Text processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the field of artificial intelligence and the field of financial science and technology, and relates to a text processing method and device based on artificial intelligence, computer equipment and a storage medium. Performing coding processing on the initial text based on the target coding type to obtain first coding data; performing language recognition processing on the initial text to obtain corresponding language information; encoding the language information based on the target encoding type to obtain second encoding data; splicing the first coded data and the second coded data to obtain a model input text; calling a regularization processing model to carry out reasoning processing on the model input text to generate a corresponding regularization text result; and returning a regularized text result to the user. In addition, the regularized text result can be stored in the block chain. Based on the use of the regularization processing model, the complexity and dimension difficulty of the text regularization system are effectively reduced, and the multi-language universality of the text regularization system is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of financial technology, and in particular to artificial intelligence-based text processing methods, devices, computer equipment and storage media. Background Art

[0002] In the field of speech synthesis technology, text regularization processing, as a key step, plays a vital role in improving the accuracy and naturalness of speech output. This process covers multiple aspects such as spelling error correction, grammatical problem adjustment, number and special character processing, etc., aiming to ensure the consistency and fluency of speech output in a multilingual environment, thereby optimizing the user experience. The effective implementation of text regularization is not only crucial to the effectiveness of speech synthesis applications such as voice assistants and voice navigation, but also has a far-reaching impact on improving the accessibility of technology.

[0003] However, the current field of text regularization processing faces many challenges, especially in multilingual processing. Traditionally, rule-based multilingual text regularization systems require the design of complex rule sets for each language, which not only increases the complexity of the system, but also greatly increases the difficulty of maintenance. In addition, the existing rule-based multilingual text regularization has obvious deficiencies in adapting to multilingual texts, making it difficult to flexibly respond to diverse language scenarios and unable to achieve true multilingual versatility. This limitation results in the inability of existing technologies to achieve ideal regularization effects when processing texts containing multiple language elements, which in turn affects the overall performance of speech synthesis and user experience. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a text processing method, apparatus, computer equipment and storage medium based on artificial intelligence to solve the technical problem that the existing rule-based regular processing of multilingual text requires a complex set of rules to be designed separately for each language, which not only increases the complexity of the system, but also greatly increases the difficulty of maintenance, and it is difficult to flexibly respond to diverse language scenarios and cannot achieve true multilingual compatibility.

[0005] In order to solve the above technical problems, the present application embodiment provides a text processing method based on artificial intelligence, which adopts the following technical solutions:

[0006] Receive initial text input from the user;

[0007] Encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data;

[0008] Performing language recognition processing on the initial text to obtain language information corresponding to the initial text;

[0009] Encoding the language information based on the target encoding type to obtain corresponding second encoding data;

[0010] Concatenating the first coded data and the second coded data to obtain a corresponding model input text;

[0011] Calling a pre-built regularization processing model to perform inference processing on the model input text to generate a regularized text result corresponding to the model input text;

[0012] The regularized text result is returned to the user.

[0013] Furthermore, the step of encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data specifically includes:

[0014] Obtaining a target processing method corresponding to the target encoding type;

[0015] Performing encoding processing on the initial text based on the target processing method to obtain corresponding target encoding data;

[0016] The target encoded data is used as the first encoded data.

[0017] Furthermore, the step of performing language recognition processing on the initial text to obtain language information corresponding to the initial text specifically includes:

[0018] Call the preset language recognition tool;

[0019] Performing language recognition processing on the initial text by using the language recognition tool to obtain a language label corresponding to the initial text;

[0020] The language tag returned by the language identification tool is received, and the language tag is used as the language information corresponding to the initial text.

[0021] Furthermore, the step of concatenating the first coded data and the second coded data to obtain a corresponding model input text specifically includes:

[0022] Get the preset splicing order;

[0023] splicing the first coded data and the second coded data based on the splicing order to obtain spliced ​​coded data;

[0024] The concatenated encoded data is used as the model input text.

[0025] Furthermore, before the step of calling the pre-built regularization processing model to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text, the method further includes:

[0026] Construct multilingual text sample data based on pre-collected multilingual text datasets;

[0027] Call the preset language model;

[0028] Pre-training the language model based on the multilingual text sample data to obtain a corresponding first language model;

[0029] Get pre-built multilingual text regularization task data;

[0030] Fine-tune the first language model based on the multilingual text regularization task data to obtain a corresponding second language model;

[0031] The second language model is used as the regular processing model.

[0032] Furthermore, the step of constructing multilingual text sample data based on the pre-collected multilingual text dataset specifically includes:

[0033] Call the preset multilingual text data corpus;

[0034] Acquiring a specified amount of multilingual text data from the multilingual text data corpus to obtain the multilingual text data set;

[0035] Filtering the multilingual text data set based on a preset data filtering rule to obtain a processed multilingual text data set;

[0036] The processed multilingual text data set is used as the multilingual text sample data.

[0037] Furthermore, after the step of calling the pre-built regularization processing model to perform reasoning processing on the model input text and generating a regularized text result corresponding to the model input text, the method further includes:

[0038] Call the pre-trained speech synthesis model;

[0039] Performing speech synthesis processing on the regularized text result based on the speech synthesis model to obtain speech data corresponding to the regularized text result;

[0040] Optimizing the speech data to obtain a corresponding target speech;

[0041] The target speech is returned to the user.

[0042] In order to solve the above technical problems, the embodiment of the present application further provides a text processing device based on artificial intelligence, which adopts the following technical solution:

[0043] A receiving module, used for receiving the initial text input by the user;

[0044] A first processing module, used for encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data;

[0045] A recognition module, used to perform language recognition processing on the initial text to obtain language information corresponding to the initial text;

[0046] A second processing module, configured to encode the language information based on the target encoding type to obtain corresponding second encoded data;

[0047] A concatenation module, used for concatenating the first coded data and the second coded data to obtain a corresponding model input text;

[0048] An inference module, used for calling a pre-built regularization processing model to perform inference processing on the model input text, and generating a regularized text result corresponding to the model input text;

[0049] The first returning module is used to return the regularized text result to the user.

[0050] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0051] Receive initial text input from the user;

[0052] Encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data;

[0053] Performing language recognition processing on the initial text to obtain language information corresponding to the initial text;

[0054] Encoding the language information based on the target encoding type to obtain corresponding second encoding data;

[0055] Concatenating the first coded data and the second coded data to obtain a corresponding model input text;

[0056] Calling a pre-built regularization processing model to perform inference processing on the model input text to generate a regularized text result corresponding to the model input text;

[0057] The regularized text result is returned to the user.

[0058] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0059] Receive initial text input from the user;

[0060] Encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data;

[0061] Performing language recognition processing on the initial text to obtain language information corresponding to the initial text;

[0062] Encoding the language information based on the target encoding type to obtain corresponding second encoding data;

[0063] Concatenating the first coded data and the second coded data to obtain a corresponding model input text;

[0064] Calling a pre-built regularization processing model to perform inference processing on the model input text to generate a regularized text result corresponding to the model input text;

[0065] The regularized text result is returned to the user.

[0066] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0067] The present application first receives the initial text input by the user; then encodes the initial text based on the preset target encoding type to obtain the corresponding first encoded data; then performs language recognition processing on the initial text to obtain the language information corresponding to the initial text; and encodes the language information based on the target encoding type to obtain the corresponding second encoded data; subsequently, the first encoded data and the second encoded data are spliced ​​to obtain the corresponding model input text; further, a pre-built regular processing model is called to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text; and finally, the regularized text result is returned to the user. The present application encodes the initial text based on the target encoding type to obtain the first encoded data in the input format of the regular processing model, and encodes the language information of the initial text based on the target encoding type to obtain the second encoded data in the input format of the regular processing model, and splices the first encoded data and the second encoded data to obtain the model input text, and then performs reasoning processing on the model input text based on the pre-built regular processing model, so as to achieve the rapid and accurate generation of the regularized text result corresponding to the model input text. Based on the characteristics of the target encoding type, this application can support language texts used in most countries in the world, unify the training and reasoning of various languages, and enable a single regular processing model to automatically adapt to the text regularization rules of various languages, and automatically and accurately output regularized text results corresponding to texts in different languages, effectively reducing the workload of designing independent text regularization rules for each language, reducing the complexity and dimensionality difficulty of the text regularization system, and improving the multilingual versatility of the text regularization system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0070] Figure 2 A flowchart of an embodiment of an artificial intelligence-based text processing method according to the present application;

[0071] Figure 3 is a structural schematic diagram of an embodiment of an artificial intelligence-based text processing device according to the present application;

[0072] Figure 4It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0074] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0075] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0076] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0077] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0078] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, and the like.

[0079] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0080] It should be noted that the artificial intelligence-based text processing method provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based text processing device is generally arranged in a server / terminal device.

[0081] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0082] Continue to refer Figure 2 , showing a flowchart of an embodiment of an artificial intelligence-based text processing method according to the present application. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted. The artificial intelligence-based text processing method provided in the embodiment of the present application can be applied to any scenario that requires text regularization processing, and the artificial intelligence-based text processing method can be applied to products in these scenarios, for example, regularization processing of financial text data in the field of finance and insurance. The artificial intelligence-based text processing method comprises the following steps:

[0083] Step S201, receiving initial text input by a user.

[0084] In this embodiment, the text processing method based on artificial intelligence is executed on the electronic device (e.g. Figure 1The server / terminal device shown in the figure) can obtain the initial text through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, Wi-Fi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (Ultra Wi-Fi Band) connection, and other wireless connection methods currently known or to be developed in the future. The executor of the present application may specifically be a text regularization processing system. The above-mentioned initial text is the text that needs to be regularized. The initial text may contain characters, numbers, special symbols, etc. in multiple languages. Exemplarily, in the business scenario of finance and insurance, the initial text may include transaction text data, payment text data, claims text data, and so on.

[0085] Step S202: encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data.

[0086] In this embodiment, the above-mentioned target encoding type is an encoding type that matches the regular processing model. By encoding the initial text based on the target encoding type to obtain the first encoding data, it can be ensured that the regular processing model can accurately process the various languages ​​and special characters contained in the initial text. Among them, the above-mentioned encoding process based on the preset target encoding type to obtain the corresponding first encoding data is further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here.

[0087] Step S203: performing language recognition processing on the initial text to obtain language information corresponding to the initial text.

[0088] In this embodiment, the above-mentioned specific implementation process of performing language recognition processing on the initial text to obtain the language information corresponding to the initial text will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated in detail here.

[0089] Step S204: encoding the language information based on the target encoding type to obtain corresponding second encoded data.

[0090] In this embodiment, the specific implementation process of encoding the language information based on the target encoding type to obtain the corresponding second encoded data can refer to the subsequent encoding process of the initial text based on the preset target encoding type to obtain the corresponding first encoded data, which will not be elaborated here.

[0091] Step S205: concatenate the first encoded data and the second encoded data to obtain a corresponding model input text.

[0092] In this embodiment, the above-mentioned splicing process of the first encoded data and the second encoded data to obtain the corresponding model input text is further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here. Among them, by using the language information of the initial text as the prefix of the model input text, additional contextual information can be provided for the subsequent regularization processing of the regularization model text regularization, and an inference process that can guide the regularization model to perform correct text regularization can be performed.

[0093] Step S206, calling a pre-built regularization processing model to perform inference processing on the model input text, and generating a regularized text result corresponding to the model input text.

[0094] In this embodiment, by inputting the model input text of the language information corresponding to the initial text into a pre-built regular processing model, the regular processing model will perform text regularization tasks according to the input initial text and language information, such as spelling error correction, grammatical problem processing, normalization of numbers and special characters, etc., and output the regularized text result to obtain the above-mentioned regularized text result.

[0095] Step S207: returning the regularized text result to the user.

[0096] In this embodiment, the normalized text result can be returned to the user based on the communication information obtained by obtaining the user's communication information, wherein the communication information can include a mobile phone number or an email address.

[0097] The present application first receives the initial text input by the user; then encodes the initial text based on the preset target encoding type to obtain the corresponding first encoded data; then performs language recognition processing on the initial text to obtain the language information corresponding to the initial text; and encodes the language information based on the target encoding type to obtain the corresponding second encoded data; subsequently, the first encoded data and the second encoded data are spliced ​​to obtain the corresponding model input text; further, a pre-built regular processing model is called to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text; and finally, the regularized text result is returned to the user. The present application encodes the initial text based on the target encoding type to obtain the first encoded data in the input format of the regular processing model, and encodes the language information of the initial text based on the target encoding type to obtain the second encoded data in the input format of the regular processing model, and splices the first encoded data and the second encoded data to obtain the model input text, and then performs reasoning processing on the model input text based on the pre-built regular processing model, so as to achieve the rapid and accurate generation of the regularized text result corresponding to the model input text. Based on the characteristics of the target encoding type, this application can support language texts used in most countries in the world, unify the training and reasoning of various languages, and enable a single regular processing model to automatically adapt to the text regularization rules of various languages, and automatically and accurately output regularized text results corresponding to texts in different languages, effectively reducing the workload of designing independent text regularization rules for each language, reducing the complexity and dimensionality difficulty of the text regularization system, and improving the multilingual versatility of the text regularization system.

[0098] In some optional implementations, step S202 includes the following steps:

[0099] Obtain a target processing method corresponding to the target encoding type.

[0100] In this embodiment, the target encoding type is specifically the UTF-8 byte encoding type. The target processing method is specifically the str.encode('utf-8') method.

[0101] The initial text is encoded based on the target processing method to obtain corresponding target encoded data.

[0102] In this embodiment, the str.encode('utf-8') method is executed to perform UTF-8 byte encoding processing on the initial text, so as to convert the initial text into the corresponding UTF-8 encoding, that is, the above-mentioned target encoding data.

[0103] The target encoded data is used as the first encoded data.

[0104] The present application obtains a target processing method corresponding to the target encoding type; then encodes the initial text based on the target processing method to obtain corresponding target encoding data; and subsequently uses the target encoding data as the first encoding data. The present application obtains a target processing method corresponding to the target encoding type, and then encodes the initial text based on the use of the target processing method, so that the first encoding data that conforms to the target encoding type can be generated quickly and accurately, that is, the initial text is intelligently converted into an acceptable input format for a regular processing model, effectively ensuring that the subsequent regular processing model can smoothly process the first encoding data.

[0105] In some optional implementations of this embodiment, step S203 includes the following steps:

[0106] Call the preset language recognition tool.

[0107] In this embodiment, the language recognition tool may specifically adopt a general language recognition library, such as langdetect (Python library) or other similar tools.

[0108] The language recognition tool is used to perform language recognition processing on the initial text to obtain a language label corresponding to the initial text.

[0109] In this embodiment, the initial text is passed to a language recognition tool to perform language recognition on the initial text, and the language recognition tool returns a recognized language label, such as "en" for English, "zh" for Chinese, etc.

[0110] The language tag returned by the language identification tool is received, and the language tag is used as the language information corresponding to the initial text.

[0111] The present application calls a preset language recognition tool; then uses the language recognition tool to perform language recognition processing on the initial text to obtain a language label corresponding to the initial text; subsequently receives the language label returned by the language recognition tool, and uses the language label as the language information corresponding to the initial text. The present application uses a language recognition tool to perform language recognition processing on the initial text, thereby achieving rapid and accurate generation of language information corresponding to the initial text, thereby improving the efficiency and accuracy of language information generation.

[0112] In some optional implementations, step S205 includes the following steps:

[0113] Get the preset splicing order.

[0114] In this embodiment, there is no specific limitation on the determination of the splicing order, which can be set according to actual business needs. Specifically, the splicing order can be: adding the second coded data corresponding to the language information before the first coded data corresponding to the original text, or can also be: adding the second coded data corresponding to the language information after the first coded data corresponding to the original text.

[0115] The first coded data and the second coded data are spliced ​​based on the splicing order to obtain spliced ​​coded data.

[0116] In this embodiment, the first coded data and the second coded data may be spliced ​​according to the sorting content of the splicing order, so as to obtain the spliced ​​coded data. For example, if the first coded data corresponding to the original text is b"he llo", and the second coded data corresponding to the language information is "en", the spliced ​​coded data may be b"enhe llo" or b"he ll oen".

[0117] The concatenated encoded data is used as the model input text.

[0118] The present application obtains a preset splicing order; then splices the first coded data and the second coded data based on the splicing order to obtain the spliced ​​coded data; and subsequently uses the spliced ​​coded data as the model input text. The present application splices the first coded data and the second coded data based on the acquired splicing order, thereby realizing intelligent and rapid generation of the corresponding model input text, thereby improving the generation efficiency and generation intelligence of the model input text.

[0119] In some optional implementations, before step S206, the electronic device may further perform the following steps:

[0120] Construct multilingual text sample data based on pre-collected multilingual text datasets.

[0121] In this embodiment, the specific implementation process of constructing multilingual text sample data based on the pre-collected multilingual text data set will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0122] Call the preset language model.

[0123] In this embodiment, the above-mentioned language model specifically adopts the ByT5 model. ByT5 is a pre-trained language model based on the Transformer architecture, which converts all NLP tasks into text-to-text format. Compared with mT5 (multilingual text-to-text transfer converter), ByT5 has made some key changes in the architecture to directly process byte sequences instead of subword token sequences. This includes omitting the SentencePiece vocabulary and directly inputting UTF-8 bytes into the model without any text preprocessing. The ByT5 model uses UTF-8bytes encoded input tokens, a move designed to enhance the model's adaptability to multilingual character sets. By using UTF-8bytes encoding, the model is able to more accurately and comprehensively represent characters in various languages, thereby better handling the challenges of multilingual text.

[0124] The language model is pre-trained based on the multilingual text sample data to obtain a corresponding first language model.

[0125] In this embodiment, training tasks corresponding to the ByT5 model are pre-constructed. The training tasks include: 1) Self-supervised learning tasks: ByT5 is pre-trained using self-supervised learning methods, which include fill-in-the-blank tasks, masked language model tasks, etc. In the masked language model task, the model needs to predict randomly masked words or byte sequences in the text. 2) UTF-8 byte encoding: ByT5 operates directly on UTF-8 byte sequences instead of using a traditional vocabulary. This allows the model to process arbitrary text sequences, including text containing unknown words or special characters.

[0126] Specifically, the training process of the ByT5 model includes: using the constructed multilingual text sample data to train the ByT5 model. During the training process, the ByT5 model will continuously iterate and optimize its parameters to better capture the semantic information and language features in the text. After the training is completed, the obtained first language model will have the ability to process texts in multiple languages ​​and provide high-quality text representation for downstream tasks.

[0127] In the pre-training stage, ByT5 inherited the self-supervised learning framework of mT5 and trained on a large amount of multilingual parallel text data through self-supervised learning methods such as fill-in-the-blank tasks and mask tasks. In this stage, the model learns to map UTF-8 bytes encoded multilingual text to target language text, which not only strengthens the general representation learning of multilingual text, but also improves the ability to flexibly respond to different language character sets and grammatical structures.

[0128] Get pre-built multilingual text regularization task data.

[0129] In this embodiment, in order to apply the ByT5 model to the text regularization task and optimize the processing capability of the ByT5 model for the text regularization task, it is necessary to fine-tune the ByT5 model using a labeled data set containing multilingual text regularization tasks. The labeled data set contains input text and expected output text related to the text regularization task. In addition, the labeled data set is preprocessed (including text encoding, word segmentation and other steps) to ensure that its format is consistent with the model input requirements, thereby obtaining the above-mentioned multilingual text regularization task data.

[0130] The first language model is fine-tuned based on the multilingual text regularization task data to obtain a corresponding second language model.

[0131] In this embodiment, fine-tuning tasks with the ByT5 model are pre-built. The fine-tuning tasks include: 1) Text regularization task: During the fine-tuning stage, the ByT5 model will be trained for the text regularization task. This includes correction of spelling errors, handling of grammatical issues, normalization of numbers and special characters, etc. 2) Task-specific parameter adjustment: According to the requirements of the fine-tuning task, certain parameters of the model can be adjusted to optimize the performance of the model on the task.

[0132] Specifically, the process of fine-tuning the first language model includes: fine-tuning the first language model using the multilingual text regularization task data. During the fine-tuning process, the model will learn how to convert the input text into the desired output text, and continuously optimize its parameters to improve performance. After the fine-tuning is completed, the obtained second language model will be optimized for a specific task and can achieve better performance on the task.

[0133] In the fine-tuning phase, ByT5 further optimized the model by fine-tuning using multilingual text regularization task data. Multilingual text encoded in UTF-8 bytes was introduced, and the model was adjusted to perform text regularization tasks such as correction of spelling errors, handling of grammatical issues, normalization of numbers and special characters, etc. The purpose of fine-tuning is to fine-tune the model parameters to make it more suitable for the specific needs of multilingual text regularization, so as to achieve higher accuracy and consistency when processing multilingual mixed text.

[0134] The second language model is used as the regular processing model.

[0135] In this embodiment, the training process of the ByT5 model is a process from general to specific. In the pre-training stage, the model learns general text representation capabilities through a large-scale multilingual text dataset; in the fine-tuning stage, the model is trained and optimized for a specific task (text regularization task) to improve performance on the task. This training method enables the ByT5 model to have the ability to process texts in multiple languages ​​and to efficiently process specific tasks.

[0136] Among them, the ByT5 model is used to model the regularization of multilingual text, which can simplify the process of the entire text regularization system. With the characteristics of UTF-8, the text regularization system can support texts in languages ​​used in most countries in the world. The training and reasoning of various languages ​​are unified, so that the system can automatically adapt to the text rules of various languages ​​through a single model without manual design and adjustment. In addition, the introduction of multilingual data helps to reduce the model's dependence on specific languages. The model will simultaneously learn the distribution of each language in the same feature space, improving its consistency and effect when processing different languages. In addition, the introduction of multilingual regularization data makes the model more robust, and through the synergy of regular modeling of each language, the model's generalization ability is improved, thereby improving overall performance and user experience.

[0137] The present application constructs multilingual text sample data based on a pre-collected multilingual text data set; then calls a preset language model; then pre-trains the language model based on the multilingual text sample data to obtain a corresponding first language model; subsequently obtains pre-constructed multilingual text regularization task data; further fine-tunes the first language model based on the multilingual text regularization task data to obtain a corresponding second language model; and finally uses the second language model as the regularization processing model. The present application obtains a first language model by pre-training a language model using multilingual text sample data pre-constructed based on a pre-collected multilingual text data set, and then fine-tunes the first language model using pre-constructed multilingual text regularization task data, so that a regularization processing model that can quickly and accurately train the ability to process texts in multiple languages ​​and can efficiently process text regularization tasks can be achieved, effectively improving the construction intelligence of the regularization processing model and ensuring the model effect of the regularization processing model. In addition, by introducing multilingual regularization data, the model is more multilingual adaptable, can more accurately process texts in different languages, and ensures the accuracy and naturalness of speech synthesis in a multilingual environment. In addition, the use of regular processing models reduces the workload of designing independent text regularization rules for each language, and improves the versatility of the text regularization system. It helps to reduce the deployment and maintenance costs of speech synthesis applications. In a multilingual environment, the improved text regularization system can process texts from different languages ​​and provide more universal and extensive speech synthesis services to global users, which is particularly important for cross-cultural and multi-context applications.

[0138] In some optional implementations of this embodiment, constructing multilingual text sample data based on a pre-collected multilingual text dataset includes the following steps:

[0139] Call the preset multilingual text data corpus.

[0140] In this embodiment, the ByT5 model is usually pre-trained using a large multilingual text data set. The multilingual text data corpus may specifically be the mC4 corpus, which is a large unlabeled multilingual text data corpus.

[0141] A specified amount of multilingual text data is acquired from the multilingual text data corpus to obtain the multilingual text data set.

[0142] In this embodiment, there is no specific limitation on the value of the above specified number, which can be set according to the actual model training requirements. The above multilingual text dataset contains a large amount of diverse texts, covering multiple languages.

[0143] The multilingual text data set is filtered based on a preset data filtering rule to obtain a processed multilingual text data set.

[0144] In this embodiment, the data filtering rules include data filtering processing to remove noise, duplicates and useless information. By using the data filtering rules to filter the multilingual text dataset, the quality and diversity of the processed multilingual text dataset can be ensured.

[0145] The processed multilingual text data set is used as the multilingual text sample data.

[0146] The present application calls a preset multilingual text data corpus; then obtains a specified amount of multilingual text data from the multilingual text data corpus to obtain the multilingual text data set; then filters the multilingual text data set based on preset data filtering rules to obtain a processed multilingual text data set; and subsequently uses the processed multilingual text data set as the multilingual text sample data. The present application obtains a specified amount of multilingual text data from a preset multilingual text data corpus to obtain the multilingual text data set, and then filters the multilingual text data set based on preset data filtering rules, thereby realizing the rapid and accurate construction of multilingual text sample data for pre-training a language model, effectively ensuring the quality and diversity of the obtained multilingual text sample data.

[0147] In some optional implementations of this embodiment, after step S206, the electronic device may further perform the following steps:

[0148] Call a pre-trained speech synthesis model.

[0149] In this embodiment, there is no specific limitation on the selection of the above-mentioned speech synthesis model, which can be determined according to the actual speech synthesis task requirements. For example, any one of a speech synthesis model based on statistical parameter modeling (such as HMM-based), a speech synthesis model based on deep learning (such as RNN, CNN, Transformer, etc.), and an end-to-end speech synthesis model (such as Tacotron, FastSpeech, etc.) can be used as the above-mentioned speech synthesis model. Specifically, the training process of the speech synthesis model includes: using a large amount of annotated speech data for model training. The annotated speech data includes text and corresponding speech waveforms. During the training process, the model will learn the mapping relationship between text and speech, including pronunciation, intonation, speaking speed and other features, and finally obtain a trained speech synthesis model.

[0150] The regularized text result is subjected to speech synthesis processing based on the speech synthesis model to obtain speech data corresponding to the regularized text result.

[0151] In this embodiment, the regularized text result is input into the speech synthesis model, and then the speech synthesis model generates a corresponding speech waveform according to the input regularized text result, that is, the above-mentioned speech data. The speech waveform generation process includes multiple steps such as text conversion to phonemes, acoustic feature prediction, and waveform generation.

[0152] The speech data is optimized to obtain the corresponding target speech.

[0153] In this embodiment, the above-mentioned optimization processing may include denoising processing and smoothing processing. By performing the optimization processing on the speech data, the speech quality of the obtained target speech can be effectively improved.

[0154] The target speech is returned to the user.

[0155] In this embodiment, the target voice may be returned to the user based on the communication information obtained from the user, wherein the communication information may include a mobile phone number or a communication address.

[0156] This application calls a pre-trained speech synthesis model; then performs speech synthesis processing on the regularized text result based on the speech synthesis model to obtain speech data corresponding to the regularized text result; then optimizes the speech data to obtain the corresponding target speech; and subsequently returns the target speech to the user. After calling a pre-built regular processing model to perform inference processing on the model input text and generating a regularized text result corresponding to the model input text, this application will also intelligently call a pre-trained speech synthesis model to perform speech synthesis processing on the regularized text result to obtain speech data corresponding to the regularized text result, and optimize the speech data to obtain the corresponding target speech, and then return the target speech to the user, thereby automatically completing the speech synthesis processing of the initial text input by the user, so that the user does not need to perform speech synthesis processing again, which is conducive to improving the user's work efficiency and usage experience.

[0157] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0158] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned regularized text results, the above-mentioned regularized text results can also be stored in a node of a blockchain.

[0159] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0160] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0161] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0162] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0163] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0164] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a text processing device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0165] like Figure 3 As shown, the text processing device 300 based on artificial intelligence described in this embodiment includes: a receiving module 301, a first processing module 302, a recognition module 303, a second processing module 304, a splicing module 305, a reasoning module 306 and a first return module 307. Among them:

[0166] Receiving module 301, used for receiving initial text input by a user;

[0167] A first processing module 302 is used to encode the initial text based on a preset target encoding type to obtain corresponding first encoded data;

[0168] The recognition module 303 is used to perform language recognition processing on the initial text to obtain language information corresponding to the initial text;

[0169] A second processing module 304 is used to encode the language information based on the target encoding type to obtain corresponding second encoded data;

[0170] A concatenation module 305 is used to concatenate the first coded data and the second coded data to obtain a corresponding model input text;

[0171] The reasoning module 306 is used to call the pre-built regularization processing model to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text;

[0172] The first returning module 307 is used to return the regularized text result to the user.

[0173] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0174] In some optional implementations of this embodiment, the first processing module 302 includes:

[0175] A first acquisition submodule, used to acquire a target processing method corresponding to the target encoding type;

[0176] An encoding submodule, used for encoding the initial text based on the target processing method to obtain corresponding target encoding data;

[0177] The first determining submodule is used to use the target coded data as the first coded data.

[0178] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0179] In some optional implementations of this embodiment, the identification module 303 includes:

[0180] The first calling submodule is used to call a preset language recognition tool;

[0181] An identification submodule, configured to perform language identification processing on the initial text by using the language identification tool to obtain a language label corresponding to the initial text;

[0182] The second determination submodule is used to receive the language label returned by the language identification tool, and use the language label as the language information corresponding to the initial text.

[0183] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0184] In some optional implementations of this embodiment, the splicing module 305 includes:

[0185] The second acquisition submodule is used to acquire a preset splicing order;

[0186] A splicing submodule, configured to splice the first coded data and the second coded data based on the splicing order to obtain spliced ​​coded data;

[0187] The third determination submodule is used to use the concatenated encoded data as the model input text.

[0188] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0189] In some optional implementations of this embodiment, the text processing device based on artificial intelligence further includes:

[0190] A construction module, used to construct multilingual text sample data based on a pre-collected multilingual text dataset;

[0191] A first calling module, used to call a preset language model;

[0192] A pre-training module, configured to perform pre-training processing on the language model based on the multi-language text sample data to obtain a corresponding first language model;

[0193] The acquisition module is used to obtain pre-built multilingual text regularization task data;

[0194] A fine-tuning module, configured to fine-tune the first language model based on the multilingual text regularization task data to obtain a corresponding second language model;

[0195] A determination module is used to use the second language model as the regular processing model.

[0196] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0197] In some optional implementations of this embodiment, the building blocks include:

[0198] The second calling submodule is used to call a preset multilingual text data corpus;

[0199] A third acquisition submodule is used to acquire a specified amount of multilingual text data from the multilingual text data corpus to obtain the multilingual text data set;

[0200] A filtering submodule, configured to filter the multilingual text data set based on a preset data filtering rule to obtain a processed multilingual text data set;

[0201] The fourth determining submodule is configured to use the processed multilingual text data set as the multilingual text sample data.

[0202] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0203] In some optional implementations of this embodiment, the text processing device based on artificial intelligence further includes:

[0204] The second calling module is used to call the pre-trained speech synthesis model;

[0205] A synthesis module, used for performing speech synthesis processing on the regularized text result based on the speech synthesis model to obtain speech data corresponding to the regularized text result;

[0206] An optimization module, used for optimizing the speech data to obtain a corresponding target speech;

[0207] The second returning module is used to return the target speech to the user.

[0208] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based text processing method in the aforementioned implementation, and will not be repeated here.

[0209] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0210] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0211] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0212] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of a text processing method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0213] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the computer-readable instructions stored in the memory 41 or process data, such as running the computer-readable instructions of the artificial intelligence-based text processing method.

[0214] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0215] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0216] In an embodiment of the present application, the initial text is encoded based on the target encoding type to obtain first encoded data that conforms to the input format of the regular processing model, and the language information of the initial text is encoded based on the target encoding type to obtain second encoded data that conforms to the input format of the regular processing model, and the first encoded data and the second encoded data are spliced ​​to obtain the model input text, and then the model input text is inferred based on the pre-built regular processing model, so that the regularized text result corresponding to the model input text can be generated quickly and accurately. Based on the characteristics of the target encoding type, the present application can support language texts used in most countries in the world, unify the training and reasoning of various languages, so that a single regular processing model can automatically adapt to the text regularization rules of various languages, and automatically and accurately output regularized text results corresponding to texts in different languages, effectively reducing the workload of designing independent text regularization rules for each language, reducing the complexity and dimensionality difficulty of the text regularization system, and improving the multi-language versatility of the text regularization system.

[0217] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based text processing method.

[0218] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0219] In an embodiment of the present application, the initial text is encoded based on the target encoding type to obtain first encoded data that conforms to the input format of the regular processing model, and the language information of the initial text is encoded based on the target encoding type to obtain second encoded data that conforms to the input format of the regular processing model, and the first encoded data and the second encoded data are spliced ​​to obtain the model input text, and then the model input text is inferred based on the pre-built regular processing model, so that the regularized text result corresponding to the model input text can be generated quickly and accurately. Based on the characteristics of the target encoding type, the present application can support language texts used in most countries in the world, unify the training and reasoning of various languages, so that a single regular processing model can automatically adapt to the text regularization rules of various languages, and automatically and accurately output regularized text results corresponding to texts in different languages, effectively reducing the workload of designing independent text regularization rules for each language, reducing the complexity and dimensionality difficulty of the text regularization system, and improving the multi-language versatility of the text regularization system.

[0220] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0221] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A text processing method based on artificial intelligence, characterized in that: The steps include: Receive initial text input from the user; Encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data; Performing language recognition processing on the initial text to obtain language information corresponding to the initial text; Encoding the language information based on the target encoding type to obtain corresponding second encoding data; Concatenating the first coded data and the second coded data to obtain a corresponding model input text; Calling a pre-built regularization processing model to perform inference processing on the model input text to generate a regularized text result corresponding to the model input text; The regularized text result is returned to the user.

2. The text processing method based on artificial intelligence according to claim 1, characterized in that: The step of encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data specifically includes: Obtaining a target processing method corresponding to the target encoding type; Performing encoding processing on the initial text based on the target processing method to obtain corresponding target encoding data; The target encoded data is used as the first encoded data.

3. The text processing method based on artificial intelligence according to claim 1, characterized in that: The step of performing language recognition processing on the initial text to obtain language information corresponding to the initial text specifically includes: Call the preset language recognition tool; Performing language recognition processing on the initial text by using the language recognition tool to obtain a language label corresponding to the initial text; The language tag returned by the language identification tool is received, and the language tag is used as the language information corresponding to the initial text.

4. The text processing method based on artificial intelligence according to claim 1, characterized in that: The step of concatenating the first coded data and the second coded data to obtain a corresponding model input text specifically includes: Get the preset splicing order; splicing the first coded data and the second coded data based on the splicing order to obtain spliced ​​coded data; The concatenated encoded data is used as the model input text.

5. The text processing method based on artificial intelligence according to claim 1, characterized in that: Before the step of calling the pre-built regularization processing model to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text, the method further includes: Construct multilingual text sample data based on pre-collected multilingual text datasets; Call the preset language model; Pre-training the language model based on the multilingual text sample data to obtain a corresponding first language model; Get pre-built multilingual text regularization task data; Fine-tune the first language model based on the multilingual text regularization task data to obtain a corresponding second language model; The second language model is used as the regular processing model.

6. The text processing method based on artificial intelligence according to claim 5, characterized in that: The step of constructing multilingual text sample data based on the pre-collected multilingual text dataset specifically includes: Call the preset multilingual text data corpus; Acquiring a specified amount of multilingual text data from the multilingual text data corpus to obtain the multilingual text data set; Filtering the multilingual text data set based on a preset data filtering rule to obtain a processed multilingual text data set; The processed multilingual text data set is used as the multilingual text sample data.

7. The text processing method based on artificial intelligence according to claim 1, characterized in that: After the step of calling the pre-built regularization processing model to perform reasoning processing on the model input text to generate a regularized text result corresponding to the model input text, the method further includes: Call the pre-trained speech synthesis model; Performing speech synthesis processing on the regularized text result based on the speech synthesis model to obtain speech data corresponding to the regularized text result; Optimizing the speech data to obtain a corresponding target speech; The target speech is returned to the user.

8. A text processing device based on artificial intelligence, characterized in that: include: A receiving module, used for receiving the initial text input by the user; A first processing module, used for encoding the initial text based on a preset target encoding type to obtain corresponding first encoded data; A recognition module, used to perform language recognition processing on the initial text to obtain language information corresponding to the initial text; A second processing module, configured to encode the language information based on the target encoding type to obtain corresponding second encoded data; A concatenation module, used for concatenating the first coded data and the second coded data to obtain a corresponding model input text; An inference module, used for calling a pre-built regularization processing model to perform inference processing on the model input text, and generating a regularized text result corresponding to the model input text; The first returning module is used to return the regularized text result to the user.

9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based text processing method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the artificial intelligence-based text processing method as described in any one of claims 1 to 7 are implemented.

Images