A) FIELD OF THE INVENTION
[0002] The present invention, in general, relates to transliteration. More particularly, the present invention relates to a system and a method for transliteration based on phonetic mapping. Further, the present invention provides systems and methods for inputting language into a computing device based on a phonetic-based scheme.

B) BACKGROUND OF THE INVENTION
[0003] In a general sense, transliteration is defined as a process of translation of text of one literature to other based on pronunciation. Transliteration is an im-portant process in many multilingual natural language tasks. An essential component of transliteration approaches is a verification mechanism that evaluates if the two words of different languages are phonetically accurate translations of each other. Alt-hough many systems have transliteration generation (recognition) as a component, stand-alone verification is relatively new. Most of the existing transliteration meth-ods follow word to word mapping, for example, latin words were directly mapped to words in a native language. Transliteration using word to word mapping does not provide accurate results and also deceit the users of a good typing experience. Fur-ther, verification has been used as an essential step for transliteration and the existing prior-art fail to provide an effective validation process which are time ef-ficient and accurate. Also, in many cases, there is no agreed upon standard romaniza-tion system, leading to an increase in ambiguity and noise when decoding to the tar-get words in the native script.
[0004] Hence, there is a need for a system and a method that yields substan-tial accuracy improvements and latency reductions over the existing transliteration methods.

C) OBJECT OF THE INVENTION
[0005] A primary object of the present invention is to develop a system and a method for transliteration based on grapheme to phoneme mapping and cross-lingual pronunciation mapping models.
[0006] Another object of the present invention is to utilize a single, pre-trained transliteration model for all different languages for reducing time required for training multiple different artificial intelligence (AI) models.
[0007] Yet another object of the present invention is to transliterate text in any input language (or first language) to text comprising characters of a base language (or a second language) based on pronunciation.
[0008] Yet another object of the present invention is to utilize a conventional word mapping algorithm along with the pretrained transliteration model.
[0009] Yet another object of the invention is to utilize a character mapping algorithm along with the pretrained transliteration model
[0010] The objects disclosed above will be realized and achieved at least by the elements, features, and combinations particularly pointed out in the detailed de-scription of the present invention. The objects disclosed above have outlined, rather broadly, the features of the present invention in order that the detailed description that follows may be better understood. The objects disclosed above are not intended to determine the scope of the claimed subject matter and are not to be construed as limiting the present invention. Additional objects, features, and advantages of the present invention are disclosed below. The objects disclosed above, which are be-lieved to be characteristic of the present invention, both as to its organization and method of operation, together with further objects, features, and advantages, will be better understood and illustrated by the technical features broadly embodied and de-scribed in the following description when considered in connection with the accom-panying drawings.

D) SUMMARY OF THE INVENTION
[0011] The following details present a simplified summary of the embodi-ments herein to provide a basic understanding of the several aspects of the embodi-ments herein. This summary is not an extensive overview of the embodiments herein. It is not intended to identify key/critical elements of the embodiments herein or to delineate the scope of the embodiments herein. Its sole purpose is to present the con-cepts of the embodiments herein in a simplified form as a prelude to the more detailed description that is presented later.
[0012] The other objects and advantages of the embodiments herein will become readily apparent from the following description taken in conjunction with the accompanying drawings. It should be understood, however, that the following descriptions, while indicating preferred embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the spirit thereof, and the embodiments herein include all such modifications.
[0013] According to an embodiment of the present invention, a method for converting text in one of a plurality of input languages into a text in a second lan-guage using phonetic based transliteration is provided. The method includes receiving an input text in a first script from a user. Each character of the input text is phoneti-cally mapped with a second script corresponding to the second language. The permu-tations of mapping of each input character with each character of the second script is validated and the input text in the first script is transliterated into an output text in the second script.
[0014] According to an embodiment of the present invention, the step of transliterating includes performing a machine transliteration using an artificial intelli-gence (AI)-based transliteration engine executable by at least one processor for con-verting text in any input language into output text.
[0015] According to an embodiment of the present invention, the step of transliterating includes transliterating text in an input language to a text including one or more characters of a base language.
[0016] According to an embodiment of the present invention, the step of transliterating includes transliterating text using a speech transliteration engine, and wherein a text includes Latin or English characters to a text including characters of Devanagari or Hindi characters, based on mapping of the phonetics instead of word mapping using a pre-trained artificial intelligence (AI) model.
[0017] According to an embodiment of the present invention, the AI based transliteration engine is integrated into an input interface of a user device.
[0018] According to an embodiment of the present invention, the AI based transliteration engine utilizes a Unicode symbol sequence.
[0019] According to an embodiment of the present invention, the AI based transliteration engine is configured to utilize expectation maximization (EM) algo-rithm as an approach for performing maximum likelihood estimation in the presence of latent variables. The latent variables are the variables not directly observed and are actually inferred from the values of the other observed variables.
[0020] According to an embodiment of the present invention, the expectation maximization (EM) algorithm is used for the latent variables to predict the values with the condition that a general form of probability distribution governing the latent variables is known.
[0021] According to an embodiment of the present invention, the method fur-ther includes performing a grapheme to phoneme (G2P) conversation using a per-symbol alignment of an input string and an output string. Grapheme-to-Phoneme (G2P) conversion is a technique related to Natural Language Processing, Speech Recognition and Spoken Dialog Systems development.
[0022] According to an embodiment of the present invention, a primary goal of G2P conversion is to accurately predict the pronunciation or transliteration of a novel input word given the spelling. The G2P conversion process is typically broken down into several sub-processes. The subprocesses includes (1) Sequence alignment, (2) Model training and, (3) Decoding processes. The goal of Sequence alignment pro-cess is to align the grapheme and phoneme sequence pairs in a training dictionary. The goal of Model training process is to produce a model to generate new transliteration for novel words. The goal of Decoding process is to find the most likely pronuncia-tion given the model.
[0023] According to an embodiment of the present invention, a method for converting input text into output text is provided. The method comprises the step of checking if the user device is downloaded with the weighted finite-state transducer (WFST) algorithm. The input text is provided to the WFST model. The input text is converted to a first output text, when the WFST algorithm is downloaded. The meth-od still further includes proceeding with the word mapping algorithm as the fallback when the WFST model is not downloaded in the user device. The method further in-cludes checking when the input text matches with a prestored native words through direct mapping process. The method further includes forwarding the input text through the word mapping algorithm and generating a second output text if input matches with the prestored native words and forwarding the input text through the character mapping algorithm and generating a third output text if the input does not match with the prestored native words.
[0024] According to an embodiment, the WFST is a finite-state machine in-cludes two memory tapes, including an input tape and an output tape for tuning ma-chines.
[0025] According to an embodiment, the word mapping algorithm is based on direct mappings of the words.
[0026] According to an embodiment, the character mapping algorithm implies mapping of characters of the input text to phonetically similar sounding characters of the output text.
[0027] In another aspect a system for phonetic-based transliteration is provid-ed. The system includes a memory for storing one or more executable modules and a processor for executing the one or more executable modules for phonetic-based trans-literation. The one or more executable modules includes a transliteration engine con-figured to transliterate input text of first language into the output text of second lan-guage. The transliteration engine includes a data reception module for receiving an input text in an input language, a data transformation module is used for transforming the input text into a transliterated text including one or more characters in a second language, a training module for training a pre-trained model and decoding an output text of the trained model to generate text including characters of the second lan-guage, an inference module for executing the inference stage by receiving a text file as input, processing the input text data through the pre-trained language model and generating output text data in a second language; and a database is used for storing text files received as input text for transliteration and a corpus containing large da-tasets of curated and augmented texts.
[0028] According to an embodiment, the encoder is configured to train a pre-trained model with the data files and corresponding transliterated text using transfer learning.
[0029] According to an embodiment, the decoder is configured to perform decoding and the decoder improves the accuracy of the generated text including characters of the second language.
[0030] According to an embodiment, the data transformation module translit-erates the generated text to output text including characters in the second language.
[0031] According to an embodiment, the transliteration engine is executed by the processor and causes the processor to transliterate input text of first language into the output text of second language.

E) BRIEF DESCRIPTION OF THE DRAWINGS
[0032] The foregoing summary, as well as the following detailed description, is better understood when read in conjunction with the appended drawings. For illus-trating the present invention, exemplary constructions of the present invention are shown in the drawings. However, the present invention is not limited to the specific methods and components disclosed herein. The description of a method step or a component referenced by a numeral in a drawing is applicable to the description of that method step or component shown by that same numeral in any subsequent draw-ing herein.
[0033] FIG. 1 illustrates a block diagram of a system for transliterating text in one of a plurality of input languages into out text using machine transliteration, ac-cording to an embodiment of the present invention.
[0034] FIG. 2 illustrates a fragment topology of a bigram pair language model as a WFST, according to an embodiment of the present invention.
[0035] FIG. 3 illustrates a flowchart of an implementation of a method for converting input text into output text using combination of algorithms of WFST, character mapping algorithm, and the word mapping algorithm, according to an em-bodiment of the present invention.
[0036] FIG. 4A- 4C exemplarily illustrates a graphical representation dis-played on a display unit of an electronic device, showing a transliterated text sugges-tions on a suggestion bar interface, according to an embodiment of the present inven-tion.
[0037] FIG. 5 illustrates an architectural block diagram of an exemplary im-plementation of a system for transliterating text in one of a plurality of input lan-guages into out text using machine transliteration, according to an embodiment of the present invention.
[0038] FIG. 6A- 6C illustrates the finite-state machine, finite-state transducer, and a weighted finite-state transducer, according to an embodiment of present inven-tion.
[0039] FIG. 7 illustrates various semiring types, according to an embodiment of present invention.
[0040] FIG. 8 illustrates a flowchart of a method for converting text in one of a plurality of input languages into a second language text using phonetic based trans-literation, according to an embodiment of the present invention.
[0041] FIG. 9 illustrates a flowchart of a method for converting input text in-to output text, according to an embodiment of the present invention.

F) DETAILED DESCRIPTION OF THE EMBODIMENTS
[0042] The detailed description of various exemplary embodiments of the dis-closure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly com-municate the disclosure. However, the amount of details provided herein is not in-tended to limit the anticipated variations of embodiments; on the contrary, the inten-tion is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.
[0043] It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
[0044] While the disclosure is susceptible to various modifications and alter-native forms, specific embodiment thereof has been shown by way of example in the drawings and will be described in detail below. It should be understood, however that it is not intended to limit the disclosure to the forms disclosed, but on the contra-ry, the disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.
[0045] The embodiments herein provide a system and method for translitera-tion based on graphene to phoneme mapping and cross-lingual pronunciation mapping models. The present invention transliterates text in any input language (or first lan-guage) to text comprising characters of a base language (or a second language) based on pronunciation. The system and method utilize a conventional word mapping algo-rithm along with the pretrained transliteration model. In an embodiment, the present invention uses a character mapping algorithm along with the pretrained transliteration model for transliterating.
[0046] According to an embodiment of the present invention, a method for converting text in one of a plurality of input languages into a text in a second lan-guage using phonetic based transliteration is provided. The method comprising the steps pf receiving (802) an input text in a first script from a user; phonetically mapping (804) each character of the input text with a second script corresponding to the second language; validating (806) permutations of mapping of each input charac-ter with each character of the second script; and transliterating (808) the input text in the first script into an output text in the second script.
[0047] According to an embodiment of the present invention, the step of transliterating comprises performing a machine transliteration using an artificial intel-ligence (AI)-based transliteration engine (106) executable by at least one processor for converting text in any input language into output text.
[0048] According to an embodiment of the present invention, the step of transliterating comprises transliterating text in an input language to a text comprising one or more characters of a base language.
[0049] According to an embodiment of the present invention, the step of transliterating comprises using a speech transliteration engine to transliterate a text comprising Latin or English characters to a text comprising characters of Devanagari or Hindi characters, based on mapping of the phonetics instead of word mapping us-ing a pre-trained artificial intelligence (AI) model.
[0050] According to an embodiment of the present invention, the AI based transliteration engine is integrated into an input interface (401) of a user device.
[0051] According to an embodiment of the present invention, the AI based transliteration engine utilizes a Unicode symbol sequence.
[0052] According to an embodiment of the present invention, the AI based transliteration engine is configured to utilize Expectation Maximization (EM) algorithm as an approach for performing maximum likelihood estimation in the pres-ence of latent variables.
[0053] According to an embodiment of the present invention the latent variables are the variables not directly observed and are actually inferred from the values of the other observed variables.
[0054] According to an embodiment of the present invention, an expectation maximization (EM) algorithm is used for latent variables to predict the values with the condition comprising a general form of probability distribution governing the latent variables is known.
[0055] According to an embodiment of the present invention, the method further comprises performing a grapheme to phoneme (G2P) conversation using a per-symbol alignment of an input string and an output string.
[0056] According to an embodiment of the present invention, a primary goal of G2P conversion is to accurately predict the pronunciation or transliteration of a novel input word given the spelling.
[0057] According to an embodiment of the present invention, a method for converting input text into output text is provided. The method comprising the steps of checking (301)/ (902) if the user device is downloaded with a weighted finite state transducer (WFST) algorithm; providing (302)/ (904) the input text to the WFST model and converting the input text to the output text-1, if the WFST algorithm is downloaded proceeding (906) with the word mapping algorithm as the fallback if the WFST model is not downloaded in the user device; checking (303)/ (908) if the input text matches with a prestored native words through direct mapping process; forward-ing (304)/ (910) the input text through the word mapping algorithm and generating output text-2, if input matches with the prestored native words; and forwarding (305)/ (912) the input text through the character mapping algorithm and generating output text-3, if the input does not match with the prestored native words.
[0058] According to an embodiment of the present invention, the weighted finite-state transducer (WFST) is a finite-state machine comprising two memory tapes, following the terminology for tuning machines comprising an input tape and an out-put tape.
[0059] According to an embodiment of the present invention, the word mapping algorithm is implemented for direct mappings of the words.
[0060] According to an embodiment of the present invention, the character mapping algorithm is executed for mapping of characters of the input text to phonetically similar sounding characters of the output text.
[0061] According to an embodiment of the present invention, a system for phonetic-based transliteration is provided. The system comprises a memory (102) for storing one or more executable modules; and a hardware processor (104) for execut-ing the one or more executable modules for phonetic-based transliteration, The one or more executable modules comprises a transliteration engine (106) configured to trans-literate input text of first language into the output text of second language, the trans-literation engine comprising: a data reception module (108) for receiving an input text in an input language; a data transformation module (110) for transforming the input text into a transliterated text comprising one or more characters in a second language; a training module (112) comprising an encoder (112a) and a decoder (112b) for train-ing a pre-trained model and decoding an output text of the trained model to generate text comprising characters of the second language; and an inference module (114) for executing the inference stage by receiving a text file as input, processing the input text data through the pre-trained language model, and generating output text data in a second language.
[0062] According to an embodiment of the present invention, the encoder (112a) is configured to train a pre-trained model with the data files and corresponding transliterated text using transfer learning.
[0063] According to an embodiment of the present invention, the decoder (112b) is configured to perform decoding and decoder improves the accuracy of the generated text comprising characters of the second language.
[0064] According to an embodiment of the present invention, the data trans-formation module (110) transliterates the generated text to output text comprising characters in the second language.
[0065] According to an embodiment of the present invention, the translitera-tion engine (106) is executed by the processor and causes the processor to transliterate input text of first language into the output text of second language.
[0066] FIG. 1 illustrates a block diagram of a system for transliterating text in one of a plurality of input languages into out text using machine transliteration, ac-cording to an embodiment of the present invention. In an embodiment, the system 100 for phonetic-based transliteration, includes:
- a memory (102) for storing one or more executable modules; and
- a processor (104) for executing the one or more executable modules for phonetic-based transliteration, the one or more executable modules comprising:
-a transliteration engine (106) configured to transliterate input text of first language into the output text of second language, the translitera-tion engine comprising:
- a data reception module (108) for receiving an input text in an input language;
-a data transformation module (110) for transforming the input text into a transliterated text comprising one or more characters in a second language;
-a training module (112) comprising an encoder (112a) and a decoder (112b) for training a pre-trained model and decoding an out-put text of the trained model to generate text comprising characters of the second language; and
-an inference module (114) for executing the inference stage by receiving a text file as input, processing the input text data through the pre-trained language model, and generating output text data in a sec-ond language.
[0067] In an embodiment, the modules of the transliteration engine 106 are stored in the memory unit 102. The processor 104 is operably and communicatively coupled to the memory unit 102 for executing the computer program instructions de-fined by the modules of the transliteration engine 106. The transliteration engine 106 is not limited to employing the processor 104. In an embodiment, the transliteration engine 106 employs one or more controllers or microcontrollers. The transliteration engine 106 comprises modules defining computer program instructions, which when executed by the processor 104, cause the processor 104 to transliterate input text of first language into the output text of second language. The database 116 stores, for example, text files received as input text for transliteration and a corpus containing large datasets of curated and augmented texts. The data reception module 108 re-ceives an input text in any input language, for example, Latin characters of English language. The data transformation module 110 transforms the input text into translit-erated text comprising characters of a second language. In an embodiment, the train-ing module 112 comprises an encoder 112a and a decoder 112b. The encoder 112a trains a pre-trained model with the data files and corresponding transliterated text using transfer learning. The acoustic model is pre-trained on multiple datasets of the base language. The decoder 112b performs decoding, for example, an output text of the trained model to generate text comprising characters of the second language. In an embodiment, decoder 112b improves the accuracy of the generated text compris-ing characters of the second language, for example, Hindi, by using a pre-trained cus-tomized language model. The data transformation module 110 then transliterates the generated text to output text comprising characters in the second language. The infer-ence module 114 executes the inference stage, where the inference module 114 re-ceives a text file as input, processes the input text data through the pre-trained lan-guage model, and in an embodiment, through the pre-trained customized language model, and generates output text data in a second language.
[0068] FIG. 2 illustrates a fragment topology of a bigram pair language model as a weighted finite-state transducer (WFST), according to an embodiment of the present invention. The WFST is a state machine which validates every input character and if the input is matched, there is some output that corresponds to every input state. As per FIG. 2, the finite state machine converts the input sequence into the output sequence of native characters which, on reaching a final state, provide the final out-put. The bigram has the incoming arcs into any state which are labeled with the same pair symbol, e.g., state 4. In one embodiment, the WFST model is trained with openfst which helps in creating a big state machine from all the valid inputs provided at the time of training the model that is further decompiled using the same openfst. For example, sample sets of Hindi training data to create a state machine model for Hindi language is shown in the below table as per FIG. 2.

TABLE. 1
[0069] FIG. 3 illustrates a flowchart of an implementation of a method for converting input text into output text using combination of algorithms of WFST, character mapping algorithm, and the word mapping algorithm, according to an em-bodiment of the present invention. As per FIG. 3, in one embodiment, the translitera-tion engine is configured with a fallback combination of three algorithms of WFST, character mapping algorithm, and the word mapping algorithm. In one embodi-ment, the user device is configured with the transliteration engine wherein the trans-literation engine, at 301, is configured to check if the user device is downloaded with the WFST model. If yes, the transliteration engine 106 is configured to provide the input text to the WFST model 302 and convert the input text to the output text 1 as described further in FIG. 8. Alternatively, if the WFST model 302 is not downloaded in the user device, the transliteration engine 106 is configured to proceed with the word mapping algorithm as the fallback. The word mapping algorithm is based on di-rect mappings of the words. For example, the word mapping utilizes few direct map-pings of Latin words as to a corresponding native word such as Devanagari word. If the user input Latin word matches exactly to the native word, the corresponding na-tive word is shown as the output. At 303, the transliteration engine 106 checks if the input text matches with a prestored native words through direct mapping process. If yes, at 304, the transliteration engine 106 forwards the input text through the word mapping algorithm and generates output text 2. If no, at 305, the translitera-tion engine 106 forwards the input text through the character mapping algorithm and generates output text 3.
[0070] In one embodiment, the character mapping algorithm implies mapping of characters of the input text to phonetically similar sounding characters of the out-put text. In one example, the characters of the Latin alphabet are mapped to Devana-gari language characters. The process of mapping incudes a set of Latin (English) mapped words are taken for any particular language, such as Devanagari. proceeding further, word-by-word, the phonetics of the word are mapped to the Latin characters' phonemes. A list of words, corresponding native word mapping is given to a native language expert who breaks the word phonetics mapping for us after which the below file is made which is called phonetics character mapping file. For an example of the Devanagari word, consider below mapping table,

TABLE. 3
[0071] The above part is creation of phonetics mapping which is known as en-coding. After this the decoding is carried out by the transliteration engine which mainly contributes to the conversion of input txt to corresponding transliterated out-put text. In one example, for decoding, the Latin input is taken from the user, and the word is broken down into the characters with all the possible permutations availa-ble in the Latin character mapping set. For example, if input is received as ‘kuch’ then the possible permutations would be “k u c h”, “k u ch''. Then all the permutations for the possible phonemes will be calculated by replacing the Devanagari phoneme map-ping for each Latin sound available. For the input text “kuch” below available map-pings are the possible permutations,

TABLE. 2
[0072] As per Table. 2, possible permutations for “k u c h” => , and possible permutations for “k u ch” => , . Out of given permutations, the only output which is rele-vant here is , which is validated from a given dictionary of the respective lan-guage. After validation, all the other outputs will be discarded, and the final output will be “ ” for input Latin word “kuch.”
[0073] In one embodiment, the transliteration engine 106 checks if the keyboard of the user device is a QWERTY keyboard. If not, the transliteration engine 106 is configured to proceed with the word mapping algorithm as the fallback. The word mapping algorithm is based on direct mappings of the words. The transliteration engine 106 checks if the input text matches with a prestored native words through direct mapping process. If yes, the transliteration engine 106 forwards the input text through the word mapping algorithm and generates output text 2. If no, the translit-eration engine 106 forwards the input text through the character mapping algorithm and generates output text 3.
[0074] FIG. 4A- 4C exemplarily illustrates a graphical representation dis-played on a display unit of an electronic device, showing a transliterated text sugges-tions on a suggestion bar interface 403, according to an embodiment of the present invention. When a user invokes an input interface 401, for example, the keyboard in-terface 402, through a user application, the transliteration engine 106 displays a prede-termined number of transliterated suggestions for the input text default in the sugges-tion bar 403 positioned in a row of the keyboard interface 402. In an embodiment, the transliteration engine 106 displays transliterated suggestions and predictions above the keyboard interface 402. For example, the user input the word ‘sheershak’ in the typing bar 404 and the transliteration engine 106 generates a plurality of suggestions in the suggestion bar 403.
[0075] FIG. 5 illustrates an architectural block diagram of an exemplary im-plementation of the system 100 for converting input text in first language into output text in the second language using machine transliteration, in a computing device 501, according to an embodiment of the present invention. In an embodiment, the AI-based transliteration engine 106 (used interchangeably with the term transliteration engine 106) of the system 100 disclosed herein is deployed in the computing device 501 as exemplarily illustrated in FIG. 5. The computing device 501 is a computer sys-tem programmable using high-level computer programming languages. The compu-ting device 501 is an electronic device, including for example, one or more of a per-sonal computer, a tablet computing device, a mobile computer, a mobile phone, a smart phone, a portable computing device, a laptop, a personal digital assistant, a wearable computing device such as smart glasses, a smart watch, a touch centric de-vice, a workstation, a client device, a server, a portable electronic device, a network enabled computing device, an interactive network enabled communication device, an image capture device, any other suitable computing equipment, combinations of mul-tiple pieces of computing equipment, and the like. In an embodiment, the translitera-tion engine 106 is implemented in the computing device 501 using a programmed and purposeful hardware. In an embodiment, the transliteration engine 106 is a comput-er-embeddable system that converts text of first language into output text of sec-ond language using machine transliteration.
[0076] In an embodiment, the transliteration engine 106 is accessible to users, for example, through a broad spectrum of technologies and user devices such as smart phones, tablet computing devices, endpoint devices, and the like, with access to a network, for example, a short-range network or a long-range network. The network is, for example, one of the internets, an intranet, a wired network, a wireless network, a network that implements Wi-Fi® of Wi-Fi Alliance Corporation, a mobile telecommunication network, etc., or a network formed from any combination of these networks.
[0077] As illustrated in FIG. 5, the computing device 501 comprises at least one processor 104 and a non-transitory, computer-readable storage medium, for example, a memory unit 102, for storing computer program instructions defined by modules, for example, 108, 110, 112, 114, etc., of the transliteration engine 106. In an embodiment, the modules, for example, 108, 110, 112, 114, etc., of the transliteration engine 106 are stored in the memory unit 102 as illustrated in FIG. 5. The processor 104 is operably and communicatively coupled to the memory unit 102 for executing the computer program instructions defined by the modules, for example, 108, 110, 112, 114, etc., of the transliteration engine 106. The processor 104 refers to any one or more microprocessors, central processing unit (CPU) devices, finite state machines, computers, microcontrollers, digital signal processors, logic, a logic device, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a chip, etc., or any combination thereof, capable of executing computer programs or a series of commands, instructions, or state transitions. The transliteration engine 106 is not limited to employing the processor 104. In an embodiment, the transliteration engine 106 employs one or more controllers or microcontrollers.
[0078] As illustrated in FIG. 5, the computing device 501 comprises a data bus 513, a display unit 503, a network interface 504, and common modules 505. The data bus 513 permits communications between the modules, for example, 502, 503, 504, 505, and 506. The display unit 503, via a graphical user interface (GUI) 401, displays information, display interfaces, user interface elements such as checkboxes, input text fields, etc., for example, for allowing a user to invoke and execute the transliteration engine 106, input data and perform input actions for triggering various functions such as configuring a beam width for beam search decoding, and the like.
[0079] The network interface 504 enables connection of the transliteration engine 106 to the network. The network interface 504 is, for example, one or more of infrared interfaces, interfaces implementing Wi-Fi® of Wi-Fi Alliance Corporation, universal serial bus interfaces, FireWire® interfaces of Apple Inc., interfaces based on transmission control protocol/internet protocol, interfaces based on wireless communi-cations invention such as satellite invention, radio frequency invention, near field communication, etc. The common modules 505 of the computing device 501 com-prise, for example, input/output (I/O) controllers, input devices, output devices, fixed media drives such as hard drives, removable media drives for receiving removable media, etc. Computer applications and programs are used for operating the translitera-tion engine 106. The programs are loaded onto fixed media drives and into the memory unit 102 via the removable media drives. In an embodiment, the computer applications and programs are loaded into the memory unit 102 directly via the network.
[0080] In an embodiment, the transliteration engine 106 comprises modules defining computer program instructions, which when executed by the hardware pro-cessor 104, cause the processor 104 to transliterate input text of first language into the output text of second language. In an embodiment, the modules of the transliteration engine 106 comprise a data reception module 108, a data transformation module 110, a training module 112, an inference module 114, and a database 116. The data-base 116 stores, for example, text files received as input text for transliteration and a corpus containing large datasets of curated and augmented texts. The data reception module 108 receives an input text in any input language, for example, Latin characters of English language. The data transformation module 110 transforms the input text into transliterated text comprising characters of a second language, for example, Hin-di language with Devanagari characters, using transliteration, for example, WFST model, along with word mapping model, and character mapping, as disclosed in the detailed description of FIG. 3.
[0081] In an embodiment, the training module 112 comprises an encoder 112a and a decoder 112b. The encoder 112a trains a pre-trained model with the data files and corresponding transliterated text using transfer learning. The acoustic model is pre-trained on multiple datasets of the base language. The decoder 112b performs decoding, for example, an output text of the trained model to generate text compris-ing characters of the second language, for example, Hindi, as disclosed in the detailed description of FIGS. 1-3. In an embodiment, decoder 112b improves the accuracy of the generated text comprising characters of the second language, for example, Hindi, by using a pre-trained customized language model. The data transformation module 110 then transliterates the generated text to output text comprising characters in the second language, for example, Devanagari characters. The inference module 114 executes the inference stage, where the inference module 114 receives a text file as input, processes the input text data through the pre-trained language model, and in an embodiment, through the pre-trained customized language model, and generates out-put text data in a second language.
[0082] The data reception module 108, the data transformation module 110, the training module 112, and the inference module 114 are disclosed above as soft-ware executed by the processor 104. In an embodiment, the modules, for example, 508, 509, 510, and 511 of the transliteration engine 106 are implemented completely in hardware. In another embodiment, the modules of the transliteration engine 106 are implemented by logic circuits to perform their respective functions disclosed above. In another embodiment, the transliteration engine 106 is also implemented as a com-bination of hardware and software including one or more processors, for example, 502, that are used to implement the modules, for example, 108, 110, 112 and 114 of the transliteration engine 106. The processor 104 retrieves instructions defined by the data reception module 108, the data transformation module 110, the training module 112, and the inference module 114 from the memory unit 102 for performing respec-tive functions disclosed above. The non-transitory, computer-readable storage medi-um disclosed herein stores computer program instructions executable by the proces-sor 104 for converting text using machine transliteration.
[0083] FIG. 6A- 6C illustrates the finite-state machine, finite-state transducer, and a weighted finite-state transducer, according to an embodiment of present inven-tion. A finite-state transducer (FST) is a finite-state machine with two memory tapes, following the terminology for Tuning machines: an input tape and an output tape. This contrasts with an ordinary finite-state automaton, which has a single tape. An FST is a type of finite-state automaton (FSA) that maps between two sets of symbols. An FST is more general than an FSA. An FSA defines a formal language by defining a set of accepted strings, while an FST defines relations between sets of strings. An FST will read a set of strings on the input tape and generate a set of relations on the output tape. An FST can be thought of as a translator or relater between strings in a set. When FSTs are added with weights, where each transition is labeled with a weight in addition to the input and output labels.
[0084] A Weighted Finite State Transducer (WFST) over a set K of weights can be defined as an 8-tuple T= (Q, S, G, I, F, E, ?, ?), where:
Q is a finite set, the set of states;
S is a finite set, called the input alphabet;
G is a finite set, called the output alphabet;
I is a subset of Q, the set of initial states;
F is a subset of Q, the set of final states; and
E ? Q × (S ? { ?}) × (G ? { ?}) × Q × K (where e is the empty string) is the finite set of transitions;
?: I ? K maps initial states to weights;
?: F ? K maps final states to weights.
In order to make certain operations on WFSTs well-defined, it is convenient to require the set of weights to form a semiring. Two typical semirings used in practice are the log semiring and tropical semiring: nondeterministic automata may be regarded as having weights in the Boolean semiring.
Stochastic FSTs (also known as probabilistic FSTs or statistical FSTs) are a form of weighted FST.
[0085] FIG. 7 illustrates various semiring types, according to an embodiment of present invention. The OpenFST is an open-source library for weighted finite-state transducers (WFSTs). The OpenFST consists of a C++ template library with efficient WFST representations and over 25 operations for constructing, combining, optimiz-ing, and searching them. At the shell-command level, there are corresponding trans-ducer file representations and programs that operate on them. The OpenFST is de-signed to be both very efficient in time and space and to scale to exceptionally large problems. This library has key applications in speech, image, and natural language processing, pattern and string matching and machine learning. The OpenFST Library closely parallels its mathematical foundations in the theory of rational power series. The library user can define the alphabets and weights that label transitions. The weights may represent any set so long as they form a semiring. A semiring (K, ?, ?, 0, 1) is specified by a set of values K, two binary operations ? and ?, and two designated values 0 and 1. The operation ? is associative, commutative, and has 0 as identity. The operation ? is associative, has identity 1, distributes with respect to ?, and has 0 as annihilator: for all a ? K, a ? 0 = 0 ? a = 0. If ? is also commuta-tive, we say that the semiring is commutative.
[0086] A WFST, T = (A, B, Q, I, F, E, ?, ?) over a semiring K is specified by a finite input alphabet A, a finite output alphabet B, a finite set of states Q, a set of initial states I ? Q, a set of final states F ? Q, a finite set of transitions E ? Q × (A ? {e}) × (B ? {e}) × K × Q, an initial state weight assignment ? : I ? K, and a final state weight assignment ? : F ? K. E[q] denotes the set of transitions leaving state q ? Q.
[0087] Given a transition e ? E, p[e] denotes its origin or previous state, n[e] its destination or next state, i[e] its input label, o[e] its output label, and w[e] its weight. A path p = e1 · · · ek is a sequence of consecutive transitions: n[ei-1] = p[ei], i = 2, . . ., k. The functions n, p, and w on transitions can be extended to paths by set-ting: n[p] = n[ek] and p[p] = p[e1] and by defining the weight of a path as the ?-product of the weights of its constituent transitions: w[p] = w[e1] ? · · · ? w[ek ]. More generally, w is extended to any finite set of paths R by setting w[R] = ? p?R w[p]; if the semiring is closed, this is defined even for infinite R. We denote by P (q, q') the set of paths from q to q' and by P (q, x, y, q') the set of paths from q to q' with input label x ? A* and output label y ? B*. These definitions can be extended to sub-sets R, R' ? Q by P (R, R') = ?q?R, q' ?R' P (q, q'), P (R, x, y, R') = ?q?R, q' ?R' P (q, x, y, q').
[0088] A transducer T is regulated if the weight associated by T to any pair of input- output string (x, y) given by:
[[T]] (x, y) = ? ?[p[p]] ? w[p] ? ?[n[p]] p?P (I, x, y, F)
is well-defined and in K. If P (I, x, y, F) = Ø, then T (x, y) = 0. A weighted transducer without e-cycles is regulated.
[0089] FIG. 8 illustrates a flowchart of a method for converting text in one of a plurality of input languages into a second language text using phonetic based trans-literation, according to an embodiment of the present invention. The method dis-closed herein employs an artificial intelligence (AI)-based transliteration engine exe-cutable by at least one processor for converting text in any input language into out-put text using machine transliteration. For purposes of illustration, the detailed description refers to a text input in an input language, for example, Latin characters, being converted into text comprising Devanagari characters; however, the scope of the method and the system disclosed herein is not limited to the output text being Devanagari characters but may be extended to include any Indian or Indic language or languages of other countries. The transliteration engine is configured to transliterate text in any input language to text comprising characters of a base language. In one embodiment, the speech transliteration engine is configured to transliterate text comprising Latin or English characters to text comprising characters of Devanagari or Hindi characters, based on mapping of the phonetics instead of word mapping using a pre-trained AI model.
[0090] The transliteration engine is integrated into an input interface of a user device. As used herein, “input interface” refers to an interface rendered on the user device, for example, a smartphone, for receiving one or more inputs from a user. For example, the input interface is a keyboard or a virtual keyboard that is invoked on the user device when a user clicks on an input field such as a text field provided by a user application such as a messaging application or a chat application. In the method dis-closed herein, the transliteration engine is integrated within the input interface invoked on the user device, independent of a user application, for example, a messag-ing or messenger application, a chat application, etc. As the input interface such as a virtual keyboard is opened and closed only during an input action into an input field of the user application, the operations and functions the engine is configured to be independent of the user application or any application that is present in the fore-ground.
[0091] In the method disclosed herein, the transliteration engine, at step 802, receives an input text in a first script from the user, for example, Latin or English characters. As used herein, the term “first script” refers to words or characters of a first language. For example, if the first language is English then the corresponding first script refers to Latin characters. The transliteration engine which is integrated with the input interface is configured to receive the input text in the first script which is further configured to be phonetically transliterated into a “second script.” For ex-ample, consider the input text contains the word ‘SANSKRIT’ which is inputted into the input interface by the user. The input text is further configured to be converted into output text of the second script. For example, consider the output text contains the word ‘???????’ transliterated from the ‘SANSKRIT.’
[0092] At 804, the transliteration engine is configured to phonetically map each grapheme (or character) of the input text with a second script. In one embodiment, for mapping each character of the input text with the phonemes based matching characters of the second script. For example, the transliteration engine is configured to align a Devanagari script of word ‘???????’ with its romanization ver-sion ‘Sanskrit.’ In one embodiment, the transliteration engine utilizes the Unicode symbol sequence, for example, the Unicode symbol sequence for the input text ‘San-skrit,’
‘??????????’: s: ?, a: ?, n: ??, s: ?, ?: ??, k: ?, r: ??, i: ?, t: ?
[0093] The Unicode symbols sequence on either the input or the output may not directly correspond to a symbol on the other side, that is, in the present example, ‘a,’ ‘i’ and ??, which is represented with an ? on the other side of the transduction. We make use of the Unicode symbol sequence (??????????) with its Romanised word Sanskrit.
s: ? a: e n: ?? s: ? e: ?? k: ? r: ?? i: e t: ?
[0094] Note that symbols on either the input or the output may not directly correspond to a symbol on the other side (such as ‘a,’ ‘i’ and ?? in the above exam-ple), which we represent with an e on the other side of the transduction. This explains the method of creating a training set for models based on define vocab. Expectation maximization (EM) is used to learn effective alignments of this sort. We built an n-gram model to produce joint probabilities over sequences of such pairs.
[0095] In one embodiment, the transliteration engine is configured to utilize Expectation Maximization (EM) algorithm as an approach for performing maximum likelihood estimation in the presence of latent variables. It can be used for the latent variables (variables that are not directly observable and are actually inferred from the values of the other observed variables) too in order to predict their values with the condition that the general form of probability distribution governing those latent variables is already known. This algorithm is actually at the base of many unsuper-vised clustering algorithms in the field of machine learning.
The algorithm comprises following steps:
1. Given a set of incomplete data, consider a set of starting parameters.
2. Expectation step (E–step): Using the observed available data of the dataset, estimate (guess) the values of the missing data.
3. Maximization step (M–step): Complete data generated after the expectation (E) step is used in order to update the parameters.
4. Repeat Expectation step and Maximization step until convergence.
5. Given a lexicon of words and their transliterations, expectation maximization (EM) is used to learn effective alignments of the input: output pairs sorting.
[0096] The lexicon of input words and their pronunciations or transliterations, for example, ‘sanskrit’ is a romanization of ???????, are straightforwardly used to learn effective alignments of input words with that of output words. In one embodiment, for grapheme to phoneme conversion, wherein the grapheme to phoneme conversation is a method of per-symbol alignment of both the input string and the output string. For example, the word “phlegm” is pronounced F L EH M and one natural alignment between the grapheme and phoneme sequences is: p: ? h: F l: L e:EH g: ? m: M. Grapheme-to-Phoneme (G2P) conversion is an important problem related to Natural Language Processing, Speech Recognition and Spoken Dialog Systems development. The primary goal of G2P conversion is to accurately predict the pronunciation or transliteration of a novel input word given only the spelling. The G2P conversion problem is typically broken down into several sub-problems: (1) Sequence alignment, (2) Model training and, (3) Decoding. The goal of (1) is to align the grapheme and phoneme sequence pairs in a training dictionary. The goal of (2) is to produce a model able to generate new transliteration for novel words. The goal of (3) is to find the most likely pronunciation given the model. The alignment comprises the proposed toolkit that implements a modified WFST-based version of the EM-driven multiple-to- multiple alignment algorithm. This algorithm is capable of learning natural G-P relationships like Sanskrit-> ? ? ? ? ? ? ? which were not possible with previous 1-to-1 algorithms. The Joint Sequence N-gram model is the transliteration model implemented by the toolkit is a straightforward joint N-gram model. The training corpus is constructed by extracting the best alignment for each entry. The training procedure is then,
(1) Convert aligned sequence pairs to sequences of aligned joint label pairs, (g 1: p 1, g 2: p 2, ..., g n: p n);
(2) Train an N-gram model from (1);
(3) Convert the N-gram model to a WFST.
Step (3) may be performed with any language modeling toolkit. In this invention MITLM is utilized.
[0097] The decoding comprises the proposed toolkit that provides varying support for three different decoding schemes. The default decoder provided by the distribution simply extracts the shortest path through the phoneme lattice created via composition with the input word. The recurrent Neural Network (RNN) Language Models have recently enjoyed a resurgence in popularity in the context of ASR applications. N-best reranking is then accomplished with the toolkit by configuring the decoder to output the N-best joint G-P sequences and employing RNNLM to re-rank the N-best joint sequences.
[0098] The aligned letter to phoneme sequence of input: output pairs (or alter-natively referred as ‘permutations’), for example, symbols such as e:EH, are used to build an n-gram model to produce joint probabilities over sequences of the pairs, wherein the n-gram models are referred as pair language models (alternatively called as ‘joint multi-gram models’). By conditioning the probability of the input: output mappings on the prior context, the transliteration engine appropriately condi-tions the probability of h: F on whether the previous mapping was p: ?. As stated above, results of these models yield remarkably similar performance to more complex and compute-intensive modeling methods, and they can be directly encoded as Weighted Finite state Transducers (WFSTs), making them excellent candidates for low resource, low-latency models for mapping graphemes of the input words with the phonemes of the output words.
[0099] At 806, the transliteration engine validates the permutations of map-ping of each input character with phoneme of every character of the second script. The transliteration engine is further configured to build n-gram model to produce joint probabilities over sequences of the pairs. In one embodiment, the transliteration en-gine is configured to score each permutation probability of input: output pair. For ex-ample, consider below scenarios where the score is calculated using joint probability of input: output pairs (s: ? a: e n: ?? s: ? e: ?? k: ? r: ?? i: e t: ?)
Input Word: Score: Output Transliteration sequence
Sanskrit: 11.56: ? ? ? ? ? ? ?
Sanskrit: 13.78: ? ? ? ? ? ? ? ?
Sanskrit: 19.34: ? ? ? ? ? ? ?
[0100] In an embodiment, for any weighted finite-state transducer, WFST is represented by ‘T’, wherein T=(S,?,Q,I,F,E,K) that is the ‘T’ includes input (S) and output (?) vocabularies, a finite set of states (Q), of one is the initial state (I), and a subset of states (F? Q) are final states; a weight semiring K; and a set of transitions(q,s,d,w,q') ? E, where q,q' ? Q are, respectively, the source and destina-tion states of the transition, are at s ?S,d ?? and w?K. Further, alteratively, a weighted finite-state automaton is a special case where S=? and, for every transition, the condition is (q, s, d, w, q') ? E, s=d. In an exemplary embodiment, the OpenFst library is used to encode and manipulate WFSTs, and, unless otherwise stat-ed, use the tropical semiring for weights. As per FIG. 8, WFSTs generate input: out-put pairs and the engine builds n-gram model to produce joint probabilities over se-quences of such pairs. At 808, the transliteration engine transliterates the input text in the first script into the output text of second script based on the WFST model.
[0101] Additionally, in one embodiment, the transliteration engine is config-ured to personalize the input text transliteration. As established, the Indian languages transliteration is very fuzzy. For example, “bahar” can be transliterated as “????” or “????.” The transliteration engine, by default, transliterate “Bahar” as “????” but the engine suggests the user with “????”as well. If the user picks the suggested word, then the engine stores the user choice of word and next time whenever user types “bahar” the default transliteration is “????.”In one embodiment, the translit-eration engine is configured with a character level filtration process to detect invalid words detection. For example, input text is “ok,” the first generated output for the word “ok” is ? ?, which is an Invalid word and after filtration, the suggested output is ??.
[0102] FIG. 9 illustrates the flowchart of a method for converting input text into output text, according to an embodiment of the present invention. At step 902, the user device is checked if it is downloaded with a weighted finite state transducer (WFST) algorithm. At step 904, the input text is provided to the WFST model, and the input text is converted to the output text 1 if the WFST algorithm is downloaded. At step 906, the word mapping algorithm proceeds as the fallback if the WFST model is not downloaded in the user device. At step 908, the input text is checked to see if it matches with a prestored native words through direct mapping process. At step 910, the input text is forwarded through the word mapping algorithm and output 2 is generated if input matches with the prestored native words. At step 912, the input text is forwarded through the character mapping algorithm and output 3 is generated if the input does not match with the prestored native words.
[0103] The present invention is based on an AI-based transliteration engine for conversion of text in first language into text in second language based on combi-nation of algorithms. The present invention has multiple applications involving text-to-text conversations from Latin to Hindi, other Indic languages such as Tamil, Telu-gu, Kannada, Malayalam, or any other language spoken in the world. The present in-vention can be used by third parties, research industries, firms or academic institutions working on transliteration, businesses requiring data-driven strategies, research-based industries, software sectors, cloud-based companies, AI-based conversation media entities, etc. The present invention precludes the need for investing substantial amounts of money, time, and human resources on building AI models for speech recognition for multiple languages.
[0104] The foregoing examples and illustrative implementations of various embodiments have been provided merely for explanation and are in no way to be con-strued as limiting the present invention. While the present invention has been de-scribed with reference to various embodiments, illustrative implementations, draw-ings, and techniques, it is understood that the words, which have been used herein, are words of description and illustration, rather than words of limitation. Furthermore, although the present invention has been described herein with reference to means, ma-terials, embodiments, techniques, and implementations, the present invention is not intended to be limited to the particulars disclosed herein; rather, the present invention extends to all functionally equivalent structures, methods and uses, such as are within the scope of the appended claims. It will be understood by those skilled in the art, having the benefit of the teachings of this specification, that the present invention is capable of modifications and other embodiments may be effected and changes may be made thereto, without departing from the scope and spirit of the present invention.

I/We Claim :

1. A method for converting text in one of a plurality of input languages into a text in a second language using phonetic based transliteration, the method comprising:
receiving (802) an input text in a first script from a user;
phonetically mapping (804) each character of the input text with a second script corresponding to the second language;
validating (806) permutations of mapping of each input character with each character of the second script; and
transliterating (808) the input text in the first script into an output text in the second script.

2. The method as claimed in claim 1, wherein transliterating comprises performing a machine transliteration using an artificial intelligence (AI)-based transliteration engine (106) executable by at least one processor for converting text in any input language into output text.

3. The method as claimed in claim 2, wherein the transliterating comprises transliterat-ing text in an input language to a text comprising one or more characters of a base language.

4. The method as claimed in claim 2, wherein transliterating comprises transliterating using a speech transliteration engine, a text comprising Latin or English characters to a text comprising characters of Devanagari or Hindi characters, based on mapping of the phonetics instead of word mapping using a pre-trained artificial intelligence (AI) model.

5. The method as claimed in claim 2, wherein the AI based transliteration engine is integrated into an input interface (401) of a user device.

6. The method as claimed in claim 1, wherein the AI based transliteration engine uti-lizes a Unicode symbol sequence.

7. The method as claimed in claim 1, wherein the AI based transliteration engine is configured to utilize Expectation Maximization (EM) algorithm as an approach for performing maximum likelihood estimation in the presence of latent variables.

8. The method as claimed in claim 7, wherein the latent variables are the variables not directly observable and are actually inferred from the values of the other observed variables.

9. The method as claimed in claim 7, wherein an expectation maximization (EM) algo-rithm is used for latent variables to predict the values with the condition comprising a general form of probability distribution governing the latent variables is known.

10. The method as claimed in claim 1, further comprising performing a grapheme to phoneme (G2P) conversation using a per-symbol alignment of an input string and an output string.

11. The method as claimed in claim 10, wherein a primary goal of G2P conversion is to accurately predict the pronunciation or transliteration of a novel input word given the spelling.

12. A method for converting input text into output text, the method comprising:
checking (301)/ (902) if the user device is downloaded with a weighted finite state transducer (WFST) algorithm;
providing (302)/ (904) the input text to the WFST model and converting the input text to the output text 1 if the WFST algorithm is downloaded;
proceeding (906) with the word mapping algorithm as the fallback if the WFST model is not downloaded in the user device;
checking (303)/ (908) if the input text matches with a prestored native words through direct mapping process;
forwarding (304)/ (910) the input text through the word mapping algorithm and generating output 2 if input matches with the prestored native words; and
forwarding (305)/ (912) the input text through the character mapping algorithm and generating output 3 if the input does not match with the prestored native words.

13. The method as claimed in claim 12, wherein the weighted finite-state transducer (WFST) is a finite-state machine comprising two memory tapes, following the termi-nology for tuning machines comprising an input tape and an output tape.

14. The method as claimed in claim 12, wherein the word mapping algorithm is based on direct mappings of the words.

15. The method as claimed in claim 12, wherein the character mapping algorithm im-plies mapping of characters of the input text to phonetically similar sounding charac-ters of the output text.

16. A system for phonetic-based transliteration, the system comprising:
a memory (102) for storing one or more executable modules; and
a processor (104) for executing the one or more executable modules for phonet-ic-based transliteration, the one or more executable modules comprising:
a transliteration engine (106) configured to transliterate input text of first lan-guage into the output text of second language, the transliteration engine comprising:
a data reception module (108) for receiving an input text in an input lan-guage;
a data transformation module (110) for transforming the input text into a transliterated text comprising one or more characters in a second language
a training module (112) comprising an encoder (112a) and a decoder (112b) for training a pre-trained model and decoding an output text of the trained model to generate text comprising characters of the second language; and
an inference module (114) for executing the inference stage by receiving a text file as input, processing the input text data through the pre-trained lan-guage model, and generating output text data in a second language.

17. The system as claimed in claim 16, wherein the encoder (112a) is configured to train a pre-trained model with the data files and corresponding transliterated text us-ing transfer learning.

18. The system as claimed in claim 16, wherein the decoder (112b) is configured to perform decoding and decoder improves the accuracy of the generated text compris-ing characters of the second language.

19. The system as claimed in claim 16, wherein the data transformation module (110) transliterates the generated text to output text comprising characters in the second language.

20. The system as claimed in claim 16, wherein the transliteration engine (106) is exe-cuted by the processor and causes the processor to transliterate input text of first lan-guage into the output text of second language.