Speech recognition self-learning method for man-machine speech dialogue, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a voice recognition self-learning method and system for a man-machine voice dialogue, electronic equipment and a storage medium, and the method comprises the steps: recording a man-machine dialogue log which at least comprises a user identifier, an input text, an input type and a dialogue task completion state; classifying the logs according to the user identifier, screening an input text of which the task completion state is successful, and generating a user-level corpus; training a user exclusive language model based on the user-level corpus; and in the speech recognition process, synchronously loading the system-level language model and the user exclusive language model for recognition processing. According to the method, online logs and user-level-oriented training corpora are analyzed and submitted to the self-learning system, self-learning and user-level language model updating are carried out, self-learning is achieved, and the voice recognition effect of the dialogue system is continuously improved.

Description

Technical Field

[0001] This invention belongs to the field of human-computer interaction technology, and particularly relates to a speech recognition self-learning method, speech recognition self-learning system, electronic device and storage medium for human-computer voice dialogue. Background Technology

[0002] Related technologies include: 1. A human-computer dialogue method and system based on a self-learning dialogue model, which mainly utilizes a self-learning dialogue model to train a role information memory network using predefined role information. When predicting responses, the role information memory network is first used to generate role information with the highest relevance to the current context, and then the role information is input into the dialogue generation network. By appropriately incorporating role information, it better conforms to the dialogue habits between users, making the responses generated by the model more personalized and diverse. 2. An intent recognition method and system with self-learning capabilities, which performs self-learning based on the feature distribution of historical intent recognition data in the actual use scenarios of intelligent voice dialogue systems and terminals, dynamically adjusting the intent recognition strategy to improve the accuracy of intent recognition in the use scenarios. 3. A speech recognition method and device, which obtains representation information matching the target speech from a pre-constructed memory, wherein the memory stores a large number of sample speaker representation results and / or sample speaking environment representation results, and then the target speech can be recognized based on the representation information obtained from the memory. 4. A personalized voice interaction method and system, which, in response to a user's voice interaction command, collects user behavior data; wherein the behavior data includes at least voice data; inputs the behavior data into a preset feature recognition model for feature recognition; obtains the user's personalized features and the total score of the behavior data based on the preset score value corresponding to each user behavior; extracts text data from the voice data based on a preset text generation model; generates a response text matching the total score of the behavior data based on the text data and personalized features, and based on the feature tags and score tags of each text in a preset corpus; and converts the response text into audio data.

[0003] The inventors discovered that existing related technologies all improve the user experience of voice dialogue systems from dimensions such as personalization and self-learning. They all optimize the response text of the dialogue system to make it more diverse and personalized, or improve the intent recognition effect of intelligent voice dialogue systems through self-learning methods, and improve the overall speech recognition effect and efficiency through the representation results of a large number of samples of speech and environment.

[0004] The inventors also discovered that existing related technologies suffer from the following problems: 1) user-level personalized speech recognition issues; 2) the lack of a user-level continuous self-learning process; and 3) the self-learning process does not consider more diverse inputs. Because of these three reasons—the absence of user-level speech recognition, the lack of a user-level self-learning process, and the failure to consider more diverse inputs—these deficiencies have long existed in voice dialogue system applications, failing to provide users with a better experience. Summary of the Invention

[0005] This invention provides a speech recognition self-learning method for human-computer speech dialogue, as well as a speech recognition self-learning system, electronic device, and storage medium, to at least solve one of the above-mentioned technical problems.

[0006] In a first aspect, embodiments of the present invention provide a speech recognition self-learning method for human-computer voice dialogue, comprising: recording human-computer dialogue logs, wherein the logs include at least a user identifier, input text, input type, and dialogue task completion status; classifying the logs according to the user identifier, filtering input texts with a task completion status of success, and generating a user-level corpus; A user-specific language model is trained based on the user-level corpus; during speech recognition, the system-level language model and the user-specific language model are simultaneously loaded for recognition processing.

[0007] Secondly, embodiments of the present invention also provide a speech recognition self-learning system for the speech recognition self-learning method for human-computer speech dialogue described in the first aspect, comprising: a human-computer dialogue subsystem, including an automatic speech recognition module, a natural language understanding module, a natural language generation module, a speech synthesis module, and a dialogue management module; a log subsystem, used to store dialogue records associated with user identifiers and task completion status; a self-learning subsystem, including a log analysis module, which periodically extracts user input text with a successful task completion status from the log subsystem and generates a corpus dataset according to user identifiers; a language model training module, which trains a user-specific language model based on the corpus dataset; and the automatic speech recognition module is configured to simultaneously call the system-level language model and the user-specific language model to perform speech recognition.

[0008] Thirdly, embodiments of the present invention also provide a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of the speech recognition self-learning method for human-computer speech dialogue according to any embodiment of the present invention.

[0009] Fourthly, embodiments of the present invention also provide an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method described in the first aspect.

[0010] Fifthly, embodiments of the present invention also provide a storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method described in the first aspect.

[0011] In the method of this application embodiment, online logs are analyzed to provide user-level training corpus, which is then submitted to a self-learning system for self-learning and updating of the user-level language model, thereby achieving self-learning and continuously improving the speech recognition performance of the dialogue system. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart of an embodiment of a speech recognition self-learning method for human-computer speech dialogue provided by an embodiment of the present invention; Figure 2 A schematic diagram of a speech recognition self-learning optimization method and system structure for human-computer speech dialogue provided by the present invention; Figure 3 This is a schematic diagram of system-level language model processing provided by the present invention; Figure 4 A schematic diagram illustrating the system-level language model combined with the user-level language model processing provided by this invention; Figure 5 The flowchart illustrates the implementation of the speech recognition self-learning method for human-computer speech dialogue provided by this invention. Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0016] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, elements, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0017] In this invention, terms such as "module," "device," and "system" refer to relevant entities applied to a computer, such as hardware, combinations of hardware and software, software, or software in execution. More specifically, for example, an element can be, but is not limited to, a process running on a processor, a processor, an object, an executable element, an execution thread, a program, and / or a computer. Furthermore, an application program or script running on a server, and the server itself, can also be an element. One or more elements may be in an execution process and / or thread, and elements may be localized on a single computer and / or distributed across two or more computers, and may be run on various computer-readable media. Elements can also communicate via local and / or remote processes based on signals having one or more data packets, for example, signals from data interacting with another element in a local system, a distributed system, and / or interacting with other systems via signals over a network on the Internet.

[0018] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising" or "including" include not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0019] This invention provides a self-learning speech recognition method for human-computer voice dialogue, which can be applied to electronic devices. The electronic device can be a computer, server, or other electronic product, etc., and this invention does not limit it to these.

[0020] Please refer to Figure 1 This illustrates a speech recognition self-learning method for human-computer speech dialogue provided by an embodiment of the present invention.

[0021] like Figure 1 As shown, in step 101, a human-computer dialogue log is recorded, which includes at least the user identifier, input text, input type, and dialogue task completion status. In step 102, if the logs are classified according to the user identifier, input texts with a task completion status of success are filtered and a user-level corpus is generated; In step 103, a user-specific language model is trained based on the user-level corpus; In step 104, during the speech recognition process, the system-level language model and the user-specific language model are loaded simultaneously for recognition processing.

[0022] In this embodiment, for step 101, a human-computer dialogue log is recorded. The log includes at least the user identifier, input text, input type, and dialogue task completion status. The online human-computer dialogue system serves user interaction and question-and-answer, and the relevant logs can record detailed logs of each dialogue or interaction. To achieve a user-level speech recognition self-learning system, the log system not only records detailed interaction logs but also records the specific user ID and the completion status of each dialogue task. An example of the log is shown in the table below: the first column is the user ID; the second column is the text input into the dialogue system, including the result after speech recognition or the text input result on the interface; the third column is the type of text input; the fourth column indicates the completion status of the dialogue scenario task, such as: in a voice navigation scenario, completing navigation is "completed," and not completing the navigation task is "not completed."

[0023] Next, in step 102, if the logs are categorized according to the user identifier, input texts with a "task completed successfully" status are filtered to generate a user-level corpus. The log analysis module periodically collects relevant user-level corpus from historical records in the dialogue logs. First, it categorizes logs based on the user ID, grouping logs with the same user ID together; second, it only filters input texts indicating "task completed"; finally, it forms user-level corpus data. The user-level corpus generated after this module's modification is shown in the previous table below.

[0024] Then, for step 103, a user-specific language model is trained based on the user-level corpus. The ASR (Automatic Speech Recognition) language model training module periodically trains a user-level language model based on each user's corpus. The speech recognition module consists of two important models: the acoustic model is responsible for recognizing the input audio into basic "pronunciation units," while the language model converts the basic "pronunciation units" output by the acoustic model into text, playing a role in distinguishing homophones and checking language standardization.

[0025] Finally, in step 104, during the speech recognition process, both the system-level language model and the user-specific language model are simultaneously loaded for recognition processing. The online ASR module not only uses the system-level language model but also loads a user-level language model for speech recognition, achieving personalized speech recognition at the user level.

[0026] The method in this application embodiment analyzes online logs, provides user-level training corpora, submits them to a self-learning system for self-learning and updating of the user-level language model, thereby achieving self-learning and continuously improving the speech recognition performance of the dialogue system.

[0027] It should be noted that before adopting the solution of this invention, the online ASR module used the same system-level language model for speech recognition. All users' audio was recognized using the same system-level language model, which made it impossible to achieve self-learning and improve the speech recognition effect of the dialogue system.

[0028] In some optional embodiments, the logging step includes real-time recording of speech recognition results and interface text input results, where the input type distinguishes between speech input and text input. The log can record detailed logs of each dialogue or interaction. The logging system not only records detailed interaction logs but also the specific user ID and the completion status of each dialogue task. The log includes the text input into the dialogue system, as well as the results of speech recognition or the text input results on the interface.

[0029] The training cycle for the user-specific language model is a predetermined time interval, and the training process is independent of the system-level language model update process. By analyzing online logs, user-level training corpora are submitted to the self-learning system for self-learning and updating of the user-level language model, thus achieving self-learning.

[0030] This application also provides a speech recognition self-learning system in one embodiment. This speech recognition self-learning system is used in the aforementioned speech recognition self-learning method for human-computer speech dialogue.

[0031] The human-computer dialogue subsystem includes an automatic speech recognition module, a natural language understanding module, a natural language generation module, a speech synthesis module, and a dialogue management module. The log subsystem is used to store conversation records associated with user identifiers and task completion status; The self-learning subsystem includes a log analysis module, which periodically extracts user input text with a successful task completion status from the log subsystem and generates a corpus dataset according to the user identifier. The language model training module trains user-specific language models based on corpus datasets. The automatic speech recognition module is configured to simultaneously call both the system-level language model and the user-specific language model to perform speech recognition.

[0032] The logging subsystem records the completion status of dialogue tasks, including navigation completion and incomplete indicators in navigation scenarios.

[0033] When the log analysis module generates corpus, it excludes records with text input.

[0034] The language model training module retains the acoustic model parameters unchanged during training, and only updates the language model parameters.

[0035] The user-specific language model is used to distinguish homophones and verify language standardization. Its output weights are dynamically weighted and fused with the output weights of the system-level language model.

[0036] Online human-computer dialogue systems consist of several basic technical modules, including: ASR (Automatic Speech Recognition), NLU (Natural Language Understanding), NLG (Natural Language Generation), TTS (Text To Speech), and DM (Dialogue Management).

[0037] The online human-computer dialogue system serves users' interactive question and answer; the log system not only records detailed interaction logs, but also records the specific user ID and the status of each dialogue task completion.

[0038] The user-level ASR language model self-learning system consists of two core steps: 1) The log analysis module periodically collects relevant user-level corpus from historical dialogue logs. First, it categorizes logs by user ID, grouping logs from the same user ID together; second, it only filters out input text indicating "task completed"; finally, it generates user-level corpus data.

[0039] 2) The ASR language model training module will periodically train a user-level language model based on each user's corpus. The speech recognition module consists of two important models: the acoustic model is responsible for recognizing the input audio into basic "pronunciation units", while the language model converts the basic "pronunciation units" output by the acoustic model into text, which plays a role in distinguishing homophones and checking language standardization.

[0040] The inventors found the following main problems with existing similar technologies: 1) User-level personalized speech recognition issues.

[0041] In human-computer dialogue systems, speech recognition results are closely related to each user; even audio with similar pronunciations can yield different speech recognition results in different user scenarios. Here are two examples: Scenario 1: When asking "How's the weather in Hengshan?", the meaning of the word "hengshan" can lead to different speech recognition results depending on the user context. User A might want to ask "How's the weather in Hengshan?", while user B might want to ask "How's the weather in Hengshan?". Similarly, the similar pronunciation "sanqi" should be recognized as "three phases" in a financial credit card installment scenario, but as "three-seven" in a traditional Chinese medicine pharmacy scenario. The speech recognition result must be closely related to the user's current context.

[0042] Scenario 2: It's not just homophones; the recognition results for many words with similar pronunciations are also closely related to the user. For example, if a Beijing user, C, lives in "Juyuan," and says "daohang qu juyuan" during voice navigation on their car or mobile device, it's highly likely the system won't recognize it correctly.

[0043] 2) There is no continuous self-learning process at the user level.

[0044] When users find that the speech recognition result is not what they want, they usually rephrase their speech and engage in a second human-computer interaction to correct the recognition result.

[0045] In scenario 1, user B actually meant to ask "How is the weather in Hengshan?", but the voice dialogue system recognized it as "How is the weather in Hengshan?". User B might ask again "How is the weather in Hengshan, Shanxi?" to correct the voice recognition result in the voice dialogue system.

[0046] Without a continuous self-learning process at the user level, if user B asks "How is the weather in Hengshan?" again, the speech recognition will still interpret it as "How is the weather in Hengshan?" and will not be able to learn from the previous conversation history with the user at the user level.

[0047] 3) The self-learning process does not take into account more diverse inputs.

[0048] In addition to voice input, voice dialogue systems also offer other input methods during terminal interaction, such as direct text input.

[0049] In scenario 2, when user C was using voice navigation, they found that "Juyuan" was not correctly recognized. In fact, they could enter the Chinese characters "Juyuan" into the input bar of the map interface that popped up on the in-vehicle terminal or mobile terminal using the input method, and then proceed with navigation.

[0050] If the self-learning system does not consider other inputs in the relevant scenario, and the next time user B asks "dao hang qujuyuan", the voice dialogue system still cannot learn from the previous user's historical dialogue records and thus correctly perform speech recognition.

[0051] Therefore, the inventors discovered that the above three reasons—lack of user-level speech recognition, lack of a user-level self-learning process, and failure to consider more diverse inputs—have long plagued voice dialogue systems. These deficiencies exist in mobile voice dialogue systems, in-vehicle voice dialogue systems, and smart speaker dialogue systems, among others.

[0052] It should be noted that the method in this application can improve the accuracy of speech recognition by providing real-time scene information or contextual hot words when calling the speech recognition system. However, this method follows a different approach; the approach in this application focuses on improving recognition performance by training a user-specific language model using a user-level corpus.

[0053] Please refer to Figure 2 This application proposes a speech recognition self-learning optimization method for human-computer speech dialogue and a schematic diagram of the system architecture. The system architecture diagram is shown in Figure 2, which consists of three subsystems: a human-computer dialogue system, a log system, and a user-level ASR language model self-learning system.

[0054] Please refer to this again. Figure 3 , Figure 3 This is a schematic diagram of system-level language model processing.

[0055] Before adopting the solution of this invention, online ASR modules performed speech recognition using the same system-level language model. All users' audio was processed using the same system-level language model for speech recognition, such as... Figure 3 As shown.

[0056] Please refer to this again. Figure 4 , Figure 4 This diagram illustrates the process of combining a system-level language model with a user-level language model.

[0057] After adopting the solution of this invention, the online ASR module not only uses a system-level language model, but also loads a user-level language model to perform speech recognition, thereby realizing personalized speech recognition at the user level.

[0058] Please refer to Figure 5 The document presents a flowchart illustrating the implementation of the speech recognition self-learning method for human-computer speech dialogue according to the present invention.

[0059] like Figure 5 As shown, taking user C's voice navigation scenario as an example, the implementation flowchart of the method of this application is illustrated. In subsequent embodiments, "day xx" is not limited to a specific time but can refer to a certain stage, which will not be elaborated upon here.

[0060] On the first day, when User C used the map navigation, they first used voice input. The voice dialogue system recognized the command as "Navigate to the theater." Because this was not the result User C wanted, they did not proceed with the navigation. Instead, they directly entered "Juyuan" into the interface text to complete the navigation. These interactions were all logged in the log system.

[0061] The next day, when User C used map navigation, they simply said, "Navigate to Malianwa Chrysanthemum Garden." The voice dialogue system correctly recognized this and completed the navigation task. The interaction log was also recorded. At this point, the ASR self-learning system followed up with relevant corpora (including: "chrysanthemum garden," "Malianwa Chrysanthemum Garden") to train a language model, updating the user-level language model for that user through self-learning.

[0062] On the third day, when User C was using map navigation, he directly said "Navigate to Juyuan". This time, the voice dialogue system loaded User C's user-level language model, correctly recognized it, and completed the navigation task.

[0063] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0064] In some embodiments, the present invention provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions, which can be read and executed by electronic devices (including but not limited to computers, servers, or network devices, etc.) to perform any of the above-described speech recognition self-learning methods for human-computer voice dialogue of the present invention.

[0065] In some embodiments, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform any of the above-described speech recognition self-learning methods for human-computer speech dialogue.

[0066] In some embodiments, the present invention also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a speech recognition self-learning method for human-computer speech dialogue.

[0067] Figure 6 This is a schematic diagram of the hardware structure of an electronic device that performs a speech recognition self-learning method for human-computer speech dialogue, according to another embodiment of this application. Figure 6 As shown, the device includes: One or more processors 610 and memory 620, Figure 6 Take the 610 processor as an example.

[0068] The device for implementing a speech recognition self-learning method for human-computer speech dialogue may further include an input device 630 and an output device 640.

[0069] The processor 610, memory 620, input device 630, and output device 640 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0070] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the speech recognition self-learning method for human-computer voice dialogue in the embodiments of this application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby realizing the speech recognition self-learning method for human-computer voice dialogue in the above-described method embodiments.

[0071] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created during the use of the speech recognition self-learning method device for human-computer voice dialogue. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories can be connected to the speech recognition self-learning method device for human-computer voice dialogue via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0072] The input device 630 can receive input numerical or character information, and generate signals related to user settings and function control of the speech recognition self-learning method device for human-computer voice dialogue. The output device 640 may include a display device such as a display screen.

[0073] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, they execute the speech recognition self-learning method for human-computer speech dialogue in any of the above method embodiments.

[0074] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0075] The electronic devices in this application embodiments exist in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0076] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include: PDAs, MIDs, and UMPCs, etc.

[0077] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players, handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.

[0078] (4) Other airborne electronic devices with data interaction capabilities, such as vehicle-mounted systems installed on vehicles.

[0079] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A self-learning speech recognition method for human-computer speech dialogue, comprising: Record human-computer dialogue logs, which include at least the user identifier, input text, input type, and dialogue task completion status; The logs are categorized according to the user identifier, and input text with a successful task completion status is filtered to generate a user-level corpus. Train a user-specific language model based on the aforementioned user-level corpus; During speech recognition, both the system-level language model and the user-specific language model are loaded simultaneously for recognition processing.

2. The method according to claim 1, characterized in that, The logging step includes real-time recording of speech recognition results and interface text input results, wherein the input type distinguishes between speech input and text input.

3. The method according to claim 1, characterized in that, The training cycle of the user-specific language model is a predetermined time interval, and the training process is independent of the system-level language model update process.

4. A speech recognition self-learning system, characterized in that, include: The human-computer dialogue subsystem includes an automatic speech recognition module, a natural language understanding module, a natural language generation module, a speech synthesis module, and a dialogue management module. The log subsystem is used to store conversation records associated with user identifiers and task completion status; The self-learning subsystem includes a log analysis module, which periodically extracts user input text with a task completion status of success from the log subsystem and generates a corpus dataset according to user identifiers. The language model training module trains a user-specific language model based on the corpus dataset. The automatic speech recognition module is configured to simultaneously call the system-level language model and the user-specific language model to perform speech recognition.

5. The system according to claim 4, characterized in that, The log subsystem records the completion status of dialogue tasks, including navigation completion and incomplete indicators in navigation scenarios.

6. The system according to claim 4, characterized in that, When the log analysis module generates the corpus, it excludes records with text input.

7. The system according to claim 4, characterized in that: The language model training module retains the acoustic model parameters unchanged during training, and only updates the language model parameters.

8. The system according to any one of claims 4 to 7, characterized in that, The user-specific language model is used to distinguish homophones and verify language standardization. Its output weights are dynamically weighted and fused with the output weights of the system-level language model.

9. An electronic device comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the method of claim 2 or 3.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method of claim 2 or 3.

Images