Abjad: Towards Interactive Learning Approach to Arabic Reading Based on Speech Recognition

Abjad: Towards Interactive Learning Approach to Arabic Reading Based on Speech Recognition

ScienceDirect ScienceDirect Procedia Computer Science 00 (2018) 000–000 Available online at www.sciencedirect.com Available online at www.sciencedir...

1MB Sizes 0 Downloads 8 Views

ScienceDirect ScienceDirect Procedia Computer Science 00 (2018) 000–000

Available online at www.sciencedirect.com

Available online at www.sciencedirect.com

www.elsevier.com/locate/procedia

ScienceDirect

www.elsevier.com/locate/procedia

Procedia Computer Science 00 (2018) 000–000

Procedia Computer Science 142 (2018) 198–205

The 4th International Conference on Arabic Computational Linguistics (ACLing 2018), November 17-19 2018, Dubai, United Arab Emirates” The 4th International Conference on Arabic Computational Linguistics (ACLing 2018), November 17-19 2018, Dubai, United Arab Emirates”

Abjad: Towards Interactive Learning Approach to Arabic Reading Abjad: Towards Interactive Learning Approach to Arabic Reading Based on Speech Recognition Speech Recognition a Norah Alsunaidia, LobnaBased Alzeera *on , Maha Alkatheiri , Alaa Habbabaha, Marwah Alattasa,

Malak Aljabria, Mona Altassana Norah Alsunaidia, Lobna Alzeera *, Maha Alkatheiria, Alaa Habbabaha, Marwah Alattasa, a a Malak Aljabri Mona Altassan College of Computer Science & Information Technology , Immam, Abdalrahman Bin Faisal University, Dammam 31433, Saudi Arabia a

a

College of Computer Science & Information Technology , Immam Abdalrahman Bin Faisal University, Dammam 31433, Saudi Arabia

Abstract

Speech recognition technology has gained more attention recently especially with the increasing use of mobile phones, and tablets. Abstract Yet, researches in speech recognition technology for Arabic language is still limited. This paper presents our experience with Arabic speech recognition using SpeechRecognizer Program Interfaceuse(API) through the design and Speech recognition technology has Android’s gained more attention recently Application especially with the increasing of mobile phones, and tablets. implementation Yet, researches of in Abjad. speech recognition technology for Arabic language is still limited. This paper presents our experience with Abjad a mobile applicationusing for learning Arabic reading which mainly targets Program children atInterface the age of six years old. It Arabicisspeech recognition Android’s SpeechRecognizer Application (API) through theinteracts design with and the users in a real-time manner by prompting them to record their voice while reading. The voice is then instantly evaluated and implementation of Abjad. vocal is provided according the accuracy of the user reading. addition, the users to assess progress in Abjadfeedback is a mobile application for learning Arabic reading which mainlyIntargets children at thecan agedooftests six years old.their It interacts with reading. goalsmanner of Abjadbyare to providethem the Arabic language learner in general, children in particular the usersThe in amain real-time prompting to record their voice while reading.and Theprimary voice isone then instantly evaluatedwith and interactive technology to shorten theirthe learning journey. to enrichthe theusers Arabic recognition researches with vocal feedback is provided according accuracy of theMore user importantly, reading. In addition, canspeech do tests to assess their progress in our experience using SpeechRecpgnizer API for the Arabic language, including thegeneral, optimization techniques we developed to increase reading. The main goals of Abjad are to provide Arabic language learner in and primary one children in particular with the accuracy rate of Arabic speechtheir recognition 44.44% up to 94.44% with adults and 80% withrecognition children. researches with interactive technology to shorten learning from journey. Moretoimportantly, to enrich the Arabic speech our experience using SpeechRecpgnizer API for Arabic language, including the optimization techniques we developed to increase the2018 accuracy rate of Arabic speech from 44.44% to up to 94.44% with adults and 80% with children. © The Authors. Published by recognition Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) © Published Elsevier B.V. © 2018 2018 The The Authors. Authors. Published by byof Elsevier B.V. Peer-review under responsibility the scientific committee of the 4th International Conference on Arabic Computational This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Linguistics. Peer-review under responsibility of the scientific committee of the 4th International Conference on Arabic Computational Linguistics. Peer-review under responsibility of the scientific committee of the 4th International Conference on Arabic Computational Keywords: Speech Recognition; Arabic; Reading; Mobile Application; Learning; interactive; Android Application. Linguistics. Keywords: Speech Recognition; Arabic; Reading; Mobile Application; Learning; interactive; Android Application.

* Corresponding author. E-mail address: [email protected], [email protected] * Corresponding author. E-mail address: [email protected], [email protected] 1877-0509 © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review the scientific committee 1877-0509 ©under 2018responsibility The Authors. of Published by Elsevier B.V.of the 4th International Conference on Arabic Computational Linguistics. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the scientific committee of the 4th International Conference on Arabic Computational Linguistics.

1877-0509 © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the scientific committee of the 4th International Conference on Arabic Computational Linguistics. 10.1016/j.procs.2018.10.476

2

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

199

1. Introduction In the era of smartphones, children are more enthusiastic in playing interactive games than traditional toys. Therefore, it is very important for the education system to keep up with the technologies rapid changes and integrate new tools to enhance the traditional teaching approaches. Since teaching the reading skill is very important for the primary student, and due to the lack of Arabic resources1, we have designed and developed Abjad to serve as a fun practicing tool for Arabic reading for children. The application in its first version targets primary one students and its content is based on their school book” ‫"(” ﻟﻐﺘﻲ‬Loghati - My language") which is the official Arabic book for the primary one students in Saudi Arabia. The main goal of Abjad is to provide children with a simple interactive environment that excites the students and encourages them to read Arabic. Several studies demonstrated that children can learn languages more efficiently when incorporating the hearing sense in the learning process [10][7]. For that, the learning methodology for Abjad is to teach the children how to read by listening to the correct pronunciation of the words. The interactive environment is implemented by giving a feedback in a real-time manner after the user records his/her voice while reading the displayed word. Although this interactive manner is not totally new, most of the existing applications suffer from low accuracy rate with Arabic language such as Rosetta Stone application that is discussed in section 2. The application’s design is simple accompanied with illustrative pictures, not overwhelmed with too many colors and images, and the font style is clear and readable for children. The main contribution of this paper is to describe the design, implementation and performance evaluation of Abjad. Moreover, to discuss the accuracy level of SpeechRecongnizer API for Arabic language, reasons behind the poor accuracy of the original implementation of the API, and the optimization techniques we implemented to raise this level by 50.5% with adults and 35.56% with children. To the best of our knowledge, no similar speech recognition application is available for practicing Arabic reading. 2. Literature Review To construct a comprehensive knowledge, we consider several aspects that fall under two main parts: speech recognition technologies where we look into the different approaches of speech recognition and the applications related to Abjad, while in the physiological part we discuss the educational environment and how to ensure the application design is appropriate for children. 2.1. Speech Recognition APIs and Applications Speech recognition technology can be implemented in two different methods. Voice to voice comparison and Voice to text comparison. The first method is to recognize an audio sample by matching it with the original audio file stored in a database. This method is mostly used in music recognition subfield [4][6], while the second method is to convert spoken words into textual format. The generated text is then used for comparison with the original text in database [5]. Figures 1 and 2 show the basic idea for both methods.

Figure 1 General steps for Voice to voice comparison.

Figure 2 General steps for Voice to text comparison.

Nadiah amal sharqy. "Importance of teaching reading for kindergarten kids",Edutrapedia, published Oct. 15,2010, http://www.edutrapedia.illaf.net/arabic/show_article.thtml?id=629

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

200

3

Voice to text comparison is implemented in many API’s (Application Programming Interface) like Google, Microsoft and Apple. The main features and limitations of those API's are summarized in Table 1. Table 1 Speech Recognition API's API name Speech library

2

Google Cloud Speech3

Speech Recognizer4

Required SDK

Supported languages

Features

Limitations

IOS 10.0+

● ●

Swift Objective-C

● Free ● Convert speech to text in real time ● Support Arabic



Any software written by the supported languages and runnable on any device (PC, template, Mobile). Android OS.

● ● ● ● ● ● ● ●

Go Java Node.gs Ruby PHP NET Python Java

● ● ●

● ●

● Android OS ● IOS ● Desktop application

● ● ●

Microsoft Azure 5



C# JavaScript Java for Android only Objective-C

● ● ● ● ● ● ● ● ●

Convert speech to text in real time Support Arabic High accuracy with 9% Word Error Rate (WER) Use deep learning neural network Inappropriate Content Filtering Noise Robustness Free. Convert speech to text in real time Support Arabic Convert speech to text in real time Has 18% WER Support Arabic



● ● ●

Not more than 1 min in every recording Needs internet connection Not free Needs internet connection

Works offline for all languages except for Arabic Not more than 10 min in each recording Not free

Android’s API is used to develop the application as it is freely available. Moreover, the recording time supported is more than 2 minutes, while apple’s API covers recording length of up to one minute. It also has wider resources, references and documentations to support the developers along their journey while developing their applications. Table 2 represents a sample of related mobile applications that have the same main functionality of Abjad and listing their benefits and drawbacks based on user experience and App stores description provided by the developers Table 2 Related Applications Supported languages 23 languages, Arabic not included

Pros

Cons

Duolingo (Learning a language)6

Targeted age 4 years and above

Providing real time feedback

  

Shazam (Music configuration)7

12 years and above

Any language, including Arabic

 

Application name

2

High detection rate for the songs Using voice-to-voice comparison

 

Redundant words Low accuracy Needs internet connection Can't recognize similar songs Needs internet connection

“Speech”, Purchase and Activation - Support - Apple Developer, last accessed Oct. 25,2017, https://developer.apple.com/documentation/speech. 3 “Cloud Speech-to-Text Documentation | Cloud Speech-to-Text API | Google Cloud”, Google, last accessed Oct. 24, 2017, https://cloud.google.com/speech/docs. 4 "SpeechRecognizer | Android Developers”, Android Developers, last accessed Oct. 24,2017, https://developer.android.com/reference/android/speech/SpeechRecognizer. 5 Zhou wang." Get Started with Microsoft Speech Recognition API Using Client Libraries ", Microsoft Docs, published Sep. 15,2017, https://docs.microsoft.com/en-us/azure/cognitive-services/speech/getstarted/getstartedclientlibraries. 6 Duolingo." Duolingo on the App Store ", Apple Music, Apple Inc, published Nov. 13, 2012, https://itunes.apple.com/us/app/duolingo/id570060128?mt=8. 7 Shazam Entertainment Ltd. “ ,”Shazam on the App StoreApple Music, Apple Inc, published July 11,2008, https://itunes.apple.com/us/app/shazam/id284993459?mt=8.

4

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000 Application name Rosetta Stone (learning a language)8

Targeted age 4 years and above

Supported languages More than 17 languages, including Arabic.

Pros  

201

Cons Providing real-time coloured feedback Breakdown reading procedure

  

Recoding starts immediately Low accuracy. Needs internet connection

One distinguished feature for Abjad compared with the applications presented in Table 2 is that it provides real time vocal feedback that changes based on the child’s attempt score. By this way, children can easily understand their mistake and try their best to overcome that mistake. Moreover, Abjad offers educator supervision account that helps in tracking the child’s performance and knowing his/her strengths and weaknesses. Furthermore, Abjad is based on “‫"( ”ﻟﻐﺗﻲ‬Loghati - My language") school book content which makes it more suitable to use in a real classroom where students can practice their lessons within a fun environment. 2.2. Psychological and Educational Aspects Abjad has included positive feedback as recommended by Schmidgall et al. [7] to motivate children while reading correctly. Due to age closeness in the study conducted by Fisher et al. [11] and Abjad end users, the Abjad’s interfaces were designed based on the authors recommendations. In particular, they are designed to be fun and simple without over decoration, so that the learning environment can still grab the child’s attention. The study conducted by Tan et al.[9] showed that there is a high acceptance from primary students to use Mobile Based Interactive Learning Environment which supports the intention of building Abjad as a mobile learning application. Furthermore, the study conducted by Baker et al. [1] gave a convincing evidence of the importance and benefits of using speech recognition in classroom to enhance the students’ reading level. Therefore, the application was implemented as an extension of this study to incorporate speech recognition within the traditional learning method. Based on the study's results of Salvador at el.[8], the lessons provided by Abjad are based on the primary one book “‫"(” ﻟﻐﺗﻲ‬Loghati - My language") which is written in Arabic, the native language for the intended children. 3. Design and Implementation Abjad has two classes of users: child and educator. Both can access the system through the application interfaces. Although each class has different functionalities, there are common functionalities shared between the two classes which are: Sign in, Reset password and Sign out. The application provides the educator with seven functions, some of them are: Signing up as a new educator, Adding a new child to the educator’s account, Viewing a child’s progress information, and Reporting a problem from any interface. The child, which is the main user, has the following services: Viewing units and selecting one of them, Practicing unlocked lessons, and Taking tests, which encompasses different exercises such as: Reading test, Matching test, True or false test, and Selecting the heard word test.

Figure 3 Unit Interface

8

Figure 4 Lesson Interface

Rosetta Stone, Ltd. “ ,”Rosetta Stone: Learn Languages on the App StoreApple Music, Apple Inc, published June 2,2011, https://itunes.apple.com/us/app/rosetta-stone/id435588892?mt=8.

202

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

5

Achieving Abjad core functionality starts with recording the child voice as he/she reads the displayed phrase. The speech recognition process is divided into three main phases and implemented using SpeechRecognizer API.  The first phase is getting the child's voice while reading the word/sentence.  The second phase is converting the voice into text format.  The third phase is the evaluation phase, where the application compares the converted string with the expected string (desired phrase) and calculates the matching percentage to provide an appropriate feedback to the child. The API converts the output and saves it in an ArrayList variable called matches. Evaluating child's reading is achieved as follows: 1. Iterate over the matches elements to check if there is an identical matching between the desired phrase and any element in the matches. If so, set the child’s score to full score which is 7. 2. If no matching with (1) and if the desired phrase is a word, a checking of word repetition will be done to see if the child has read the same word twice or more. This is implemented by splitting each array element that is generated by the API into separated words and saving them in an array called duplicates. Then, iterate over duplicates elements to compare each element with the desired word. If a repetition is detected, the child's reading is considered as 100% correct. 3. If no matching with (1) or (2), then check the special cases that we have prepared specifically for Abjad application’s content. There are 3 cases for app’s words and 5 cases for its sentences. Example of a special case with words is "‫"( "ﻟﻌﺑﺕ‬La'ebat - She played"). After several times of trials, we discovered that, the API usually converts the beginning of this word correctly, but the ending is always wrong. Figure 6 shows the elements of the matches array that are generated by the API. As indicated by Figure 6, the closest element to the desired word is found in index "1" with similarity percentage 0.8. Based on that, the chosen phrase becomes "‫"( " ﻟﻌﺑﺎﺕ‬Loa'bat – toys"). Then, the comparison will be between the desired word and the chosen phrase. 4. If no matching with (1) or (2) or (3), the similarity percentage will be calculated between all matches elements and the desires phrase using Levenshtein Distance algorithm [3]. The algorithm gives the number of different letters in two strings which are stored in globaCost variable and the similarity percentage which is stored in max_match variable. The element with the highest percentage will be saved in choosenPhrase variable. Based on the value of globalCost, max_match, and the word length (with words only), an appropriate feedback will be played. For sentences, there are 7 different cases. For example:  Case 1: check globalCost value, if it is 1, that means the child read correctly because based on our trials, usually, the API adds an additional letter to the sentences. Figure 5 shows an example of this case where the desired sentence is "‫"( ﺃﺣﻼﻡ ﺗﺮﺗﺐ ﺍﻟﻜﺘﺐ‬Ahlam toratteb alkotob – Ahlam arranges the books"). The heard feedback will be “‫ ﺃﺣﺴﻨﺖ‬،‫"(“ ﺭﺍﺋﻊ ﺭﺍﺋﻊ‬Raaea' raaea' ahsant – Great great, well done").  Case 2: if max_match >= 0.89, the child score is 6 out of 7, and the heard feedback is " ‫ﻗﺮﺍءﺗﻚ ﺟﻴﺪﺓ ﻭﻟﻜﻦ ﻟﻴﺴﺖ‬ ‫"("ﺻﺤﻴﺤﺔ ﺗﻤﺎﻣﺎ‬Qeraa'toka jayedah wa laken laisat saheehah tamaman – Your reading is good ,but not totally correct").

Figure 5 Example of adding addtional letter to the converted sentence by the API.

Figure 6 API output of the word "‫" ﻟﻌﺒﺖ‬

6

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

203

4. Results and Discussion We tested the application with 25 persons (3 adults and 22 children). The application was tested on adults firsthand to test the application in ideal conditions, as adults’ sounds are clearer than children and usually they read fairly well. Also, to avoid the children behaviors such as playing while recording, or repeating words or sentences, etc. .This was very important because at this first stage, the aim of the test was to check the pure functionality of the API before testing it on the targeted children.

Figure 7 Example of the API translating error

The accuracy obtained with original Android SpeechRecongnizer API was 44.44%, which is very low. After investigating the reasons behind this low accuracy rate, we were able to improve the API performance. The API optimization techniques we implemented solved the three main problems affecting the API accuracy, which are: 1. When converting speech to text, the API always writes “ ‫( “ ﺓ‬taa' marbootah – a different form of the Arabic letter " ‫ "ﺕ‬,that comes at the end of the word ) as “ ‫"(“ ﻩ‬haa' - a different form of the Arabic letter " ‫ "ﻫـ‬,that comes at the end of the word ") and “ ‫"( “ ﺃ‬alef – an Arabic letter") as “ ‫"(“ ﺍ‬hamzat wasel – a special case of the Arabic letter ‫ )" ﺃ‬which results in wrong evaluation when comparing text to text. So, the API was not able to figure out these two Arabic letters correctly. The solution we added is a temporary change for the word/sentence that the user read to a form that is accepted by the API, so the comparison succeeds. The change is implemented only in the backend and does not affect the displayed word/sentence. Figure 7 shows an example of this problem, the displayed word is “ ‫"(” ﺇﺷﺎﺭﺓ‬Isharah - Signal"), but the API translates it as "

‫" ﺍﺷﺎﺭﻩ‬.

Integrating a spell checker in the processing steps along with the API would be an efficient solution to fix misspelled words detected by the API, and to reduce the amount of special cases. An example on that is correcting the word translated by the API "‫ "ﻁﺎﻭﻟﻪ‬to be "‫"(" " ﻁﺎﻭﻟﺔ‬Tawelah – Table") easily . However, the spell checker may not be effective enough as in some cases the word detected by the API is not translated accurately even though it is spelled correctly. For example the word ‫ "( ﻟﻌ َﺒﺖ‬La'ebat – She played")is translated by the API as ‫"( ﻟﻌﺒﺎﺕ‬Loa'bat – toys "), in this case even if we count on a spell checker, the word might be corrected to ‫ "( ﻻﻋﺒﺎﺕ‬La'eebaat – Female players) which is totally a different word than "‫ "ﻟﻌﺒﺖ‬and results in providing the wrong feedback because the two words are not matched. Moreover, since we noticed that the API has problems only with two letters ( ‫ﺃ‬: Alef & ‫ﺓ‬: taa' marbootah ) and few words, it was better to handle these two cases by using a simple code rather than importing an additional API for spell checking, which might slow down the evaluating process. 2. Giving an error to the user if she/he finishes reading but doesn’t release the microphone button. This was solved by controlling the execution of the error function that is called by the API automatically. Once the user finishes her/his speech, a Boolean flag value will be set as TRUE, so every time the error function is called, the flag value will be checked. If it is true, the function will return with no error executed.   3. Stopping of speech recognition service when the user clicks on the microphone button multiple times sequentially. This was solved by cancelling the current state of speech recognizer object and starting the listening process again.   After our optimization, the accuracy raised to 94.5% with adults and 80% with children. The reasons behind this variation between adults and children accuracy could be:  Children face difficulty with long sentences such that they stop several times while reading or skip some words in the sentence which affect the evaluation result. 

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

204

 

7

Sometimes children press the microphone button after they start reading or release it before they finish reading. Since Abjad application is for primary one children who are six years old only, some children do not pronounce some Arabic letters correctly. For example, " ‫"(" َﺟ َﻣﻝ‬Jamal - Camel") pronounced as " ‫" ﺯﻣﻝ‬ ("Zamal – A word with no meaning in Arabic).

Interestingly however, we have faced some other issues with API which cannot be solved and still affecting the accuracy level. We summarize those issues as follows:    The API interrupts the speech to text conversion process by calling its Error function for reasons that are out of developer control such as "bind to recognition service failed" and "StartListening method is failed"9.  The API sometimes fails to recognize a match for the user speech even though there is an actual match to what the user has said without clear reasons.  The API has a feature of working on the offline mode. However, this feature is not available for the Arabic language.  The API does not allow the programmer to have access to the voice data that comes from the user. Due to this limitation, we were not able to implement “child hears his/her voice after reading” function.  The API does not permit access to the length of silence which is the default time that the API remains waiting after the user stops speaking. By default, the API converts the user input into text after a very short time of silence. We have tried to use certain constants to control the length of silence, but these constants values have no effect as demonstrated in the SpeechRecognizer implementation guide 10. The consequence of this issue is that, the children usually speak slowly and may stop several times when they read a long sentence which results in wrong feedback. 5. Conclusion Abjad is interactive mobile application targets primary one students. It was designed and implemented to assist the children to practice Arabic reading. It uses speech recognition technology to evaluate the user reading, and to provide an instant feedback accordingly. In addition, the application offers tests to assess the child progress. Android Speech recognizer API used to develop Abjad. With the original implementation of the API the accuracy was very low, then after implementing new optimizations techniques, the accuracy level improved by 50.5% for adults and 35.56% for children. We believe that our experience with Abjad greatly added to Arabic language technologies and assist applications developers to benefit from the results we achieved, the optimizations techniques we added, and the issues of the SpeechRecognizer API we demonstrated. Because of the promising results we have achieved, efforts shall be made next to enhance abjad usability and functionality. For example, using a larger dataset to study the performance based on the size of data , highlighting the words that were not read or read mistakenly, enhancing the reading level for children by using Phrase Drill error correction mechanism that showed its effectiveness [2], considering repeating some of the sentence’s words in a single attempt as correct reading since this behavior can appear when children practice reading, and customizing the application vocal feedback based on the child's gender.

9

“SpeechRecognizer.java”, Google Git, last accessed Sep.2,2018, https://android.googlesource.com/platform/frameworks/base/+/master/core/java/android/speech/SpeechRecognizer.java. 10 “RecognizerIntent | Android Developers”, Android Developers, last accessed Apr. 24,2018, https://developer.android.com/reference/android/speech/RecognizerIntent.

8

Norah Alsunaidi et al. / Procedia Computer Science 142 (2018) 198–205 Norah Alsunaidi et al. / Procedia Computer Science 00 (2018) 000–000

205

References [1] Baker, Elizabeth Betsy A. (2016) " Apps, iPads, and Literacy: Examining the Feasibility of Speech Recognition in a First-Grade Classroom." Reading Research Quarterly 52 (3): 291–310. [2] C.Begeny, John, Edward J. Daly III ,and Rachel J. Valleley. (2006) " Improving Oral Reading Fluency Through Response Opportunities: A Comparison of Phrase Drill Error Correction with Repeated Readings." Journal of Behavioral Education 15(4): 229–235. [3] Hirschberg, D. S. (1975) “A Linear Space Algorithm for Computing Maximal Common Subsequences.” Communications of the ACM 18(6): 341–343. [4] Kim, Kyu-hong, Jeong-su Kim, and Ick-sang Han. (2008) " Method and apparatus for searching for music based on speech recognition". US 2008/0249770 A1, filed Aug. 20, 2007. [5] K.saksamudre, Suman, P.P. Shrishrimal ,and R.R. Deshmukh.(2015) " A Review on Different Approaches for Speech Recognition System." International Journal of Computer Applications 115(22): 23-28. [6] Li-Chun Wang, Avery, and Julius O. Smith, III. (2006) " System And Methods For Recognizing Sound And Music Signals In High Noise And Distortion. " US 6,990,453 B2, filed Apr. 20, 2001. [7] Schmidgall, Melissa, and Laurice M. Joseph. (2007) " Comparison of phonic analysis and whole word‐reading on first graders' cumulative words read and cumulative reading rate: An extension in examining instructional effectiveness and efficiency ." Psychology in the Schools 44(4): 319-332. [8] S. Salvador, Ana Maria Margarita. (2017) “To Read or Not To Read: Listening to Children Talk about Reading Motivation.” Acta Universitatis Lodziensis. Folia Librorum 1(24): 47-65. [9] Tan, Tan-Hsu, and Tsung-Yu Liu. (2004) " The MObile-Based Interactive Learning Environment (MOBILE) and A Case Study for Assisting Elementary School English Learning." IEEE International Conference on Advanced Learning Technologies, 2004. Proceedings. [10] Timothy Sowden, Paul ,and Jim Stevenson.(1994) "Beginning reading strategies in children experiencing contrasting teaching methods." Reading and Writing: An Interdisciplinary Journal 6(2): 109-123. [11] V. Fisher ,Anna, Karrie E. Godwin ,and Howard Seltman. (2014) " Visual Environment, Attention Allocation, and Learning in Young Children: When Too Much of a Good Thing May Be Bad." Psychological Science 25(7): 1362-1370.