Method and user device for selectively enciphering and deciphering a data stream

Abstract

The present disclosure relates generally to data enciphering, and more particularly to method and user device for selectively enciphering and deciphering a data stream. The method comprises obtaining a data stream corresponding to a speech signal of a user, identifying at least one sensitive portion of the data stream, encrypting at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream, and generating a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion.

Description

METHOD AND USER DEVICE FOR SELECTIVELY ENCIPHERING AND DECIPHERING A DATA STREAM

[0001] The present disclosure relates generally to a field of data enciphering / deciphering, and more particularly, to a method and system for selectively enciphering / deciphering a data stream.

[0002] Conventionally, many systems rely on transfer of data from user devices to remote servers for the processing of queries and generation of responses. In such systems, when a user submits a query or interacts with the device, the device collects the relevant data and sends it to the server for analysis and response generation. The server, upon receiving the data, typically performs to understand the context of the query and generate a response to the query.

[0003] To properly process the data and generate response, server often decrypt the entire data stream, exposing sensitive portions of the data to potential risks. While encryption techniques are employed to protect data during transmission, conventional systems require sensitive information to be decrypted at the server to ensure that its context and meaning are preserved in the response. This creates a vulnerability, as the sensitive data is exposed during processing and may be accessible by unauthorized parties or third-party services. Additionally, when generating responses that include sensitive information, servers often include such data in a form that can be easily understood, further increasing the risk of exposing private details.

[0004] For instance, if a user asks a virtual assistant to "check my bank balance," the device sends the command to the server. To provide an accurate response, the server must decrypt the sensitive data present in the query to understand the request and generate a response. While the user benefits from the convenience of the voice assistant, the sensitive information is fully exposed during processing, raising privacy concerns. Additionally, when generating the response (e.g., "Your balance is $1,000"), the server must incorporate sensitive information into the response in a human-readable form, which further increases the risk of exposure.

[0005] The information disclosed in this background of the disclosure section is only for enhancement of understanding of the general background of the disclosure and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.

[0006] The present disclosure overcomes one or more shortcomings of the prior art and provides additional advantages discussed throughout the present disclosure. Additional features and advantages are realized through the techniques of the present disclosure. Other embodiments and aspects of the disclosure are described in detail herein and are considered a part of the claimed disclosure.

[0007] In a non-limiting embodiment of the present disclosure, a method for selectively enciphering a data stream is provided. The method comprises obtaining a data stream corresponding to a speech signal of a user. The method comprises identifying at least one sensitive portion of the data stream. The method further comprises encrypting at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. The method comprises generating a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion. The method may be performed by a user device.

[0008] In a non-limiting embodiment of the present disclosure, a first plurality of portions of the received data stream are identified using Natural language processing (NLP). The first plurality of portions includes at least one sensitive portion and at least one non-sensitive portion. A first sensitivity score for each of the first plurality of portions is determined. The sensitivity score for each of the first plurality of portions is associated with sensitivity characteristics of the respective portions. The method comprises determining a sensitivity level for each of the first plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context. The method comprises identifying the at least one sensitive portion and the at least one non-sensitive portion based on the determined sensitivity level.

[0009] In a non-limiting embodiment of the present disclosure, identifying the at least one sensitive portion and the at least one non-sensitive portion further comprises comparing a sensitivity level of a first portion of the plurality of portions with a predefined threshold sensitivity range. The predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range. The first portion is identified as a sensitive portion in response to the sensitivity level of the first portion being within at least one of the high threshold ranges, the medium threshold range, or the low threshold range. The first portion is identified as non-sensitive in response to the sensitivity level of the first portion being below the low threshold range.

[0010] In a non-limiting embodiment of the present disclosure, the method comprises generating a plurality of dynamic keys for the at least one sensitive portion based on the content surrounding the at least one sensitive portion within the data stream. At least one sensitive portion of the data is encrypted using one of the generated plurality of dynamic keys.

[0011] In a non-limiting embodiment of the present disclosure, the method comprises transmitting, to a server, the selectively encrypted data stream. The method comprises receiving, from the server, a response data stream comprising the at least one encrypted sensitive portion.

[0012] In a non-limiting embodiment of the present disclosure, the method comprises decrypting the at least one encrypted sensitive portion included in the response data stream. The method comprises modifying the decrypted response data stream based on the at least one sensitive portion to generate a modified data stream.

[0013] In a non-limiting embodiment of the present disclosure, the method comprises identifying a second plurality of portions of the decrypted response data stream using Natural language processing (NLP), wherein the second plurality of portions includes the at least one decrypted sensitive portion and at least one non-sensitive portion. The method comprises determining a third sensitivity score for each of the plurality of portions of the decrypted response data stream, wherein the first sensitivity score for each of the plurality of portions of the decrypted response data stream is associated with sensitivity characteristics of the respective portion. The method comprises determining a sensitivity level for each of the second plurality of portions based on the determined third sensitivity score and a fourth sensitivity score associated with user context and user device context. The method comprises identifying the at least one sensitive portion and at least one non-sensitive portion from the second plurality of portions based on the determined sensitivity level for each of the second plurality of portions.

[0014] In a non-limiting embodiment of the present disclosure, the method comprises comparing the sensitivity level of each of the second plurality of portions with a predefined threshold sensitivity range, wherein the predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range. The method comprises, in response to the sensitivity level of at least one portion of the second plurality of portions being within at least one of the high threshold range, the medium threshold range, or the low threshold range, modifying the at least one portion.

[0015] In a non-limiting embodiment of the present disclosure, a computer-readable storage medium storing one or more instructions is provided. The one or more instructions, when executed by at least one processor individually or collectively, cause the at least one processor to perform any one of the methods in accordance with the present disclosure.

[0016] In a non-limiting embodiment of the present disclosure, a user device for selectively enciphering a data stream is provided. The user device comprises at least one processor comprising processing circuitry. The user device comprises memory comprising one more storage media storing one or more instructions. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device to obtain a data stream corresponding to a speech signal of a user. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device identify at least one sensitive portion of the data stream. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device to encrypt the at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device to generate a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion.

[0017] The foregoing summary is illustrative only and is not intended to be in any way limiting. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features will become apparent by reference to the drawings and the following detailed description.

[0018] The embodiments of the disclosure itself, as well as a preferred mode of use, further objectives, and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings. One or more embodiments are now described, by way of example only, with reference to the accompanying drawings in which:

[0019] FIG. 1 illustrates a system of selectively enciphering / deciphering a data stream, in accordance with an embodiment of the present disclosure;

[0020] FIG. 2 illustrates the block diagram of a system for selectively enciphering a data stream, in accordance with the present disclosure;

[0021] FIG. 3 illustrates a block diagram of a user device for generating plurality of portions in the data stream using acoustic speech recognition process, in accordance with the exemplary mechanism;

[0022] FIG. 4 illustrates a pre-trained model for determining a second score associated with user context and user device context, in accordance with an embodiment of the present disclosure;

[0023] FIG. 5 illustrates an exemplary embodiment for determining a second score associated with user context and user device context by a pretrained model, in accordance with an embodiment of the present disclosure;

[0024] FIG. 6 illustrates an exemplary embodiment for determining a sensitivity level for each of the plurality of portions, in accordance with an embodiment of the present disclosure;

[0025] FIG. 7A illustrates an exemplary embodiment for an encrypting at least one sensitive portion of the data stream, in accordance with an embodiment of the present disclosure;

[0026] FIG. 7B illustrates an exemplary selectively encrypted query, in accordance with an embodiment of the present disclosure;

[0027] FIG. 8 illustrates block diagram of a server for generating a response for enciphered data stream, in accordance with an embodiment of the present disclosure;

[0028] FIG. 9 illustrates block diagram of a server for selectively deciphering a data stream, in accordance with an embodiment of the present disclosure;

[0029] FIG. 10 illustrates an exemplary embodiment for selectively deciphering a data stream received from the server, in accordance with the present disclosure;

[0030] FIG. 11 illustrates flowchart of method for selectively enciphering a data stream, in accordance with an embodiment of the present disclosure;

[0031] FIG. 12 illustrates flowchart of method for selectively deciphering a data stream, in accordance with an embodiment of the present disclosure; and

[0032] FIG. 13 illustrates flowchart of method for generating a response for enciphered data stream, in accordance with an embodiment of the present disclosure;

[0033] It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of the illustrative systems embodying the principles of the present subject matter. Similarly, it will be appreciated that any flowchart, flow diagrams, state transition diagrams, pseudo code, and the like represent various processes which may be substantially represented in computer readable medium and executed by a computer or processor, whether or not such computer or processor is explicitly shown.

[0034] In the present document, the word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

[0035] While the disclosure is susceptible to various modifications and alternative 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 particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternative falling within the scope of the disclosure.

[0036] The terms "comprises," "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a setup, device, or method that comprises a list of components or steps does not include only those components or steps but may include other components or steps not expressly listed or inherent to such setup or device or method. For example, one or more elements in a system or apparatus proceeded by "comprises ... a" does not, without more constraints, preclude the existence of other elements or additional elements in the system or apparatus.

[0037] The phrase "artificial intelligence (AI)" refers to the use and / or development of computing devices and / or systems that are able to perform tasks normally associated with human intelligence. The AI model may consist of a plurality of neural network layers. Each layer has a plurality of weight values and performs a layer operation through calculation of a previous layer and an operation of a plurality of weights. Examples of neural networks include, but are not limited to, convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), restricted Boltzmann Machine (RBM), deep belief network (DBN), bidirectional recurrent deep neural network (BRDNN), generative adversarial networks (GAN), and deep Q-networks.

[0038] The phrase "machine learning" is used throughout the disclosure. The machine learning broadly describes a function of systems that learn from data. A machine learning system, engine, or module can include a machine learning algorithm that can be trained to learn functional relationships between inputs and outputs that are currently unknown. In one or more embodiments, machine learning functionality can be implemented using the neural networks having the capability to be trained to perform a currently unknown function.

[0039] Machine learning may be classified into supervised learning, unsupervised learning, and reinforcement learning according to a learning method. The supervised learning may refer to a method of learning an artificial neural network in a state in which a label for learning data is given, and the label may mean the correct answer (or result value) that the artificial neural network must infer when the learning data is input to the artificial neural network. Unsupervised learning may refer to a method of learning an artificial neural network in a state in which a label for learning data is not given.

[0040] Here, being provided through learning means that, by applying a learning algorithm to a plurality of learning data, a predefined operating rule or AI model of a desired characteristic is made. The learning may be performed in a device / apparatus itself in which AI according to an embodiment is performed, and / or may be implemented through a separate server / system.

[0041] The learning algorithm is a method for training a device / apparatus using a plurality of learning data to cause, allow, or control the target device to make a determination or prediction. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0042] In the following detailed description of the embodiments of the disclosure, reference is made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration specific embodiments in which the description may be practiced. These embodiments are described in sufficient detail to enable those skilled in art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is, therefore, not to be taken in a limiting sense.

[0043] The terms like "at least one" and "one or more" may be used interchangeably throughout the description. The terms like "a plurality of" and "multiple" may be used interchangeably throughout the description. The terms like "network" and "communication network" may be used interchangeably throughout the description.

[0044] Turning now to the drawings, the detailed description set forth below in connection with the appended drawings is intended as a description of various configurations and is not intended to represent the only configurations in which the concepts described herein may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of various concepts with like numerals denote like components throughout the several views. However, it will be apparent to those skilled in the art that these concepts may be practiced without these specific details.

[0045] Fig. 1 illustrates a system 100 representing selectively enciphering / deciphering a data stream, in accordance with an embodiment of the present disclosure.

[0046] According to an embodiment of the present disclosure, the system 100 may comprise a user device 102 and a server 104. In some embodiment, the system 100 may comprise more elements than the elements illustrated in FIG. 1, however the same are not explained for the sake of brevity. All the elements of the system 100 may communicate with each other via wireless connection, wired connection, or combination of both.

[0047] In an embodiment, the user device 102 may be, not limited to, a wearable device, a portable electronic device, and an embedded system. The portable electronic device may be, but not limited to, a smart phone, a tablet, a laptop, a personal computer. A virtual assistant (or voice assistant) (VA) 106 may be configured in the user device 102. For example, VA 106 may be an application, a software, and / or program codes executed by the user device 102 for obtaining an input prompt (or an input query) in natural language, from the user of the user device 102, processing the input prompt or the input query, and providing response in natural language to the user based on the input prompt. The user device 102 may be configured to receive a data stream indicating speech signal of a user. The data stream may be but not limited to user query comprising sensitive portion and non-sensitive portion. In an exemplary embodiment, the data stream may be voice stream including voice or speech command associated with the user.

[0048] The user device 102 configured to perform encryption of the sensitive portion present in the data stream and send the encrypted data stream to the server 104 along with non-encrypted non-sensitive portion. Further, upon receiving the data stream, the server 104 may determine intention of the content present in the data stream without decrypting the encrypted at least one sensitive portion. Further, the server 104 may generate the response for the received data stream based on the determined intention of the speech signal. Once the response is generated by the server 104, the response is sent to the user device 102. Further, the user device 102 decrypts the response to identify if there are any sensitive portions present in the decrypted response. If there are sensitive portions, then the user device 102 may modify the decrypted data based on the identified sensitive portion to generate the modified response.

[0049] The VA 106 configured in the user device 102 have become integral to everyday life, offering convenience and hands-free operation. For complex data interpretation, VA 106 processes user data at remote place (for example, at a server end or the server 106), wherein user data contains sensitive information too, which raises significant privacy and security concerns as the server 106 becomes known to sensitive information. Furthermore, the user device 102 may perform various user-friendly functionalities such as conversation data analysis to extract TO-DO list or summary covering key points of conversation, which require to use remote place (for example, the server end or the server 106) to perform the complex data interpretation. Conversation agents such as the VA 106 on the user device 102 try to give friendly functionalities while facing problem of not able to secure user's privacy while processing the complex data at the remote place.

[0050] While the VA 106 transmit user's data to the server 106, the VA 106 may selectively encrypt sensitive parts of data stream using dynamic keys with maintaining context / meaning of sensitive parts for natural language processing (NLP) and natural language generation (NLG) at server end without decryption. Though sensitive data is preserved via encryption before sending, its context is not persevered while used in NLG during conversation along with its meaning. Since the server 106 and / or other 3-rd party application programming interface (API) associated with the server 106 decrypt received (e.g., from the VA 106 of the user device 102) data stream before performing operation at server end to understand the context of the message, the data stream including sensitive parts are became known by the VA 106 and / or other APIs for NLP and NLG. Furthermore, sensitive information is disclosed to cloud services and servers. Also, NLG is not fully user and device context aware, and not content sensitivity aware. Accordingly, there is a need of real end to end encrypted channel in the domain of voice assistants that the users may embrace and trust.

[0051] In an embodiment in accordance with the present disclosure, the user device 102 (also referred to as a client or a client device) may find (or identify) one or more sensitive parts of text data with a sensitivity level value, generate dynamic keys using text data itself, apply homomorphic encryption on sensitive parts with complexities according sensitivity level, and send the result of the homomorphic encryption to server 106. The server 106 may execute NLU model to interpret the received data without decrypting the encrypted sensitive parts. The server 106 may perform NLG to prepare a response considering the user and device context. The server 106 may encrypt sensitive parts of response and send the response to the user device 102. The user device 102 may perform the background, display, and voice actions according the received response with decrypting the encrypted parts of response.

[0052] In an embodiment in accordance with the present disclosure, the user device 102 may selectively encrypt one or more sensitive data parts in a data stream, thereby generating a selectively encrypted data stream. The user device 102 may transmit selectively encrypted data stream the server 106. The server 106 may not require to decrypt the encrypted sensitive parts of the received data stream to perform any NLP / NLG operation. Thus, the sensitive data of the user may be protected.

[0053] According to an embodiment of the present disclosure, the user device 102 may perform an automated speech recognition (ASR). The user device 102 may capture and / or collect audio data being uttered by the user in cases of 'audio' type data needs to be processed. The user device 102 may analyze the user's context based on the captured audio data. For example, the user device 102 may analyze the user context considering user's point of view to data and / or presence of other people in surrounding of device. The user device 102 may identify a sensitivity level for each portion of input text data based on the user context and the input text data. Based on the sensitivity level of each portion, the user device 102 identifies the homomorphic complexity level for each portion, and prepares the encryption cipher. The encrypted data may be transmitted to the server 104 for complex NLP operations.

[0054] The server 104 may evaluates the cipher portions in received data and performs NLU on input data (e.g. the received data from the user device 102). Based on the NLU result, the server 104 performs NLG and prepares response to be shared to the user device 102. The response includes NLG response and one or more actions to be performed on the user device 102. The server 104 may transmit the response to the user device 102. The user device 102 decrypts encrypted portions of the response. The user device 102 performs action, modifies NLG according user context an sensitivity levels of portions of response and communicates to user based on the modified NLG.

[0055] Fig 2 illustrates a block diagram of the user device 120 for selectively enciphering a data stream, in accordance with various embodiments of the present disclosure.

[0056] In some implementations, the user device 201 comprises a memory 203, a processor 207, and an Input / output (I / O) Interface 209. The user device 120 may comprise a machine learning (ML) module 205 including feature extraction module 219, sensitive data identification module 221, an encryption module 223, and other modules 225. As an example, data 211 may be stored in memory 203. In one embodiment, the data 211 may include sensitive data 213, encryption sensitive data 215 and other data 217. In the illustrated figure. 2, machine learning (ML) module 205 are described herein in detail. In some embodiments, the user device 201 may exclude at least one of these components or may further comprise at least one other component. In an embodiment, the user device 102 of fig. 1 may be implemented in a similar manner to the user device 201.

[0057] The memory 203 may include a volatile and / or non-volatile memory. For example, the memory 203 may store commands, instructions, program codes, and / or data related to at least one other component of the user device 201. According to embodiments of this disclosure, the memory 203 may store software and / or a program. The program may include, for example, a kernel, middleware, an application programming interface (API), and / or an application program (or an application).

[0058] According to an embodiment of the present disclosure, the memory 203 may include one or more storage media. The storage medium may store one or more instructions. The one or more instructions may be executed by the processor 207 and / or other processor(s). When the one or more instructions are executed by the processor 207 and / or other processor(s) individually or collectively, the one or more instructions may cause the user device 201 to perform any combination of operations described herein.

[0059] The processor 207 includes one or more processing devices or processing circuitry, such as one or more microprocessors, microcontrollers, digital signal processors (DSPs), application specific integrated circuits (ASICs), or field programmable gate arrays (FPGAs). In some embodiments, the processor 207 includes one or more of a central processing unit (CPU), an application processor (AP), a communication processor (CP), a graphics processor unit (GPU), or a neural (network) processing unit (NPU). The processor 207 is able to perform control on at least one of the other components of the user device 201 and / or perform an operation or data processing relating to encryption and / or decryption or other functions. As described in more detail below, the processor 207 may perform various operations related to selectively encrypt an input query from the user, decrypt a natural language response from the server 104, and provide visual and / or spoken response to the user.

[0060] According to an embodiment of the present disclosure, the user device 201 may comprise a plurality of processors, including the processor 207. The plurality of processors may execute one or more commands, instructions, and / or program code, thereby causing the user device 201 to perform any combination of operations described herein.

[0061] The I / O interface 209 serves as an interface that may, for example, transfer commands or data input from a user or other external devices to other component(s) of the user device 201. The I / O interface 209 may also output commands or data received from other component(s) of the user device 201 to the user or the other external device. In an embodiment, the I / O interface 209 may comprise one or more input devices such as a keyboard, a touch screen, a sensor, and / or a microphone. The I / O interface 209 may comprise one or more output devices such as a display device and / or a speaker.

[0062] In some embodiments, the components of the electronic device may communicate with each other via a bus included in the user device 201. The bus may include a circuit for connecting the components included in the user device 201 with one another and for transferring communications (such as control messages and / or data) between the components.

[0063] In some embodiments, data 211 may be stored in the memory 203 in form of various data structures. Additionally, the data 211 can be organized using data models, such as relational or hierarchical data models. The other data 217 may store data, including temporary data and temporary files, generated by the ML module 205 for performing the various functions of the user device 201.

[0064] In some embodiments, the ML module 205 may include, for example feature extraction module 219, a sensitive data identification module 221, an encryption module 223, and other module 225. The other module 225 may be used to perform various miscellaneous functionalities of the user device 201 apart from the functionalities performed by the sensitive data identification module 221 and the encryption module 223. It will be appreciated that such aforementioned module may be represented as a single module or a combination of different module.

[0065] In an embodiment, when the user wishes to check any information such as meeting scheduled for the day, presentation received from his boss, and the like, the user provides a data stream which may include sensitive portion and non-sensitive portion. For example, the user may utter "check if I have received message from my manager" or may provide the recorded data stream in form of the query. Then the feature extraction module 219 may be configured to identify the sensitive portion present in the data stream. In particular, the feature extraction module 219 may be configured to convert the received data stream into text data.

[0066] FIG. 3 illustrates a block diagram of a user device for generating plurality of portions in the data stream using acoustic speech recognition process, in accordance with the exemplary mechanism.

[0067] Now moving towards Fig. 3 in order to understand the detailed functioning of the feature extraction module in order to generate the text data. In an embodiment, the feature extraction unit 301 of the feature extraction module 219 may be configured to receive the data stream (including the speech signal) and provide the data stream to the decoding unit 307. For example, the feature extraction unit 301 projects an input speech signal (e.g., the data stream) to recognize into a multi-dimension space (also referred to as a feature space). The extracted features may be based on the short-term Fourier transform (SFT) of the speech waveform. The extracted features model a general shape of a spectral envelope, and attempt to replicate some of the psycho-acoustic properties of the human auditory system. In an embodiment, the features may be extracted based on Mel-frequency cepstral coefficients (MFCCs) and / r perceptual linear prediction (PLP).

[0068] The decoding unit 307 may use the acoustic model 303 to determine relationship between the speech signal and the corresponding phonemes. The acoustic model 303 may be the pattern-classifier that may consider the speech signal and their transcriptions and then compile them into statistical representations of the sounds for words. The acoustic model 303 provides the likelihood of a set of acoustic vectors given a word sequence. In an embodiment, the acoustic model 303 may be implemented based on a hidden Markov model (HMM).

[0069] Once the relationship between the speech signal and the corresponding phonemes are determined, the decoding unit 307 may use the language model 305 to identify the context of the data stream in order to distinguish between the plurality of portions in the data stream and the portion that may sound similar using the determined relationship between the speech signal and the corresponding phonemes. For example, it is common phenomenon that two or more phrases may be pronounced almost the same but mean very different things. These ambiguities are easier to resolve when evidence from the language model 305 is incorporated with the acoustic model 307. In an embodiment the language model 305 may be implemented based on a n-gram based language model.

[0070] Upon understanding the context of the data stream, the decoding unit 307 may be configured to generate text data (e.g., 'output text' of Fig. 3). The decoding unit 307 combines the output of the acoustic model 303 and the output of the language model 305 to convert the extracted features into a sequence of words. The decoding unit 307 search for the most likely word sequences given the input features. In an embodiment, the decoding unit 307 may be implemented based on a Beam Search, an optimization technique used to explore multiple hypotheses at once, balancing accuracy and computational efficiency.

[0071] After determining the text data, the feature extraction module 219 may be configured to identify separators within the text data. Then the feature extraction module 219 may be configured to divide the text data into the plurality of text portions based on the identified text portions and the separators within the data stream. For example, the text portions may be divided based on the identified separators which may depend based on the context of the data stream.

[0072] Once the text data is divided into the plurality of text portions, feature extraction module 219 may be configured to provide the plurality of text portions to the sensitive data identification module 221. The sensitive data identification module 221 may be configured to identify at least one sensitive portion of the data stream. In particular, to identify the sensitive portion of the data stream, the sensitive data identification module 221 may determine the first sensitivity score for each of the plurality of portions. The sensitivity score for each of the plurality of portions is associated with the sensitivity characteristics of the respective portions.

[0073] For example, to generate the first sensitive score, the sensitive data identification module 221 may be configured to convert the plurality of portions in the data stream to a word vector dimension which is given as an input to a pretrained model such as bidirectional Long-Short Term Memory (LSTM) layer. The pretrained model may give the output which is considered as sensitivity score for each of the plurality of portions. For instance, when the data stream "open my pan card with pan number ABC123" which is further converted into vector representation using a pretrained model. During the training process of the model, the plurality of words and corresponding score may be stored. When the vector representation is fed as an input to the pretrained model which may process the plurality of portions and determine the score for plurality of portions as "open my" = 0.2, "pan card" = 7.2, "with"= 0.1, "pan number ABC123"= 9.8.

[0074] Further, the sensitive data identification module 221 may be configured to identify a second sensitivity score as explained in figures 4 and 5.

[0075] FIG 4. illustrates a pre-trained model for determining a second score associated with user context and user device context, in accordance with an embodiment of the present disclosure. FIG 5. illustrates an exemplary embodiment for determining a second score associated with user context and user device context by a pretrained model, in accordance with an embodiment of the present disclosure.

[0076] Now moving towards fig. 5 in order to understand the generation of second sensitivity score. The second sensitivity score may correspond to the context of the user context and the user device context. The user device context may correspond to historic data associated with the user. For example, the user may interact in various online platforms and access information from the internet, the user may share the information with his colleagues or officers. The information may include both sensitive information and non-sensitive information that may be shared across multiple platforms. The above-mentioned data is collected and segregated. For instance, the historical data associated with the user may include, but not limited to, data domains (e.g., 'Set 1' of Fig. 4), data usage patterns (e.g., 'Set 2' of Fig. 4), user behavior data (e.g., 'Set 3' of Fig. 4), data source reliability (e.g., 'Set 4' of Fig. 4), compliance requirement (e.g., 'Set 5' of Fig. 4), external threat patterns (e.g., 'Set 6' of Fig. 4) that may be determined using user context determining module 501 of the sensitive data identification module 221. The data domains correspond to all domains of data user device 102 possesses. The data usage patterns correspond to patterns of the user's interaction with different domains of data. The user behavior data corresponds to data associated with the user's patterns such as sharing or reporting. The data source reliability corresponds to source authenticity of data that the user device 102 possesses. The compliance requirement corresponds to guidelines that need to be followed with sensitive data. The external threat patterns correspond to continuous threat patterns for the sensitive data that the user possesses.

[0077] When the historical data is determined, the data preprocessing module 403 of the sensitive data identification module 221 may initially categorize and encode the historical data in an embedding friendly format. Further, continuous features like login, data accessed, sharing events, and reporting events of the embedding friendly format may be normalized. Finally, the data preprocessing 403 may be configured to perform reshaping of the normalized data which may be rearrangement of the data without altering the context of the data. This results in 3D tensor which may be a matrix representation that may be suitable as an input for pretrained model. Further, user context analyzer 405 (or user context analysis ML model)of the sensitive data identification module 319 may use for example convolution neural network based neural network to predict a threat level associated with the user by using the reshaped data using the factors like data domains, data usage patterns, user behavior data, data source reliability, compliance requirement, external threat patterns as shown in fig. 4. For instance, consider that the user may be associated with the financial sector which may contain sensitive information. The user's day-to-day activities may involve various interaction with the sensitive information and may perform his interaction in various websites, domains, and the like which may be prone to threat. Based on the patterns of the user with different domain data, user behavior and other factors may affect the score associated with the user (e.g., user context sensitivity score). For ease of understanding, consider that the user who works in an NGO may not deal with sensitive information in his / her daily activities and hence the score associated with the user may be low. In an example, if the user is associated with a financial sector, then the score associated with user may be high as the user may deal with various sensitive information such as cheque book number, loan, interest rate, transfer of amounts, and the like. Therefore, the historical information associated with the user is initially collected to check the threat associated with the user. Then the collected data is processed to determine the user context which may be done using a pretrained model which may be a convolution neural network (CNN). The pre-trained model may be used to predict the threat level associated with the user using the historical data of the user. The second sensitivity score associated with user context and user device context may range from 1 to 10 with intervals of 0.1, with 10 being highly sensitive and 1 being the lowest. For example: if the score is 2.0 then the user is less likely to do data theft. In another example, if the score is 9.5 then the user is highly prone to data theft. For example, the second score corresponds to user's threat level of data theft and serves as a parameter for data content sensitivity score identification.

[0078] For example, referring to a table illustrated in Fig. 5, the user 1 who is associated with the financial domain may have 5 logins in various websites. Further, the user may have used 150 MB of data to retrieve some information from the official browser or internet and has shared the 2 data events with his / her friends. Based on the above information e.g., when the user accesses or logs in multiple times to the official website and accesses the information which is further shared with higher officials or staffs, the data may be prone to data theft. Thus, the processor is configured to collect the above-mentioned data associated with the user which is then fed to the pre-trained machine learning (ML) model to process the information and calculate the threat score which may be high as there are multiple data sharing and large amounts of data is accessed. Similarly, the user who is associated with the health domain may have 3 logins in various websites. Further, the user may have used 80 MB of data and has shared the 1 data event with his / her friends and 0 reporting events. Based on the above information associated with the user, the pretrained model may process the information and calculates the threat score of the user is medium. When the above-mentioned information is processed where the second score may be determined i.e., 6.8 for the user 1 as shown in figure 5. To simplify, when the historical data of the user is collected, the data is processed by the data preprocessing module 501, where the data is rearranged, without altering the context of the data which can be fed as an input to the user context analyzer 503 which may process the data to identify the user context sensitivity score as shown in figure 5.

[0079] In an embodiment, the sensitive data identification module 221 may determine the sensitivity level for each of the plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context as described above. The first sensitivity score and the second sensitivity score may be used as an input to determine the sensitivity level as shown in figure 6.

[0080] FIG. 6 illustrates an exemplary embodiment for determining a sensitivity level for each of the plurality of portions, in accordance with an embodiment of the present disclosure.

[0081] Moving on with fig. 6 illustrating that the sensitivity level identifier 601 of the sensitive data identification module 221 may identify sensitivity level based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context. Further, the sensitivity level may be identified and compared with a predefined threshold sensitivity range. The predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range.

[0082] The sensitivity Level Identifier 601 may comprise a data preprocessing block 603, a bi-LSTM based neural network 605, and an output processor 607. The data preprocessing block 603 converts the spoken text into one or more numerical vectors form. For example, the data preprocessing block 603 may use word-to-vector (Word2Vec) encoding technique. The converted vectors may be given to the bi-LSTM based neural network 605. The bi-LSTM based neural network 605 may generate sensitivity scores. As the bi-LSTM based Neural network 605 gives respective sensitivity score to input words or numerical data, the output processor 607 determines the start and end of each sensitive content based on the sensitivity scores. The output processor 607 sensitivity level processing table 609 to calculate sensitivity level based on sensitivity scores of words and user context sensitivity score. The output processor 607 structures the output content in the required format.

[0083] For example, referring to the table 609, if the first score is in range 2 to 4 and the user context and user device context is in range 1 to 4, the average of both may be considered and based on the average score is compared with each of the high threshold range, a medium threshold range and a low threshold range. In an embodiment, the first score and the second score may be combined to perform weighted average which is a calculation that takes into account the varying degrees of importance of the numbers in a data set. Further, the predefined threshold may depend on the user. For instance, if the user is not related to the banking sector, then the predefined threshold may be set high and if the user is associated with the finance sector, then the predefined threshold may be set low in comparison with the user who is not prone to data theft. For ease of understanding, if the first score is in range 7 to 10 and the user context and user device context is in range 4 to 6 then the sensitivity level may be high based on the average score in comparison with at least one of high threshold range, a medium threshold range and a low threshold range as shown in fig. 6. After identifying the sensitivity level, the sensitive data identification module 221 may identify the portion as the sensitive portion, if the sensitivity level of the portion is within at least one of the high threshold ranges, the medium threshold range, or the low threshold range. In an embodiment, the sensitive data identification module 221 may identify the portion as the non-sensitive portion, if the sensitivity level of the portion is below the low threshold range. For example, consider the data stream to be "open my pan card with pan number ABC123", then the sensitive data identification module 221 may process the data stream (e.g., at block 603) based on the determined first score and the second score as shown below in table 1.

[0084]

[0085] From the above table 1, the text portion "pan card" and the "pan number ABC123" is considered as the sensitivity level is determined as medium and high and the position of the sensitive word in the data stream may start from the location 8 and may end at location 16 in the data stream.

[0086] FIG 7. illustrates an exemplary embodiment for an encrypting at least one sensitive portion of the data stream, in accordance with an embodiment of the present disclosure.

[0087] Now moving onto fig. 7 illustrations encryption of sensitive portions using the plurality of dynamic keys. Fig. 7 illustrates a block diagram of encryption engine 700 comprising homomorphic complexity identifier 700a, key generation module 701, encryption module 703. In an embodiment, the encrypted data may correspond to encrypting the sensitive portions present in the data stream by an encryption module 223. For example, once the sensitive portion and non-sensitive portion is identified, then the encryption module 703 of the encryption engine 700 may be configured to encrypt the sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. For example, a plurality of dynamic keys for at least one sensitive portion may be generated based on the non-sensitive portion present in the data stream by the key generation module 701. Then the encryption module 223 may encrypt the sensitive portion of the data stream using one of the generated plurality of dynamic keys. Further, the encryption module 223 may be configured to not encrypt the non-sensitive portion present in the data stream. After that, the encryption module 223 may be configured to transmit the data stream comprising the encrypted at least one sensitive portion and non-encrypted non-sensitive portion. For example, when the data stream is considered as "open my pan card with pan number ABC123", then the sensitive portion in the data stream is determined as "pan card" and the "pan number ABC123" as the sensitivity level of "pan card" and the "pan number ABC123" is determined to fall in the range of medium threshold range and high threshold range. To encrypt the sensitive portion "pan card" and the "pan number ABC123", the encryption module 223, 703 may generate the plurality of dynamic for the sensitive portion "pan card" and the "pan number ABC123" based on the content surrounding the at least one sensitive portion within the data stream which may include the plurality of portions like "open my", "with".

[0088] To simplify, the encryption engine 700 may comprise a homomorphic complexity identifier 700a, a key generation module 701 and an encryption module 703. When the encryption engine 700 receives the data stream with identified sensitive levels, then the data stream with sensitive portion is further encrypted using the key generation module 701 via the homomorphic complexity identifier 700a that may be configured to identify complexity level of the sensitive portion and the non-sensitive portions and generate a parameter based on the identified complexity level. Upon generating the parameter, the homomorphic complexity identifier 700a may provide the generated parameters for key generation module 701 for generating keys. For example, the homomorphic complexity identifier 700a allocates a complexity level (e.g., low, medium or high) to each of the sensitive portion and the non-sensitive portions, based on the sensitivity level and non-sensitive contents. The homomorphic complexity identifier 700a determines one or more prerequisite parameters required for key generation. For example, the one ore more parameters may include a cyclotomic polynomial ring R, a polynomial degree N, a noise standard deviation σ, and a cipher modulus q. The parameters may be used to generate one or more keys at the key generation module 701. Further, the key generation module 701 may generate the plurality of dynamic keys such as public key 701b, private key 701a, bootstrapping key 701d and re-linearization key 701c which is used by the encryption module 703 for encrypting the sensitive portion which is transmitted to the server 104.

[0089] In an embodiment, the key generation module 701 of the encryption engine 700 may generate a plurality of keys which may include but not limited to public key 701b, private key 701a, bootstrapping key 701d and re-linearization key 701c as shown in figure 7. Generally, the public key 701b is relatively known and the private key 701a which is kept secret. While the public key 701b is known and also can be retrieved easily and is for encoding or encryption, the private key 170a is employed for decoding or decryption. Further, the re-linearization key 701c may be defined as evaluation key used to manage increase in polynomial degree of encrypted portions resulting from multiplication operations. The re-linearization key 701c helps to reduce the degree of encrypted texts back to a manageable level. The bootstrapping key 701d is used to refresh a encrypted portion or the cipher text by reducing the accumulated noise in it, allowing further homomorphic operations. For example, the bootstrapping key 701d may be used to maintain the correctness of a cipher text over successive operations. The generated key may be given to the encryption module to encrypt the sensitive portion using the generated dynamic key as described above.

[0090] Fig. 7B illustrates an exemplary selectively encrypted query, in accordance with an embodiment of the present disclosure.

[0091] An exemplary data stream comprising the encrypted sensitive portion. For example, consider the data stream is "Can you check if I got any mail from Narendra Modi as I am supposed to receive one regarding housing Project" (e.g., an input query 707). When the data stream is received, the sensitive portion is identified as per the above steps explained above and the sensitive portion may be encrypted using the non-sensitive portion or the content surrounding the sensitive portion of the data stream. In the above example, "Narendra Modi" and the "Housing project" of the input query 707 may be identified as the sensitive portions and the remaining texts in the data stream is considered as the non-sensitive portions. Thus, the encryption module 703 may check the complexity level of the sensitive portions of the data stream and allocate one or more key generation parameters (e.g., a polynomial degree, a cipher modulus, a plaintext modulus, a noise standard deviation, and / or a security level) for each sensitive portion. the encryption module 703 may generate public key (e.g, 'hpk' and 'mpk' of the selectively encrypted query 709), private key, bootstrapping key (e.g, 'hbtk' and 'mbtk' of the selectively encrypted query 709) and / or re-linearization key (e.g, 'hrlk' and 'mrlk' of the selectively encrypted query 709) for the sensitive portions "Narendra Modi" and the "Housing project", based on the allocated key generation parameters. Once each of the above-mentioned keys are generated, the encryption module may encrypt the sensitive portion of the data stream using one of the generated plurality of dynamic keys to transmit the data stream comprising the encrypted at least one sensitive portion and the encrypted portion may be reflected as the selectively encrypted query 709 illustrated in Fig. 7B.

[0092] For instance, the generation of plurality keys may be as follows. For examples the private key generation may be as follows:

[0093] ㆍ Select distribution type with small coefficients, e.g., binary, ternary, etc.

[0094] ㆍ Sample a secret polynomials(X) of degree N from the distribution which also belongs to the cyclotomic polynomial ring R.

[0095] Further, the public key may be generated as below however this should not be construed as the limitation.

[0096] ㆍ Sample a random polynomiala(X) from a uniform distribution

[0097] ㆍ Sample an error polynomiale(X) from a Gaussian distribution over R with std. deviation σ

[0098] ㆍ Compute polynomialb(X) = - (a(X) .s(X) +e(X) ) mod q

[0099] ㆍ The public key is thenpk= (a(X) ,b(X) )

[0100] Furthermore, the key generation module may generate re-linearization key which may be as per the steps below:

[0101] ㆍ Computes2(X)

[0102] ㆍ Sample new random & error polynomiala'(X) &e'(X)

[0103] ㆍ Computeb'(X) = - (a'(X) .s(X) +e'(X) -s2(X) . q )

[0104] ㆍ Then the relinearization key isrlk= (a'(X) ,b'(X) )

[0105] The key generation module may generate bootstrapping key which may be as per below steps:

[0106] ㆍ Sample new random & error polynomiala'(X) &e'(X)

[0107] ㆍ Computeb''(X) = - (a''(X) .s(X) +e''(X) -s(X) . q )

[0108] ㆍ Then the bootstrapping key isbtk= (a''(X) ,b''(X) ).

[0109] FIG 8. illustrates block diagram of a server 104 for generating a response for enciphered data stream, in accordance with an embodiment of the present disclosure.

[0110] In an embodiment, when the sensitive portion is encrypted based on the content surrounding the at least one sensitive portion within the data stream, then the encrypted sensitive portion along with non-encrypted non-sensitive portion may be transmitted to a system which is the server 800 comprising a processor 801 (also referred to as 'sensitive command processor) and memory 800c coupled to the processor 801. In some embodiments, the server 800 may exclude at least one of these components or may further comprise at least one other component. In an embodiment, the server 104 of fig. 1 may be implemented in a similar manner to the server 800.

[0111] The processor 801 includes one or more processing devices or processing circuitry, such as one or more microprocessors, microcontrollers, DSPs, ASICs, or FPGAs. In some embodiments, the processor 801 includes one or more of a CPU, an AP, CP, a GPU and / or an NPU. The processor 801 is able to perform control on at least one of the other components of the server 104 and / or perform an operation or data processing relating to natural language processing or other functions. As described in more detail below, the processor 801 may perform various operations related to generating a natural language response based on a selectively encrypted query from the user device 102 and / or a homomorphic cipher engine 800n.

[0112] According to an embodiment of the present disclosure, the server 104 may comprise a plurality of processors, including the processor 801. The plurality of processors may execute one or more commands, instructions, and / or program code, thereby causing the server 104 to perform any combination of operations described herein.

[0113] The memory 800c may include a volatile and / or non-volatile memory. For example, the memory 800c may store commands, instructions, program codes, and / or data related to at least one other component of the server 104. According to embodiments of this disclosure, the memory 800c may store software and / or a program. The program may include, for example, a kernel, middleware, an API, and / or an application program (or an application).

[0114] According to an embodiment of the present disclosure, the memory 800c may include one or more storage media. The storage medium may store one or more instructions. The one or more instructions may be executed by the processor 801 and / or other processor(s). When the one or more instructions are executed by the processor 801 and / or other processor(s) individually or collectively, the one or more instructions may cause the server 104 to perform any combination of operations described herein.

[0115] The processor 801 is configured to apply (or execute) a pre-trained machine learning (ML) model 800b as shown in fig. 8 in order to determine the intention of the speech signal based on the data stream without decrypting the encrypted at least one sensitive portion. For example, the processor 801 may determine the context of an intention of the at least one non-encrypted non-sensitive portion. The processor 801 may determine a co-relation between at least one non-encrypted non-sensitive portion and the encrypted sensitive portion based on the determined context of the non-encrypted non-sensitive portion and prestored data comprising predefined correlation between a plurality of non-encrypted non-sensitive portions and a plurality of encrypted sensitive portions. In a non-limiting example of the present disclosure, the data stream of "Make a call to &%*#%&# via WhatsApp" where the sensitive portion is "9587586487" which is encrypted as "&%*#&%" in the received data stream. For example, in determination of the correlation between the non-encrypted non-sensitive portion and the encrypted sensitive portion of the data stream, the processor 801 may be configured to analyze the context or intent associated with the non-encrypted non-sensitive portion. Based on this determined context, the processor 801 may be configured to compare the determined context with prestored data containing predefined correlations, which may map relationships between various non-sensitive portions and sensitive portions. By comparing the context of the non-encrypted non-sensitive portion with these predefined correlations, the processor 801 may be configured to identify a likely correlation between the non-sensitive non-encrypted portions and the encrypted sensitive portions.

[0116] To simplify, when the server 104 may receive the encrypted sensitive information and non-encrypted non-sensitive information and determine the intention of the speech signal, the Natural Language Understanding (NLU) engine which is associated with the server may generate the response for the received data stream based on the determined intention of the speech signal. Specifically, when the data stream comprises encrypted at least one sensitive portion and non-encrypted non-sensitive portion. The data stream corresponding to a speech signal of a user is provided to the NLU i.e., for fine tuning of the data. The fine-tuning process is used to remove the noise present in the data stream with encrypted sensitive portion and may provide the filtered data stream to further process and generate the response.

[0117] In an embodiment, the NLU results are provided to a Natural Language generation (NLG) engine associated with the server 104 to generate a response and display the response to the user. In an embodiment a response engine associated with the NLU of the server may generate the response to the query of the user. Once the response is generated at server end, the response may be sent to the user device.

[0118] In particular, when the above-mentioned data stream is sent to the server 104, the server 104 may initially understand the semantics of the data stream (including encrypted sensitive portion and non-encrypted non-sensitive portion). A Natural language understanding (NLU) may be a specialized unit which is trained on a dataset that emphasizes the contextual relationships between plain and cipher text, enabling it to grasp the overall intent of the data stream may be used. The cipher text may be encrypted sensitive portion of the data stream, and the plain text may be non-sensitive portion or the content surrounding the sensitive portion in the data stream. When the data stream is received, the NLU may retain the cipher text in its designated position, ensuring that it remains unaltered, while generating a coherent action output that aligns with the overall meaning of the data stream as shown in fig. 9 to further generate a natural language response to the user.

[0119] Fig. 9 illustrates a block diagram of the server 104 for selectively deciphering a data stream, in accordance with various embodiments of the present disclosure.

[0120] In some implementations, the server 901 comprises a memory 903, a processor 907, an Input / output (I / O) Interface 909. The server 901 may comprise an ML model 905 including sensitive data identification module 921, a decryption module 923, a data modification module 925 and other module 927. The memory may further include data 911 and ML module 905. As an example, data 911 is stored in the memory 903. In one embodiment, the data 911 may include sensitive data 913, decryption sensitive data 915 and other data 917. In the illustrated fig. 9, ML module 905 are described herein in detail. In some embodiments, the server 901 may exclude at least one of these components or may further comprise at least one other component. In some embodiments, the server 901 may exclude at least one of these components or may further comprise at least one other component. In an embodiment, the server 104 of fig. 1 may be implemented in a similar manner to the server 901.

[0121] The memory 903 may include a volatile and / or non-volatile memory. For example, the memory 903 may store commands, instructions, program codes, and / or data related to at least one other component of the server 901. According to embodiments of this disclosure, the memory 903 may store software and / or a program. The program may include, for example, a kernel, middleware, an API, and / or an application program (or an application).

[0122] According to an embodiment of the present disclosure, the memory 903 may include one or more storage media. The storage medium may store one or more instructions. The one or more instructions may be executed by the processor 907 and / or other processor(s). When the one or more instructions are executed by the processor 907 and / or other processor(s) individually or collectively, the one or more instructions may cause the server 901 to perform any combination of operations described herein.

[0123] The processor 907 includes one or more processing devices or processing circuitry, such as one or more microprocessors, microcontrollers, DSPs, ASICs, or FPGAs. In some embodiments, the processor 907 includes one or more of a CPU, an AP, CP, a GPU and / or an NPU. The processor 907 is able to perform control on at least one of the other components of the server 901 and / or perform an operation or data processing relating to natural language processing or other functions. As described in more detail below, the processor 907 may perform various operations related to generating a natural language response based on a selectively encrypted query from the user device 102.

[0124] According to an embodiment of the present disclosure, the server 901 may comprise a plurality of processors, including the processor 907. The plurality of processors may execute one or more commands, instructions, and / or program code, thereby causing the server 901 to perform any combination of operations described herein.

[0125] The I / O interface 909 serves as an interface that may, for example, transfer commands or data input from a user or other external devices to other component(s) of the server 901. The I / O interface 909 may also output commands or data received from other component(s) of the server 901 to the user or the other external device. In an embodiment, the I / O interface 909 may comprise one or more input devices such as a keyboard, a touch screen, a sensor, and / or a microphone. The I / O interface 909 may comprise one or more output devices such as a display device and / or a speaker.

[0126] In some embodiments, the components of the electronic device may communicate with each other via a bus included in the server 901. The bus may include a circuit for connecting the components included in the server 901 with one another and for transferring communications (such as control messages and / or data) between the components.

[0127] In some embodiments, data 911 may be stored in the memory 903 in form of various data structures. Additionally, the data can be organized using data models, such as relational or hierarchical data models. The other data 917 may store data, including temporary data and temporary files, generated by the ML module 905 for performing the various functions of the system 901.

[0128] In some embodiments, the module 905 may include, for example sensitive data identification module 921, a decryption module 923, data modification module 925 and other module 927. The other module 927 may be used to perform various miscellaneous functionalities of the system 901 apart from the functionalities performed by the sensitive data identification module 921 and the decryption module 923. It will be appreciated that such aforementioned module may be represented as a single module or a combination of different module.

[0129] Once the server 800 transmits the response to the system i.e., user device, the system disclosed in the figure 10 may perform the decryption of response to identify at least one sensitive portion in the response. The response comprises encrypted at least one sensitive portion and non-encrypted non-sensitive portion. Further, the sensitive data identification module 921 may be configured to identify if there are any sensitive portions present in the response transmitted from the server. If there are no sensitive portions, then the processor associated with the ML model is configured to transmit the response to the user. However, if there are any sensitive portions present in the response transmitted from the server, then the data modification module 925 may modify the response i.e., modification may be performed on the decrypted data stream based on the identified at least one sensitive portion. Once the modification is performed, then the data modification module 925 may generate the modified response in the form of a modified data stream and may be sent to the user device.

[0130] Fig. 10 illustrates an exemplary embodiment for selectively deciphering a data stream received from the server 104, in accordance with the present disclosure.

[0131] Now moving towards fig. 10 that illustrates an exemplary scenario of selectively deciphering a data stream. The user device 102 of fig. 10 may be implemented similar to the user device 201. The server 104 of fig. 10 may be implemented similarly to the server 800 or the server 901.

[0132] When the NLG generates a response, the response along with the sensitive information is provided to server 104. The response along with the encrypted sensitive portion is further provided to the decrypt engine using the decryption module 923 as shown fig. 10 to check if there is any sensitive information present in the response. The decrypt engine may decrypt the sensitive information in the response with the private key generated and send the decrypted response to the action executor. The action executor performs one or more actions per the specified action(s) in the response for input command data.

[0133] In an embodiment, the response from the server 104 may contain a speech natural language (NL) portion, a display NL portion, and one or more actions. The response may include one or more selectively encrypted parts. The decrypt engine decrypts sensitive portions of response data. One or more of the speech NL, the display NL, and / or one or more actions may contain some portions encrypted which was sent by the user device 102 to the server 104 in data sent to the server 104. For example, the encrypted sensitive portions sent by the user device 102 were used to generate the response without being decrypted at the server 104, and then included in the response as is, being sent back to the user device 102 from the server 104.

[0134] Further, the decrypted response may be analyzed by the sensitive NL modifier to check the sensitive portion in the data stream. Using user context, the sensitive NL modifier performs sensitive portion identification in the NL (e.g., the speech NL and / or the display NL) and smartly removes or replaces them with non-sensitive words. To identify sensitive portions, the sensitive data identification module may determine the first sensitivity score for each of the plurality of portions. The sensitivity score for each of the plurality of portions is associated with the sensitivity characteristics of the respective portions. For ease of understanding, consider the data stream corresponding to the speech signal of the user may be "open my pan card with pan number ABC123". When the sensitive data identification module receives the data stream then the sensitive data identification module may identify the sensitive portion in the data stream "open my pan card with pan number ABC123". Further, to generate the first sensitive score, the sensitive data determination module 921 may convert the plurality of portions in the data stream to a word vector dimension which is given as an input a pretrained model such as bidirectional LSTM layer. The pretrained model may give the output which is considered as sensitivity score for each of the plurality of portions in the data stream. Further, the sensitive data identification module 921 may identify the second sensitivity score which may correspond to the context of the data stream and the user device context to identify the sensitivity level in the data stream comprising the response. The user device context may correspond to historic data associated with the user. For instance, consider that the user may be associated with the financial sector which may contain sensitive information. The user's day-to-day activities may involve various interaction with the sensitive information and may perform his interaction in various websites, domains, and the like which may be prone to threat. Based on the patterns of the user with different domain data, user behavior and other factors may affect the score associated with the user. For ease of understanding, consider that the user who works in an NGO may not deal with sensitive information in his / her daily activities and hence the score associated with the user may be low. Then the collected data is processed to determine the user context which may be done using a pretrained model which may be a convolution neural network (CNN). The pretrained model may be used to predict the threat level associated with the user using the historical data of the user. The second sensitivity score associated with user context and user device context may range from 1 to 10 with intervals of 0.1, with 10 being highly sensitive and 1 being the lowest. For example: if the score is 2.0 then the user is less likely to data theft. In another example, if the score is 9.5 then the user is highly prone to data theft. Based on the sensitivity level and the sensitive portion present in the decrypted data stream, the sensitive data identification unit may modify the decrypted data stream based on the identified at least one sensitive portion to generate a modified data stream using a pretrained model wherein the processor of the client is configured to apply a pre-trained machine learning (ML) model to perform modification of the data stream.

[0135] The modified speech NL may be provided to a speech NL executor. The speech NL executor may provide a spoken response to the user based on the speech NL. For example, the speech NL executor may speak out the speech NL using text-to-speech (TTS) application and / or functionality configured on the user device 102. The modified display NL may be provided to a display NL executor. The display NL executor may provide a visual response to the user based on the display NL. For example, the display NL executor may display the display NL on the specified place on a display device included in or connected to the user device 102.

[0136] FIG 11. illustrates a method 1100 for selectively enciphering a data stream,　in accordance with an embodiment of the present disclosure.

[0137] Although example method 1100 depicts a particular sequence of operations, the sequence may be altered without departing from the scope of the present disclosure. For example, some of the operations depicted may be performed in parallel or in a different sequence that does not materially affect the function of method 1100.

[0138] According to some examples, the at block 1101 method 1100 includes identifying, by a processor, at least one sensitive portion of the data stream. Initially, the data stream corresponding to a speech signal of a user may be received. Further, the plurality of portions of the received data stream may be identified using NLP. The plurality of portions includes at least one sensitive portion and at least one non-sensitive portion. Based on the identified plurality of portions, the processor may determine a first sensitivity score for each of the plurality of portions. The sensitivity score for each of the plurality of portions is associated with the sensitivity characteristics of the respective portions. The processor may identify a sensitivity level for each of the plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context and identify at least one sensitive portion and the at least one non-sensitive portion based on the determined sensitivity level. Particularly, the processor may compare the sensitivity level of each of the plurality of portions with a predefined threshold sensitivity range. The predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range. The processor may further identify the portion as the sensitive portion, if the sensitivity level of the portion is within at least one of the high threshold ranges, the medium threshold range, or the low threshold range. Alternatively or additionally, the processor may identify the portion as the non-sensitive portion, if the sensitivity level of the portion is below the low threshold range.

[0139] According to some examples, at block 1103, method 1100 includes encrypting, by the processor, the at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. A plurality of dynamic keys may be generated for at least one sensitive portion based on the content surrounding at least one sensitive portion within the data stream. Based on the generated dynamic keys, the processor may encrypt one sensitive portion of the data stream

[0140] FIG 12. illustrates a method 1200 for selectively deciphering a data stream,　in accordance with an embodiment of the present disclosure.

[0141] Although example method 1200 depicts a particular sequence of operations, the sequence may be altered without departing from the scope of the present disclosure. For example, some of the operations depicted may be performed in parallel or in a different sequence that does not materially affect the function of method 1200.

[0142] According to some examples, the at block 1201 method 1200 includes decrypting the data stream to identify at least one sensitive portion in the decrypted data stream. The data stream comprises encrypted at least one sensitive portion.

[0143] According to some examples, the at block 1203 method 1200 includes modifying the decrypted data stream based on the identified at least one sensitive portion to generate a modified data stream.

[0144] FIG 13. illustrates a method 1300 for selectively deciphering a data stream,　in accordance with an embodiment of the present disclosure.

[0145] Although example method 1300 depicts a particular sequence of operations, the sequence may be altered without departing from the scope of the present disclosure. For example, some of the operations depicted may be performed in parallel or in a different sequence that does not materially affect the function of method 1300.

[0146] According to some examples, the at block 1301 method 1300 includes receiving the data stream comprising encrypted at least one sensitive portion, wherein the data stream corresponding to a speech signal of a user.

[0147] According to some examples, the at block 1303 method 1300 includes determining an intention of the speech signal based on the data stream without decrypting the encrypted at least one sensitive portion. Particularly, determining the intention of the speech signal the method comprises determining context of an intention of the at least one non-encrypted non sensitive portion. Further, the method comprises determining a co-relation between the at least one non-encrypted non-sensitive portion and the encrypted at least one sensitive portion based on the determined context of the at least one non-encrypted non sensitive portion and prestored data comprising predefined co-relation between a plurality of non-encrypted non-sensitive portions and a plurality of encrypted sensitive portions and determining the intention of the speech signal based on the determined co-relation.

[0148] According to some examples, the at block 1305 method 1300 includes generating the response for the received data stream based on the determined intention of the speech signal.

[0149] In a non-limiting embodiment of the present disclosure, a method for selectively enciphering a data stream is provided. The method comprises obtaining a data stream corresponding to a speech signal of a user. The method comprises identifying at least one sensitive portion of the data stream. The method further comprises encrypting at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. The method comprises generating a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion. The method may be performed by a user device.

[0150] In a non-limiting embodiment of the present disclosure, a first plurality of portions of the received data stream are identified using Natural language processing (NLP). The first plurality of portions includes at least one sensitive portion and at least one non-sensitive portion. A first sensitivity score for each of the first plurality of portions is determined. The sensitivity score for each of the first plurality of portions is associated with sensitivity characteristics of the respective portions. The method comprises determining a sensitivity level for each of the first plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context. The method comprises identifying the at least one sensitive portion and the at least one non-sensitive portion based on the determined sensitivity level.

[0151] In a non-limiting embodiment of the present disclosure, identifying the at least one sensitive portion and the at least one non-sensitive portion further comprises comparing a sensitivity level of a first portion of the plurality of portions with a predefined threshold sensitivity range. The predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range. The first portion is identified as a sensitive portion in response to the sensitivity level of the first portion being within at least one of the high threshold ranges, the medium threshold range, or the low threshold range. The first portion is identified as non-sensitive in response to the sensitivity level of the first portion being below the low threshold range.

[0152] In a non-limiting embodiment of the present disclosure, the method comprises generating a plurality of dynamic keys for the at least one sensitive portion based on the content surrounding the at least one sensitive portion within the data stream. At least one sensitive portion of the data is encrypted using one of the generated plurality of dynamic keys.

[0153] In a non-limiting embodiment of the present disclosure, the method comprises transmitting, to a server, the selectively encrypted data stream. The method comprises receiving, from the server, a response data stream comprising the at least one encrypted sensitive portion.

[0154] In a non-limiting embodiment of the present disclosure, the method comprises decrypting the at least one encrypted sensitive portion included in the response data stream. The method comprises modifying the decrypted response data stream based on the at least one sensitive portion to generate a modified data stream.

[0155] In a non-limiting embodiment of the present disclosure, the method comprises identifying a second plurality of portions of the decrypted response data stream using Natural language processing (NLP), wherein the second plurality of portions includes the at least one decrypted sensitive portion and at least one non-sensitive portion. The method comprises determining a third sensitivity score for each of the plurality of portions of the decrypted response data stream, wherein the first sensitivity score for each of the plurality of portions of the decrypted response data stream is associated with sensitivity characteristics of the respective portion. The method comprises determining a sensitivity level for each of the second plurality of portions based on the determined third sensitivity score and a fourth sensitivity score associated with user context and user device context. The method comprises identifying the at least one sensitive portion and at least one non-sensitive portion from the second plurality of portions based on the determined sensitivity level for each of the second plurality of portions.

[0156] In a non-limiting embodiment of the present disclosure, the method comprises comparing the sensitivity level of each of the second plurality of portions with a predefined threshold sensitivity range, wherein the predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range. The method comprises, in response to the sensitivity level of at least one portion of the second plurality of portions being within at least one of the high threshold range, the medium threshold range, or the low threshold range, modifying the at least one portion.

[0157] In a non-limiting embodiment of the present disclosure, a computer-readable storage medium storing one or more instructions is provided. The one or more instructions, when executed by at least one processor individually or collectively, cause the at least one processor to perform any one of the methods in accordance with the present disclosure.

[0158] In a non-limiting embodiment of the present disclosure, a user device for selectively enciphering a data stream is provided. The user device comprises at least one processor comprising processing circuitry. The user device comprises memory comprising one more storage media storing one or more instructions. In an embodiment, when executed by the at least one processor individually or collectively, the one or more instructions cause the user device to perform any combination of operations performed by the user device 102 and / or the user device 201 in accordance with the present disclosure.

[0159] For example, when executed by the at least one processor individually or collectively, the one or more instructions cause the user device to obtain a data stream corresponding to a speech signal of a user. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device identify at least one sensitive portion of the data stream. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device to encrypt the at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream. When executed by the at least one processor individually or collectively, the one or more instructions cause the user device to generate a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion.

[0160] In a non-limiting embodiment of the present disclosure, a server for generating a natural language response based on a selectively encrypted data stream is provided. The server comprises at least one processor comprising processing circuitry. The server comprises memory comprising one more storage media storing one or more instructions. In an embodiment, when executed by the at least one processor individually or collectively, the one or more instructions cause the server to perform any combination of operations performed by the server 104, the server 800, and / or the server 901 in accordance with the present disclosure.

[0161] While　various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art.　The various aspects and embodiments disclosed herein are for purposes of illustration and are not intended to be limiting, with the true scope and spirit being indicated by the detailed description.

[0162] The order in which the various operations of the methods are described is not intended to be construed as a limitation, and any number of the method described blocks can be combined in any order to implement the method. Additionally, individual blocks may be deleted from the methods without departing from the spirit and scope of the subject matter described herein. Furthermore, the methods can be implemented in any suitable hardware, software, firmware, or combination thereof.

[0163] It may be noted here that the subject matter of some or all embodiments described with reference to Figs. 1-13 may be relevant for the methods and the same is not repeated for the sake of brevity.

[0164] The various operations of methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and / or software component(s) and / or module(s), including, but not limited to a circuit, an application specific integrated circuit (ASIC), or processor. Generally, where there are operations illustrated in Figures, those operations may be performed by any suitable corresponding counterpart means-plus-function components.

[0165] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term "computer-readable medium" should be understood to include tangible items and exclude carrier waves and transient signals, i.e., non-transitory. Examples include Random Access Memory (RAM), Read-Only Memory (ROM), volatile memory, nonvolatile memory, hard drives, Compact Disc (CD) ROMs, Digital Video Disc (DVDs), flash drives, disks, and any other known physical storage media.

[0166] Certain aspects may comprise a computer program for performing the operations presented herein. For example, such a computer program product may comprise a computer readable media having instructions stored (and / or encoded) thereon, the instructions being executable by one or more processors to perform the operations described herein. For certain aspects, the computer program product may include packaging material.

[0167] Various components, module, or units are described in this disclosure to emphasize functional aspects of devices configured to perform the disclosed techniques, but do not necessarily require realization by different hardware units. Rather, as described above, various units may be combined in a hardware unit or provided by a collection of interoperative hardware units, including one or more processors as described above, in conjunction with suitable software and / or firmware.

[0168] As used herein, a phrase referring to "at least one" or "one or more" of a list of items refers to any combination of those items, including single members. As an example, "at least one of: a, b, or c" is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c. The terms "a", "an" and "the" mean "one or more", unless expressly specified otherwise. The terms "including", "comprising", "having" and variations thereof, when used in a claim, is used in a non-exclusive sense that is not intended to exclude the presence of other elements or steps in a claimed structure or method, unless expressly specified otherwise.

[0169] Finally, the language used in the specification has been principally selected for readability and instructional purposes, and it may not have been selected to delineate or circumscribe the inventive subject matter. It is therefore intended that the scope of the disclosure be limited not by this detailed description, but rather by any claims that issue on an application based here on. Accordingly, the embodiments of the present disclosure are intended to be illustrative, but not limiting, of the scope of the disclosure, which is set forth in the appended claims.

Claims

1.A method for selectively enciphering a data stream, comprising:obtaining, by a user device (102), a data stream corresponding to a speech signal of a user;identifying, by the user device (102), at least one sensitive portion of the data stream ;encrypting, by the user device (102), the at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream; andgenerating a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion.2.The method as claimed in claim 1, wherein identifying the at least one sensitive portion of the data stream further comprises:identifying, by the user device (102), a first plurality of portions of the data stream using Natural language processing (NLP), wherein the first plurality of portions includes the at least one sensitive portion and at least one non-sensitive portion;determining, by the user device (102), a first sensitivity score for each of the first plurality of portions, wherein the first sensitivity score for each of the first plurality of portions is associated with sensitivity characteristics of the respective portion;determining, by the user device (102), a sensitivity level for each of the first plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context; andidentifying, by the user device (102), the at least one sensitive portion and the at least one non-sensitive portion based on the determined sensitivity level.3.The method as claimed in claim 2, wherein identifying the at least one sensitive portion and the at least one non-sensitive portion further comprises:comparing, by the user device (102), a sensitivity level of a first portion of the first plurality of portions with a predefined threshold sensitivity range, wherein the predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range;identifying, by the user device (102), the first portion as a sensitive portion, in response to the sensitivity level of the first portion being within at least one of the high threshold range, the medium threshold range, or the low threshold range; andidentifying, by the user device (102), the first portion as a non-sensitive portion, in response to the sensitivity level of the first portion being below the low threshold range.4.The method as claimed in any one of claims 1 to 3, further comprising:generating, by the user device (102), a plurality of dynamic keys for the at least one sensitive portion based on the content surrounding the at least one sensitive portion within the data stream; andencrypting, by the user device (102), the at least one sensitive portion of the data stream using one of the generated plurality of dynamic keys.5.The method as claimed in any one of claims 1 to 4, further comprising:transmitting, to a server (104), the selectively encrypted data stream; andreceiving, from the server (104), a response data stream comprising the at least one encrypted sensitive portion.6.The method as claimed in 5, further comprising:decrypting, by the user device (102), the at least one encrypted sensitive portion included in the response data stream; andmodifying, by the user device (102), the decrypted response data stream based on the at least one sensitive portion to generate a modified data stream.7.The method as claimed in claim 6, wherein modifying the decrypted response data stream further comprises:identifying, by the user device (102), a second plurality of portions of the decrypted response data stream using Natural language processing (NLP), wherein the second plurality of portions includes the at least one decrypted sensitive portion and at least one non-sensitive portion;determining, by the user device (102), a third sensitivity score for each of the plurality of portions of the decrypted response data stream, wherein the first sensitivity score for each of the plurality of portions of the decrypted response data stream is associated with sensitivity characteristics of the respective portion;determining, by the user device (102), a sensitivity level for each of the second plurality of portions based on the determined third sensitivity score and a fourth sensitivity score associated with user context and user device context; andidentifying, by the user device (102), the at least one sensitive portion and at least one non-sensitive portion from the second plurality of portions based on the determined sensitivity level for each of the second plurality of portions.8.The method as claimed in claim 7,wherein modifying the decrypted response data stream further comprises:comparing, by the user device (102), the sensitivity level of each of the second plurality of portions with a predefined threshold sensitivity range, wherein the predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range; andin response to the sensitivity level of at least one portion of the second plurality of portions being within at least one of the high threshold range, the medium threshold range, or the low threshold range, modifying the at least one portion.9.A computer-readable storage medium storing one or more instructions, wherein the one or more instructions, when executed by at least one processor individually or collectively, cause the at least one processor to perform the method of any one of claims 1 to 8.10.A user device (102) for selectively enciphering a data stream, comprising:at least one processor (207) comprising processing circuitry; andmemory (203) comprising one more storage media storing one or more instructions,wherein, when executed by the at least one processor individually or collectively, the one or more instructions cause the user device (102) to:obtain a data stream corresponding to a speech signal of a user;identify at least one sensitive portion of the data stream;encrypt the at least one sensitive portion of the data stream based on a content surrounding the at least one sensitive portion within the data stream; andgenerate a selectively encrypted data stream by replacing the at least one sensitive portion within the data stream with the at least one encrypted sensitive portion.11.The user device as claimed in claim 10, wherein, when executed by the at least one processor individually or collectively, the one or more instructions further cause the user device (102) to:identify a first plurality of portions of the data stream using Natural language processing (NLP), wherein the first plurality of portions includes the at least one sensitive portion and at least one non-sensitive portion;determine a first sensitivity score for each of the first plurality of portions, wherein the first sensitivity score for each of the first plurality of portions is associated with sensitivity characteristics of the respective portion;determine a sensitivity level for each of the first plurality of portions based on the determined first sensitivity score and a second sensitivity score associated with user context and user device context; andidentify the at least one sensitive portion and the at least one non-sensitive portion based on the determined sensitivity level.12.The user device as claimed in claim 11, wherein, when executed by the at least one processor individually or collectively, the one or more instructions further cause the user device (102) to:compare a sensitivity level of a first portion of the first plurality of portions with a predefined threshold sensitivity range, wherein the predefined threshold sensitivity range comprises at least one of a high threshold range, a medium threshold range and a low threshold range;identify the first portion as a sensitive portion, in response to the sensitivity level of the first portion being within at least one of the high threshold range, the medium threshold range, or the low threshold range; andidentify the first portion as a non-sensitive portion, in response to the sensitivity level of the first portion being below the low threshold range.13.The user device as claimed in any one of claims 10 to 12, wherein, when executed by the at least one processor individually or collectively, the one or more instructions further cause the user device (102) to:generate a plurality of dynamic keys for the at least one sensitive portion based on the content surrounding the at least one sensitive portion within the data stream; andencrypt the at least one sensitive portion of the data stream using one of the generated plurality of dynamic keys.14.The user device as claimed in any one of claims 10 to 13, wherein, when executed by the at least one processor individually or collectively, the one or more instructions further cause the user device (102) to:transmit, to a server (104), the selectively encrypted data stream; andreceive, from the server (104), a response data stream comprising the at least one encrypted sensitive portion.15.The user device as claimed in claim 14, wherein, when executed by the at least one processor individually or collectively, the one or more instructions further cause the user device (102) to:decrypt the at least one encrypted sensitive portion included in the response data stream; andmodify the decrypted response data stream based on the at least one sensitive portion to generate a modified data stream.

Images