaott: 'Quranic Verse Recitation
Recognition Module for Support in j-QAF Learning: A
Review', UCSNS International Journalof ComputerScience and Network Security, VOL.8No.8,August2008, 2008 15 Nilsson, M.,and Ejoarsson,M.: 'Speech Recognitionusing Hidden MarkovModel' ,Department of Telecommunications and Speech Processing, Blekinge Institute ofTechnology,2002 16 20 14
17 Meng.J.,Zhang.
1.
,
and Zhao,H.:'Overview ofthe Speech Recognition Technology ' , in Editor (Ed.)"(Eds.): 'Book OverviewoftheSpeechRecognition Technology'(IEEE,2012, edn.),pp.199-20218 Truong. K.: 'Automatic pronunciationerror detection in Dutch as a second language:an acoustic-phonetic approach' ,
2004
19 Ahsiah, I., Noor, N.. and Idris, M.: "Tajweed checking system 10 support recitation' , in Editor (Ed.Y(Eds.): 'Book
Tajweed checkingsystem to support recitation' (IEEE, 2013, edn.),pp.1 89-193
20 Ibrahim. N.J.: 'Automated TAJWEED checking rules
engineforQuranicverse recitation',2010
21 Ahmad,M.A.,and EI Awady,R.M.: 'Phonetic recognition of Arabic alpbabetletters usingneuralnetworks ' ,International
Journal of Electric& ComputerSciences IJECS.IJENS,2011, 11,(01)
22 Yahya, A., Aighamdi, M., and Alansari, M.A.A.: 'Auto Correction oftheHolyQuran Recitation',Journal ofComputer Research .II:01-15.(inArabic), 2012
23 Zarrouk, E., Ayed, Y.B., and Gargouri, F.: 'Hybrid continuous speech recognition systems by HMM, MLP and SVM:a comparative study ',International Journalof Speech Technology,2014,pp.I-I I
24 Muhammad,W.M.,Muhammad,R.,Muhammad ,A., and