Text acquisition method and device, medium and equipment

Abstract

The embodiment of the invention discloses a text acquisition method and device, a medium and equipment. And a personalized speech recognition model is provided for each user. The scheme can be applied to a trusted execution environment. A user needs to confirm or deny the text recognized by the voice recognition model in the process of using the special voice recognition model of the user, and if the user denies the text recognized by the voice recognition model, a text error correction mechanism is triggered, so that the text which the user wants to input this time can be corrected, and the user experience is improved. And correct samples can be provided for iterative optimization of the speech recognition model.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a text acquisition method, device, medium and equipment. Background Art

[0002] Currently, users can use Automatic Speech Recognition (ASR) models for text input. The speech recognition models on the market are designed for a wide range of user groups, aiming to take into account a variety of different accents and language habits.

[0003] However, existing speech recognition models are not fully compatible with the unique accents and language habits of different individual users. When individual users use the speech recognition models, the recognized text is prone to errors. Summary of the invention

[0004] The present invention provides a method for obtaining text, including:

[0005] Receive a voice signal from a target user;

[0006] Calling a speech recognition model dedicated to the target user to process the speech signal to obtain a recognized text;

[0007] In response to a signal from the target user refusing to confirm the recognized text, calling a text error correction process to process the recognized text to obtain a corrected text;

[0008] Determine the text input by the target user according to the error-corrected text; and continue to train the speech recognition model according to the speech signal and the error-corrected text.

[0009] The embodiment of this specification provides a text acquisition device, including:

[0010] A receiving module receives a voice signal from a target user;

[0011] A recognition module, calling a speech recognition model dedicated to the target user, processing the speech signal, and obtaining a recognition text;

[0012] An error correction module, in response to a signal from the target user refusing to confirm the recognized text, calls a text error correction process to process the recognized text to obtain a corrected text;

[0013] A processing module determines the text input by the target user according to the error-corrected text; and continues to train the speech recognition model according to the speech signal and the error-corrected text.

[0014] The embodiments of the present specification also provide a computer program product, wherein the computer program product stores at least one instruction, and the at least one instruction is suitable for being loaded by a processor and executing the above method steps.

[0015] The embodiments of the present specification also provide a storage medium, wherein the storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the steps of the above method.

[0016] An embodiment of the present specification also provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the above method.

[0017] In the technical solution of this specification, a personalized speech recognition model is provided for each user. When using their own dedicated speech recognition model, the user needs to confirm or deny the text recognized by the speech recognition model. If the user denies the text recognized by the speech recognition model, the text error correction mechanism will be triggered, which can not only correct the text that the user wants to enter this time, but also provide training samples (the user's speech signal and the corrected text) for the iterative optimization of the speech recognition model.

[0018] In other words, users can have their own dedicated speech recognition model to assist in text input. At the same time, the speech recognition model needs to be iteratively tuned, and the user personally supervises the iterative tuning of the speech recognition model (by not confirming the recognized text output by the speech recognition model, triggering the text correction mechanism). As users use their own dedicated speech recognition model for longer and longer periods of time, the speech recognition capabilities of the speech recognition model become more and more compatible with the user's unique accent and language habits.

[0019] Through the above technical solution, the accuracy of users' text input using the speech recognition model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a text acquisition method provided in an embodiment of this specification;

[0021] Figure 2 It is a schematic diagram of the process of obtaining annotation information in the data annotation scenario provided in the embodiment of this specification;

[0022] Figure 3 is a structural schematic diagram of a text acquisition device provided in an embodiment of this specification;

[0023] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. This solution can be applied in a trusted execution environment. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this specification.

[0025] In modern society, with the rapid development of information technology, users have an increasing demand for text input in their daily lives and work. In many scenarios, users need to quickly and accurately input a large amount of text information, such as writing documents, sending messages, filling out forms, adding annotation information to model training samples, and so on. However, traditional manual input methods, such as using a keyboard for text input, are inefficient. Although keyboard input is currently the most common text input method, it requires users to have a certain typing speed and accuracy, which is often a challenge for some users who are not familiar with keyboard operations. In addition, long-term keyboard input may also cause hand fatigue, affecting user work efficiency and comfort.

[0026] In order to improve the efficiency of text input, speech recognition technology came into being. Speech recognition technology converts the user's voice signal into text, thereby achieving efficient text input. This technology can not only free the user's hands, but also greatly improve the speed and convenience of text input. Users only need to input through voice to quickly generate text content, which greatly improves the efficiency of work and life.

[0027] However, the core of speech recognition technology lies in its recognition accuracy. The accuracy of the speech recognition model directly determines the quality of text input and user experience. Most of the speech recognition models currently available on the market are designed for a wide range of user groups. These models take into account a variety of accents and language habits during the development process, aiming to take into account the needs of users from different regions and cultural backgrounds. Although this design can meet the usage needs of most users to a certain extent, it cannot fully accommodate the unique accent and language habits of each user.

[0028] Each person's speech characteristics are unique, including differences in speaking speed, pronunciation, intonation, etc. Therefore, when individual users use these general speech recognition models, recognition errors often occur. These errors not only affect the accuracy of text input, but may also cause users to spend extra time and energy on text proofreading and modification, reducing the efficiency of speech recognition technology and user satisfaction.

[0029] In addition, the voice input requirements in different scenarios are also different. For example, in a noisy environment, background noise may interfere with the normal operation of the voice recognition model, resulting in an increase in the recognition error rate. In another example, in some professional fields, some professional terms or specific expressions may be used, which are often not within the training data range of the general voice recognition model, resulting in recognition errors.

[0030] In summary, the existing general speech recognition model is difficult to be fully compatible with the unique accents and language habits of different users, resulting in errors in the recognized text. To solve this problem, it is necessary to develop a speech recognition model that can adapt to individual differences, thereby improving the accuracy of speech recognition and user experience.

[0031] To this end, in one or more embodiments provided in the present disclosure, a personalized speech recognition model is provided for each user. When using their own dedicated speech recognition model, the user needs to confirm or deny the text recognized by the speech recognition model. If the user denies the text recognized by the speech recognition model, the text error correction mechanism will be triggered, which can not only correct the text that the user wants to enter this time, but also provide training samples (the user's current speech signal and the text after error correction) for the iterative optimization of the speech recognition model.

[0032] In other words, users can have their own dedicated speech recognition model to assist in text input. At the same time, the speech recognition model needs to be iteratively tuned, and the user personally supervises the iterative tuning of the speech recognition model (by not confirming the recognized text output by the speech recognition model, triggering the text correction mechanism). As users use their own dedicated speech recognition model for longer and longer periods of time, the speech recognition capabilities of the speech recognition model become more and more compatible with the user's unique accent and language habits.

[0033] In addition, it should be noted that for the same user, he often uses the speech recognition model for text input in a specific scenario. For example, for the labelers of the model training samples, they often use the speech recognition model to input text (labeling information) in an office environment (with slight human noise), and an labeler is often responsible for the labeling of training sample data in a specific field, which means that the labeling information he needs to input often contains specialized terms in a specific field. Therefore, as users use their dedicated speech recognition model for longer and longer periods of time, the speech recognition capabilities of the speech recognition model will become more and more adapted to the characteristics of the specific scenario in which the user is located.

[0034] Through the above technical solution, the accuracy of text input by users using the speech recognition model can be improved.

[0035] The technical solution of the present disclosure is described in detail below in conjunction with the accompanying drawings.

[0036] Figure 1 : is a flowchart of a text acquisition method provided by the present disclosure, comprising the following steps:

[0037] S100: Receive a voice signal from a target user.

[0038] Figure 1 The method flow shown can be applied to a text acquisition platform. The text acquisition platform can be a software system that can collect a user's voice signal and has a demand for text corresponding to the voice signal.

[0039] As an example, Figure 1 The method flow shown can be applied to the model training data annotation scenario, the text acquisition platform can be a data annotation system, the target user is the annotator, and the text that the target user needs to enter is the annotation information. The annotator can log in to the data annotation system through the identity identification on the computing device (such as a mobile phone, computer), the data annotation system pushes the training sample data to be annotated to the annotator, and the annotator inputs the annotation information corresponding to the training sample data into the data annotation system. For example, the training sample data is "Why are children prone to fever at 6 months old?", and the annotation information that the annotator needs to enter is the answer to this question.

[0040] S102: calling a speech recognition model dedicated to the target user, processing the speech signal, and obtaining a recognized text.

[0041] In one or more embodiments of the present disclosure, each user uses his or her own dedicated speech recognition model, and different users use different speech recognition models. It is easy to understand that in the initialization phase, a common speech recognition model can be assigned to each user. In other words, the speech recognition models used by different users for the first time are the same. As different users continue to use their dedicated speech recognition models, Figure 1 As shown in the method flow, speech recognition models dedicated to different users will also undergo iterative adjustments, and each user-specific speech recognition model will become more and more adapted to the needs of this user.

[0042] It should be noted that in the data annotation scenario, the data annotation system can pre-assign a dedicated speech recognition model to each registered annotator. When the data annotation system interacts with a certain annotator, it will call the speech recognition model dedicated to this annotator.

[0043] As a specific example, the speech recognition model may include a speech feature extraction (SpeechFM, Speech Foundation Model) layer, a speech recognition encoder (Encoder for Automatic SpeechRecognition), and a speech recognition decoder (Decoder for Automatic Speech Recognition) connected in sequence. The speech feature extraction layer refers to the basic layer in the speech recognition model, which is mainly responsible for the preliminary processing and feature extraction of the input speech signal. The speech recognition encoder is responsible for further encoding the feature vector processed by the speech feature extraction layer into a representation suitable for processing by the speech recognition decoder. The speech recognition decoder is responsible for converting the high-level representation generated by the speech recognition encoder into readable text. It should be noted that the LoRA weight matrix of the decoder in the speech recognition model dedicated to different users is different.

[0044] S104: In response to the signal from the target user that the target user refuses to confirm the recognized text, a text error correction process is called to process the recognized text to obtain a corrected text.

[0045] If the user finds that the recognized text given by the dedicated speech recognition model has errors (such as word order errors, text errors, etc.), the user can refuse to confirm the recognized text. It should be noted that the signal that the target user refuses to confirm the recognized text can be obtained in many ways.

[0046] In some embodiments, if the target user does not perform an operation to confirm the recognized text within a specified period of time after the speech recognition model outputs the recognized text, it can be determined that the target user refuses to confirm the recognized text.

[0047] In other embodiments, if it is detected that the user has performed an operation of refusing to confirm the recognized text, it can be determined that the target user refuses to confirm the recognized text.

[0048] The above text error correction process can be implemented in a variety of ways. In some embodiments, the recognized text is displayed to the target user in an editable manner; and the corrected text is determined based on the editing operation of the target user. For example, a text box can be displayed to the target user, and the text box contains the recognized text output by the speech recognition model. The target user can modify and edit the recognized text in the text box to obtain the corrected text.

[0049] In some embodiments, the automatic error correction model can be called to process the recognized text to obtain the corrected text. It is easy to understand that the input of the automatic error correction model is the recognized text to be corrected, and the output is the corrected text. It should be noted that in the above embodiments, a trained automatic error correction model can be used.

[0050] As an example, the automatic error correction model may include a text conversion layer and a supervised fine-tuning (SFT) layer connected in sequence. The text conversion layer may be a T5 (Text-To-Text Transfer Transformer) layer, or a large language model (LLM) layer.

[0051] In other embodiments, an automatic error correction model may be called to process the recognized text to obtain a corrected intermediate text; in response to a signal from the target user refusing to confirm the corrected intermediate text, the corrected intermediate text may be displayed to the target user in an editable manner; and based on the editing operation of the target user, the corrected text may be determined.

[0052] That is to say, for the recognition text output by the speech recognition model, the automatic error correction model can be called first for processing to obtain the corrected intermediate text. If the corrected intermediate text still has errors in the eyes of the target user, the target user can continue to modify and edit the corrected intermediate text to obtain the corrected text.

[0053] In addition, in response to a signal from the target user confirming the error-corrected intermediate text, a post-corrected text may be determined based on the error-corrected intermediate text.

[0054] Furthermore, the automatic error correction model can be further trained based on the recognized text and the corrected text. That is to say, in the above embodiment, the automatic error correction model can be iteratively optimized during the use of the automatic error correction model. As the automatic error correction model is continuously iteratively optimized, the corrected intermediate text obtained by correcting the recognized text becomes more and more accurate, and the target user is less and less required to perform manual modification and editing.

[0055] It should be noted that the above automatic error correction model may not be dedicated to the target user, but may be used by multiple users (the target user and at least one other user). In this way, the automatic error correction model may be iteratively optimized by using the recognition texts and error-corrected texts generated by multiple users.

[0056] In addition, in the data labeling scenario, the corresponding domain vocabulary can be used to participate in the iterative optimization of the automatic error correction model based on the domain to which the training sample data labeled by the target user belongs.

[0057] S106: Determine the text input by the target user according to the corrected text.

[0058] In some embodiments, the corrected text may be directly used as the text input by the target user.

[0059] In other embodiments, a language model may be called to optimize the text structure of the corrected text and / or optimize the text expression of the corrected text. Then, the optimized corrected text may be used as the text input by the target user.

[0060] S108: Continue training the speech recognition model according to the speech signal and the corrected text.

[0061] It is easy to understand that the speech signal, recognition text, and corrected text generated by each speech of the target user can be formed into a triple. The speech signal and corrected text in the triple can be used to continue training the speech recognition model, and the recognition text and corrected text in the triple can be used to continue training the automatic error correction model. Based on this, it is necessary to explain the timing of iterative optimization of the speech recognition model and the automatic error correction model.

[0062] In some embodiments, whenever the number of accumulated triplets reaches a specified number, the specified number of triplets can be used to iteratively optimize the speech recognition model and the automatic error correction model once, and then the accumulated triplets can be cleared, and the specified number of triplets can be accumulated again before continuing to iteratively optimize the speech recognition model and the automatic error correction model.

[0063] In other embodiments, it is possible to count which triples have a specified frequency of occurrence, and use the triples that have reached the specified frequency to iteratively optimize the speech recognition model and the automatic error correction model.

[0064] Through the above method, the number of iterations of the speech recognition model and the automatic error correction model can be reduced, the model training cost can be reduced, and the model training efficiency can be improved.

[0065] S110: In response to a signal from the target user confirming the recognized text, determining text input by the target user according to the recognized text.

[0066] In some embodiments, the recognized text may be directly determined as the text input by the target user.

[0067] In other embodiments, a language model is called to optimize the text structure of the recognized text and / or to optimize the text expression of the recognized text; the optimized recognized text is used as the text input by the target user.

[0068] Figure 2 This is a flow chart of obtaining annotation information in the data annotation scenario provided by the present disclosure. Figure 2 The process shown in the figure introduces the interaction between the data annotation platform and the annotator as follows:

[0069] The data annotation platform pushes the training samples to be annotated to the annotators, who can clarify the annotation information they want to give based on their own knowledge and experience, or with the help of search engines and big prediction models. The annotators can express the annotation information by voice by clicking the "Start Input" button, and end the voice input by clicking the "End Input" button.

[0070] It should be noted that the annotator can record multiple voice segments in batches. That is to say, the voice signal recorded by the annotator by clicking "Start Recording" and "End Recording" once may only be a part of the annotation information given by the annotator for the current training sample.

[0071] In response to the annotator clicking "Start Input" and "End Input" once, the data annotation platform receives the voice signal, calls the voice recognition model to recognize the voice signal, obtains the recognized text, and displays the recognized text in the text box. After the annotator sees the recognized text in the text box, if he is sure that it is the text he wants to express, he can click the "Confirm" button below the text box. If you think the recognized text is wrong, you can click the "Correct" button below the text box to trigger the error correction process, or do not click the "Confirm" button and wait for a certain period of time to automatically trigger the error correction process.

[0072] The error correction process includes error correction by the automatic error correction model and manual error correction. First, the data annotation platform calls the automatic error correction model to process the recognized text, obtains the corrected intermediate text, and displays the corrected intermediate text in the text box. If the annotator is sure that the corrected intermediate text is correct, he can click the "Confirm" button below the text box to obtain the corrected text; if he thinks the recognized text is incorrect, he can modify and edit it in the text box and click the "Confirm" button to obtain the corrected text. The speech signal, recognized text, and corrected text generated by the annotator this time form a triple, which is stored by the data annotation platform for continued training of the speech recognition model and the automatic error correction model.

[0073] In short, once the annotator clicks the "Confirm" button below the text box, the data annotation platform will identify the text in the text box as the annotator's confirmation text. For the same training sample, the annotator can provide at least one confirmation text through the technical solution provided by the present disclosure to form annotation information. Users can supplement and delete the content of the annotation information (different from the error correction in speech recognition). The data annotation platform can also call a large language model to optimize the text structure and expression of the annotation information (such as text polishing).

[0074] The present disclosure also provides a structural schematic diagram of a text acquisition device, including:

[0075] Receiving module 301, receiving a voice signal of a target user;

[0076] The recognition module 302 calls the speech recognition model dedicated to the target user, processes the speech signal, and obtains the recognition text;

[0077] The error correction module 303, in response to the signal of the target user refusing to confirm the recognized text, calls a text error correction process to process the recognized text to obtain a corrected text;

[0078] The processing module 304 determines the text input by the target user according to the corrected text; and continues to train the speech recognition model according to the speech signal and the corrected text.

[0079] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0080] The embodiments of the present specification also provide a computer storage medium, which can store a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method of the embodiments of the present disclosure.

[0081] The present specification also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded by the processor to execute the method of the embodiment of the present disclosure.

[0082] The embodiments of this specification also provide Figure 4 The structural diagram of the electronic device shown in FIG. 1 is a schematic diagram of the electronic device shown in FIG. 1 . At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for other transactions. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-mentioned voice activity detection method.

[0083] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0084] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0085] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0086] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0087] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0088] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0089] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0090] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0091] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0092] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0093] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0094] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0095] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0096] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0097] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0098] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0099] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A text acquisition method, comprising: Receive a voice signal from a target user; Calling a speech recognition model dedicated to the target user to process the speech signal to obtain a recognized text; In response to a signal from the target user refusing to confirm the recognized text, calling a text error correction process to process the recognized text to obtain a corrected text; Determine the text input by the target user according to the error-corrected text; and continue to train the speech recognition model according to the speech signal and the error-corrected text.

2. The method according to claim 1, calling a text error correction process to process the recognized text to obtain a corrected text, comprising: Displaying the recognized text to the target user in an editable manner; According to the editing operation of the target user, a text after error correction is determined.

3. The method according to claim 1, calling a text error correction process to process the recognized text to obtain a corrected text, comprising: The automatic error correction model is called to process the recognized text to obtain a corrected text.

4. The method according to claim 1, calling a text error correction process to process the recognized text to obtain a corrected text, comprising: Calling the automatic error correction model to process the recognized text to obtain an error-corrected intermediate text; In response to a signal from the target user refusing to confirm the error-correcting intermediate text, presenting the error-correcting intermediate text to the target user in an editable manner; According to the editing operation of the target user, a text after error correction is determined.

5. The method of claim 4, further comprising: In response to a signal from the target user confirming the error-corrected intermediate text, a post-error-corrected text is determined based on the error-corrected intermediate text.

6. The method of claim 4, further comprising: The automatic error correction model is continued to be trained according to the recognized text and the corrected text.

7. The method of claim 6, wherein: The target user and at least one other user jointly use the automatic error correction model.

8. The method of claim 1, further comprising: In response to a signal from the target user confirming the recognized text, the text input by the target user is determined based on the recognized text.

9. The method of claim 8, wherein determining the text input by the target user according to the error-corrected text comprises: Calling a language model to optimize the text structure of the corrected text and / or to optimize the text expression of the corrected text; Using the optimized error-corrected text as the text input by the target user; Determining the text input by the target user according to the recognized text includes: Calling a language model to optimize the text structure of the recognized text and / or to optimize the text expression of the recognized text; The optimized recognized text is used as the text input by the target user.

10. The method according to any one of claims 1 to 9, wherein: The method is applied to a model training data annotation scenario, the target user is an annotator, and the text input by the target user is annotation information.

11. A text acquisition device, comprising: A receiving module receives a voice signal from a target user; A recognition module, calling a speech recognition model dedicated to the target user, processing the speech signal, and obtaining a recognition text; An error correction module, in response to a signal from the target user refusing to confirm the recognized text, calls a text error correction process to process the recognized text to obtain a corrected text; A processing module determines the text input by the target user according to the error-corrected text; and continues to train the speech recognition model according to the speech signal and the error-corrected text.

12. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

13. An electronic device, characterized in that: include: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the method as claimed in any one of claims 1 to 10.

14. A computer program product having at least one instruction stored thereon, characterized in that: When the at least one instruction is executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

Images