Voice instruction execution method and system, terminal and medium

Abstract

The invention discloses a voice instruction execution method and system, a terminal and a medium, and the method comprises the steps: receiving a voice instruction sent by a user, and determining a user intention corresponding to the voice instruction based on a preset basic large model; determining a target application program based on the user intention, and determining API calling information corresponding to the user intention based on a preset MOE model; and calling a system interface corresponding to the API calling information through a target application program, and executing the voice instruction based on the system interface. According to the method, the voice instruction sent by the user is converted into the operation which can be executed by the system, the user intention is accurately recognized by utilizing the preset basic large model and the MOE model, and the corresponding API calling information is determined, so that efficient analysis and execution of the voice instruction are realized, decoupling between application programs is ensured, independent iterative updating is supported, and the user experience is improved. The system maintenance cost is reduced, and the research and development efficiency and the user experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a voice command execution method and system. Background Art

[0002] In the field of smart home, integrating AI voice function into the system is widely used as a common practice to improve user experience. This integration aims to simplify the user's operation process through voice interaction. However, this method has low flexibility, and because the implementation of voice function is highly dependent on the close collaboration between the application layer and the system layer, and the large number of interfaces involved, it is vulnerable to system compatibility issues and difficult to upgrade or maintain independently without destroying the existing ecosystem. In addition, the design concept of the operating system and its applications of smart TVs follows the architectural principles of traditional embedded systems, realizing the linkage between functions through interface calls between the application layer and the system layer, and relying on the services provided by the underlying system software. This design does not specifically consider the dynamic environment unique to intelligent voice control. When the system or application is updated, the normal use of the intelligent voice function may be affected by version incompatibility and other issues.

[0003] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a voice command execution method and system in response to the above-mentioned defects of the prior art, aiming to solve the problems of low flexibility of AI voice functions, poor system compatibility and difficulty in independent upgrading and maintenance.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for executing a voice command, wherein the method comprises:

[0007] Receive a voice command issued by a user, and determine the user intention corresponding to the voice command based on a preset basic large model;

[0008] Based on the user intent, determine the target application, and determine the API call information corresponding to the user intent based on a preset MOE model;

[0009] The system interface corresponding to the API call information is called through the target application, and the voice instruction is executed based on the system interface.

[0010] In one implementation, the receiving of a voice command issued by a user and determining the user intention corresponding to the voice command based on a preset basic large model also includes:

[0011] Generate a natural language text describing the functional characteristics of the system program itself according to a preset natural language description template, wherein the preset natural language description template includes a role definition of the application program, an application identifier, a skill description, a skill triggering condition, and a corresponding response instruction;

[0012] The natural language text is sent as registration information to a preset basic large model, and the basic large model reports the functional characteristics of the system program according to the received registration information.

[0013] In one implementation, the receiving of a voice command issued by a user and determining the user intention corresponding to the voice command based on a preset basic large model includes:

[0014] Receive voice commands from users and convert the voice commands into corresponding text content;

[0015] Based on the text content, searching the reported registration information for skill triggering conditions in a natural language text that matches the text content through the basic large model;

[0016] Based on the skill triggering condition, a response instruction corresponding to the skill triggering condition is obtained to obtain the user intention.

[0017] In one implementation, the determining the target application based on the user intention includes:

[0018] Based on user intent, obtain system program application identifier;

[0019] Based on the system program application identifier, a target application program is determined.

[0020] In one implementation, determining the API call information corresponding to the user intention based on a preset MOE model includes:

[0021] The target application calls the AI ​​SDK to upload the user intention to the control model of the preset MOE model;

[0022] The control model is used to forward the user intent to the corresponding sub-model in the MOE model through the gated network and the domain name to understand the intent and obtain the API call information.

[0023] In one implementation, forwarding the user intent to a sub-model in a corresponding MOE model for intent understanding to obtain API call information includes:

[0024] Forwarding the user intent to the corresponding sub-model in the MOE model for intent understanding, and obtaining a function corresponding to the user intent and parameters of the function;

[0025] If the parameter has a specified value, then obtaining API call information according to the function and the specified value of the parameter;

[0026] If the parameter does not have a specified value, the API call information is obtained according to the function and the preset value of the parameter.

[0027] In one implementation, calling the system interface corresponding to the API call information through the target application and executing the voice instruction based on the system interface includes:

[0028] Based on the API call information, using the target application to call the system interface through the AI ​​SDK;

[0029] Based on the system interface, the voice command is executed.

[0030] In a second aspect, an embodiment of the present invention further provides a voice command execution system, wherein the system comprises:

[0031] A user intention determination module, used to receive a voice command issued by a user and determine the user intention corresponding to the voice command based on a preset basic large model;

[0032] A target application and API call information determination module, used to determine the target application based on the user's intention, and determine the API call information corresponding to the user's intention based on a preset MOE model;

[0033] The voice instruction execution module is used to call the system interface corresponding to the API call information through the target application program, and execute the voice instruction based on the system interface.

[0034] In a third aspect, an embodiment of the present invention further provides a terminal, wherein the terminal includes a memory, a processor, and a voice command execution program stored in the memory and executable on the processor, and when the processor executes the voice command execution program, the steps of the voice command execution method of any one of the above-mentioned schemes are implemented.

[0035] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein a voice instruction execution program is stored on the computer-readable storage medium, and when the voice instruction execution program is executed by a processor, the steps of the voice instruction execution method described in any one of the above schemes are implemented.

[0036] Beneficial effects: The present invention provides a method for executing voice commands. Compared with the prior art, the present invention first receives the voice commands issued by the user, and determines the user intent corresponding to the voice commands based on a preset basic large model, thereby ensuring the accuracy of voice recognition and the depth of natural language understanding. Next, based on the user intent, the target application is determined, and the API call information corresponding to the user intent is determined based on the preset MOE model. Through the intelligent matching mechanism, the system's support capabilities and flexibility for diverse user needs are enhanced. Then, through the target application, the system interface corresponding to the API call information is called, and the voice command is executed based on the system interface, thereby achieving seamless connection between voice control and the application, and ensuring the efficiency and reliability of command execution. The voice command execution method proposed in the present invention integrates advanced speech recognition and natural language understanding technologies, and combines the MOE model to intelligently match user intentions with the API calls of applications, thereby realizing efficient conversion from voice command input to system operation execution. It not only improves the flexibility and system compatibility of the AI ​​voice function, but also enables the voice function to be upgraded and maintained independently of other parts of the system. This improves the response speed and accuracy of AI voice assistants in devices such as smart TVs, while also simplifying the subsequent technical maintenance and function upgrade process, thereby achieving the purpose of optimizing user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention provides a flow chart of a specific implementation of the voice command execution method provided in an embodiment of the present invention.

[0038] Figure 2 The present invention provides an overall flow chart of the voice command execution method in practical application.

[0039] Figure 3 The present invention provides a flow chart of a preferred embodiment of the voice command execution method.

[0040] Figure 4 This is a functional block diagram of a voice command execution system provided by an embodiment of the present invention.

[0041] Figure 5 This is a block diagram of the internal structure principle of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] In the field of smart home, the integration of AI voice function aims to simplify the user's operation process through voice interaction, thereby improving the user experience. However, this method has low flexibility, and because the implementation of voice function is highly dependent on the close collaboration between the device layer and the system layer, and the large number of interfaces involved, it is susceptible to system compatibility issues. In addition, the design of smart home devices usually follows the traditional embedded system architecture principles, and realizes the linkage between functions through interface calls between the device layer and the system layer. This design does not specifically consider the dynamic environment required by intelligent voice control, so when the system or application is updated, the normal use of the intelligent voice function may be affected by version incompatibility and other issues.

[0044] In order to solve the above problems, this embodiment provides a voice command execution method. During specific implementation, this embodiment first receives the voice command issued by the user, and determines the user intention corresponding to the voice command based on the preset basic large model, thereby ensuring the accuracy of voice recognition and the depth of natural language understanding. Next, based on the user intention, the target application is determined, and the API call information corresponding to the user intention is determined based on the preset MOE model. Through the intelligent matching mechanism, the system's support capabilities and flexibility for diverse user needs are enhanced. Then, through the target application, the system interface corresponding to the API call information is called, and the voice command is executed based on the system interface, thereby achieving seamless connection between voice control and the application, and ensuring the efficiency and reliability of command execution. The voice command execution method proposed in the present invention integrates advanced speech recognition and natural language understanding technologies, and combines the MOE model to intelligently match user intentions with the API calls of applications, thereby realizing efficient conversion from voice command input to system operation execution. It not only improves the flexibility and system compatibility of the AI ​​voice function, but also enables the voice function to be upgraded and maintained independently of other parts of the system. This improves the response speed and accuracy of AI voice assistants in devices such as smart TVs, while also simplifying the subsequent technical maintenance and function upgrade process, thereby achieving the purpose of optimizing user experience.

[0045] For example, suppose a user is watching a smart TV and wants to adjust the volume of the TV through voice control. At this time, the user can use the method proposed by the present invention. The user can adjust the volume of the TV by saying voice commands such as "increase the volume" and "lower the volume". First, receive the voice command issued by the user, such as "increase the volume". The reception of the voice command makes the operation more intuitive and convenient. The user does not need to find a remote control or manual operation, but only needs to express the demand verbally, which greatly improves the user experience. Next, the user intention corresponding to the voice command is determined based on the preset basic large model, that is, the user wants to increase the volume of the TV. The key to this step is to be able to process natural language input. Even if the user's expression method is different, it can understand its core needs, which enhances the intelligence and adaptability of the system. Then, based on the parsed user intention, that is, to increase the volume, determine which application to call, and determine the API call information corresponding to the user intention based on the preset MOE model, that is, the API call information for adjusting the volume, which ensures the accurate execution of the command, reduces the possibility of misoperation, and also provides flexibility for the expansion of the system. Finally, through the target application, call the system interface corresponding to this API call information, and execute the voice command to increase the volume based on the system interface. In this way, users can directly control smart TVs through voice commands, which simplifies the human-computer interaction process and enables users to express their needs in a more natural way to obtain instant responses. The basic large model is used in combination with the MOE model to not only accurately capture user intentions, but also accurately match the corresponding API call information, ensuring the accuracy of operations and the efficiency of the system. This design improves the user experience and makes smart home control more intuitive, convenient and user-friendly.

[0046] This embodiment provides a voice command execution method, which can be applied to smart terminals, such as Figure 1 As shown in , the specific steps include:

[0047] Step S100: receiving a voice command issued by a user, and determining the user intention corresponding to the voice command based on a preset basic large model.

[0048] In this embodiment, first, when a user issues a command by voice, the voice command issued by the user is received, and the user intention corresponding to the voice command is determined based on the preset basic large model. This step is the starting point of the entire voice command execution process, ensuring the accuracy and pertinence of subsequent processing.

[0049] Specifically, the step S100 includes the following steps:

[0050] Step S101: receiving a voice command issued by a user, and converting the voice command into corresponding text content;

[0051] Step S102: Based on the text content, searching the reported registration information for skill triggering conditions in a natural language text that matches the text content through the basic large model;

[0052] Step S103: Based on the skill triggering condition, obtain a response instruction corresponding to the skill triggering condition to obtain the user intention.

[0053] In one implementation, Figure 2As shown, S1. Send the natural language text as registration information to the basic large model for reporting. The specific implementation process is as follows: receive the voice command issued by the user, and determine the user intention corresponding to the voice command based on the preset basic large model, which also includes: first, generate a natural language text describing the functional characteristics of the system program itself according to the preset natural language description template, the preset natural language description template includes the role definition of the application, application identifier, skill description, skill trigger conditions and corresponding response instructions, standardizes the expression of the system program function, so that the functions of different programs can be uniformly and clearly conveyed to the basic large model, and enhances the configurability and maintainability of the system. Then, the natural language text is sent as registration information to the preset basic large model, and the preset basic large model reports the functional characteristics of the system program according to the received registration information, so that the subsequent basic large model can perform intent recognition and allocation according to the registration information. For example, first, a natural language text describing the functional characteristics of the system program itself is generated according to the preset natural language description template (prompt template). The prompt template defines the information format for the system program to report registration to the preset basic large model. All running system programs need to send a natural language to the voice APP (such as AI TV housekeeper) to report what functions they have according to the prompt template when they are run for the first time or for the first time after the updated version. The prompt template rule example is as follows: Role: You are the system setting APP, responsible for receiving and completing the TV system control commands; APPID (application identifier): TV.System.Setting; Skill 1: Open the page specified by the TV; Trigger condition of skill 1: When the user's input {input} content involves system settings, network settings, network switches, Bluetooth settings, TV information, storage information, etc.; The corresponding response instruction for the trigger condition of skill 1: open + {input}. Skill 2: Turn on or off the specified function of the TV; Trigger condition of skill 2: When the user's input {input} involves issues related to camera, AI speaker, far-field voice, eye protection, etc., the corresponding response instruction of the trigger condition of skill 2: Please judge the output according to the user's intention, turn on or off the corresponding function. Example: User input input: "I want to take a photo", reply: "Turn on the camera". Skill 3: TV brightness adjustment - Trigger condition of skill 3: When the user's input {input} involves issues of light and brightness, the corresponding response instruction of the trigger condition of skill 3: Please judge and reply to the user to adjust the brightness of the TV. For example: User: {The light is a bit dim}, reply: {Turn up the screen brightness}.Skill 4: TV mode adjustment; Triggering condition of skill 4: When the user's input {input} involves the problem of TV mode adjustment, the reference content of TV mode is as follows: {Image mode: custom, sports, cinema, natural, standard, dynamic, AI picture quality, color temperature setting: bright, standard, natural, sound mode: AI sound quality, standard, music, theater, news, sports, image ratio: smart, movie, subtitle, panoramic, full clear restoration, 16:9, 4:3; Response instruction corresponding to the triggering condition of skill 4: Please reply to the user according to the above reference content to set the specified mode. After receiving the registration prompt content of the system program, the registration prompt content is sent to the preset basic large model for registration and filing of the skill domain. The basic large model will then perform intent recognition and allocation based on the registered content. This step allows the basic large model to understand and remember the specific capabilities and operating scope of each application, so that it can more accurately match the user's intent with the system program function in subsequent processing, thereby improving the accuracy and efficiency of voice command execution.

[0054] In one implementation, Figure 2 As shown, S2 determines the user intention corresponding to the voice instruction based on the basic large model. The specific implementation process is as follows: When a user issues an instruction through voice, first, the voice instruction issued by the user is received, and the voice instruction is converted into the corresponding text content, and the unstructured voice input is converted into structured text data for subsequent processing. Then, based on the text content, the skill triggering conditions in the natural language text that matches the text content are searched in the reported registration information through the basic large model, and the pre-reported application function information is utilized to ensure that the user instructions can be accurately understood and classified. Then, based on the skill triggering condition, the response instruction corresponding to the skill triggering condition is obtained to obtain the user intention, and the user demand is directly matched with the operation of the application, so as to achieve efficient and accurate instruction execution, and improve the user experience and service quality as a whole. For example, Figure 3 As shown, the user said: "The light is a little dim", and the user's voice command was converted into text content, that is, the text content is: The light is a little dim. Then, based on the text content "The light is a little dim", the basic large model is used to search for the skill triggering conditions in the natural language text that matches "The light is a little dim" in the reported prompt content, and the skill triggering condition is: The light is a little dim. Next, based on the skill triggering condition: The light is a little dim, the response instruction corresponding to the skill triggering condition of the light is a little dim is obtained, that is, turn up the screen brightness, and the user's intention is to turn up the screen brightness. This embodiment converts voice instructions into text, uses the basic large model to match the skill triggering conditions, and obtains the response instructions to parse the user's intention, thereby achieving rapid understanding and accurate response to user needs, and improving the naturalness of interaction and user experience.

[0055] Step S200: determine the target application based on the user intent, and determine the API call information corresponding to the user intent based on a preset MOE model.

[0056] In this embodiment, after obtaining the user's intention, first, the target application is determined based on the user's intention, and the application that is most suitable for processing the request is quickly located according to the user's specific needs, avoiding the wrong call of irrelevant applications and improving the response speed and accuracy of the system; then, the API call information corresponding to the user's intention is determined based on the preset MOE model. The MOE model can intelligently select the most appropriate expert (i.e., API call information) according to different scenarios and contexts, thereby achieving more accurate function calls, which not only optimizes resource usage, but also ensures the high quality of processing results, and improves the intelligence level of the system and user experience. This step ensures the accuracy and efficiency of voice command execution through precise target application selection and intelligent matching of API call information.

[0057] Specifically, the step S200 includes the following steps:

[0058] Step S201: based on the user's intention, obtain a system program application identifier;

[0059] Step S202: determining a target application program based on the system program application identifier;

[0060] Step S203: calling the AI ​​SDK through the target application to upload the user intention to the control model of the preset MOE model;

[0061] Step S204: Use the control model to forward the user intent to the corresponding sub-model in the MOE model through the gated network and the domain name to understand the intent and obtain API call information.

[0062] In one implementation, Figure 2As shown, S3. After obtaining the user intent, determine the target program based on the user intent. The specific implementation process is as follows: After obtaining the user intent, first, obtain the system program application identifier based on the user intent. Then, based on the system program application identifier, determine the target application to ensure that subsequent processing points to the correct application and avoid incorrect calls. Then, S4. Upload the user intent to the MOE model. The specific implementation process is as follows: Call the AI ​​SDK through the target application to upload the user intent to the preset MOE model control model. The AI ​​SDK is a text serialization tool used to serialize natural language text, which realizes the effective docking of user intent and MOE model, and provides a basis for subsequent intent understanding. Finally, S5. Get API call information. The specific implementation process is as follows: First, based on the user intent, get the domain name through the control model. Then, use the control model to select the corresponding sub-model in the MOE model according to the domain name through the gating network. Then, the control model is used to forward the user intent to the corresponding sub-model in the MOE model through the gated network for intent understanding, and obtain the function corresponding to the user intent and the parameters of the function; if the parameter has a specified value, the API call information is obtained according to the function and the specified value of the parameter; if the parameter does not have a value, the API call information is obtained according to the preset value of the function and the parameter. The domain name is used to identify different sub-models in the MOE model to help the control model determine which specific sub-model the received user intent should be forwarded to for processing. For example, the system control domain is used to identify the system control model, the film and television search domain is used to identify the film and television search model, and the chat encyclopedia domain is used to identify the chat encyclopedia model. Different sub-models in the MOE model match functions in different fields respectively. For example, the system control model is used to match functions in the system control domain, the film and television search model is used to match functions in the film and television search domain, and the chat encyclopedia model is used to match functions in the chat encyclopedia domain. For example, Figure 3As shown, first, based on the obtained user intent: increase the screen brightness, based on the reported prompt content, the system program application identifier is obtained as APPID: TV.System.Setting. Next, based on APPID: TV.System.Setting, based on the reported prompt content, the target application is determined to be the system settings APP. Then, the system settings APP calls the AI ​​SDK to upload the user intent (increase the screen brightness) to the control model of the preset MOE model. Finally, based on the user's intention, the domain name is obtained as the system control domain. The gated network is used to select the corresponding sub-model (i.e., the system control model) in the MOE model according to the domain name (i.e., the system control domain). The control model is used to forward the user's intention (increasing the screen brightness) to the system control model (i.e., the sub-model corresponding to the system control domain) through the gated network for intention understanding, and the function setBrightness corresponding to the user's intention and the parameter b of the function are obtained. If the parameter b has a specified value, such as b=50, the API call information is setBrightness(50), which means increasing the brightness by 50. If the parameter b does not have a specified value, the preset value of the parameter b is obtained. For example, the preset value of b is 10. At this time, the API call information is setBrightness(10), which means increasing the brightness by 10. The intention transmission of the system program is all completed by natural language, without the need for API call bundling, and does not involve system interface-related dependencies, thus realizing decoupled operations. No matter how the system program is iterated and upgraded, it will not affect the use of voice functions and the distribution of system instructions. When the subsequent system program individuals need to be updated and upgraded, they only need to complete the functional characteristics report based on the preset natural language description template after updating themselves. The preset basic large model and MOE model do not need to follow the system program for version upgrades, which realizes the decoupling between applications, reduces the difficulty of system version iteration and upgrade costs, and greatly improves the R&D efficiency and convenience of system maintenance. This embodiment obtains the application identifier, determines the target application, and then uses the AI ​​SDK and MOE model to upload and understand the intent, and finally generates API call information with appropriate parameters. This series of steps not only improves the response speed and accuracy of the system, but also enhances the ability to handle complex user requests, thereby significantly improving the user experience.

[0063] Step S300: calling the system interface corresponding to the API call information through the target application, and executing the voice instruction based on the system interface.

[0064] In this embodiment, first, the system interface corresponding to the API call information is called through the target application; then, the voice command is executed based on the system interface, ensuring that the user's intention can be accurately converted into system operation, so that the voice command can be successfully completed.

[0065] Specifically, step S300 includes the following steps:

[0066] Step S301: Based on the API call information, use the target application to call the system interface through the AI ​​SDK;

[0067] Step S302: Execute the voice command based on the system interface.

[0068] In one implementation, Figure 2 As shown, S6. Interface call, the specific implementation process is as follows: First, based on the API call information, use the target application to call the system interface through the AI ​​SDK. The AI ​​SDK provides a standardized interface call method so that different applications can consistently communicate with the underlying system. This not only simplifies the design and development of applications, but also enhances the stability and security of the system. Then, based on the system interface, the voice command is executed to ensure that the user's intention can be accurately converted into actual system behavior, improve the accuracy and reliability of the execution, and also ensure the response speed of the system and enhance the user experience. For example, Figure 3 As shown, based on the API call information setBrightness(10), the system settings APP calls the system DriverBase interface through the AI ​​SDK, and the TV brightness is set based on the DriverBase interface. The entire process of this example is carried out through standardized interfaces and predefined processes, which reduces the possibility of errors and makes the system response more timely and accurate.

[0069] In summary, this embodiment first receives the voice command issued by the user, and determines the user intent corresponding to the voice command based on the preset basic large model, ensuring the accuracy of voice recognition and the depth of natural language understanding. Then, based on the user intent, the target application is determined, and the API call information corresponding to the user intent is determined based on the preset MOE model. Through the intelligent matching mechanism, the system's support capability and flexibility for diversified user needs are enhanced. Then, through the target application, the system interface corresponding to the API call information is called, and the voice command is executed based on the system interface, realizing seamless connection between voice control and the application, and ensuring the efficiency and reliability of command execution. The voice command execution method proposed in the present invention integrates advanced voice recognition and natural language understanding technology, combines the MOE model to intelligently match the user intent and the API call of the application, and realizes the efficient conversion from voice command input to system operation execution, which not only improves the flexibility and system compatibility of the AI ​​voice function, but also enables the voice function to be upgraded and maintained independently of other parts of the system, thereby improving the response speed and accuracy of the AI ​​voice assistant in devices such as smart TVs, while also simplifying the subsequent technical maintenance and function upgrade process, achieving the purpose of optimizing user experience.

[0070] like Figure 4 As shown in , this embodiment also provides a voice instruction execution system, which includes: a user intention determination module 10, a target application and API call information determination module 20 and a voice instruction execution module 30. Specifically, the user intention determination module 10 is used to receive a voice instruction issued by a user, and determine the user intention corresponding to the voice instruction based on a preset basic large model. The target application and API call information determination module 20 is used to determine the target application based on the user intention, and determine the API call information corresponding to the user intention based on a preset MOE model. The voice instruction execution module 30 is used to call the system interface corresponding to the API call information through the target application, and execute the voice instruction based on the system interface.

[0071] In one implementation, the system further includes:

[0072] A natural language text generation unit, used to generate a natural language text describing the functional characteristics of the system program itself according to a preset natural language description template, wherein the preset natural language description template includes a role definition of the application program, an application identifier, a skill description, a skill triggering condition, and a corresponding response instruction;

[0073] The reporting unit is used to send the natural language text as registration information to the basic large model, and the basic large model reports the functional characteristics of the system program according to the received registration information.

[0074] In one implementation, the user intention determination module 10 includes:

[0075] A voice command receiving and converting unit, used to receive a voice command issued by a user and convert the voice command into corresponding text content;

[0076] A skill trigger condition acquisition unit, configured to search for a skill trigger condition in a natural language text matching the text content in the reported registration information through the basic large model based on the text content;

[0077] The user intention acquisition unit is used to acquire a response instruction corresponding to the skill triggering condition based on the skill triggering condition to obtain the user intention.

[0078] In one implementation, the target application and API call information determination module 20 includes:

[0079] An application identifier acquisition unit, used to acquire a system program application identifier based on a user's intention;

[0080] a target application determining unit, configured to determine a target application based on the system program application identifier;

[0081] A user intention uploading unit, configured to upload the user intention to a control model of a preset MOE model by calling the AI ​​SDK through the target application;

[0082] The API call information acquisition unit is used to use the control model to forward the user intention to the sub-model in the corresponding MOE model through the gated network and the domain name to understand the intention and obtain the API call information.

[0083] In one implementation, forwarding the user intent to a sub-model in a corresponding MOE model for intent understanding to obtain API call information includes:

[0084] A function and a parameter acquisition unit for the function, used to forward the user intention to a sub-model in the corresponding MOE model for intention understanding, and obtain a function corresponding to the user intention and the parameters of the function;

[0085] The first subunit for obtaining API call information is used to obtain API call information according to the function and the specified value of the parameter if the parameter has a specified value;

[0086] The second subunit for obtaining API call information is used to obtain API call information according to the function and the preset value of the parameter if the parameter does not have a value.

[0087] In one implementation, the voice instruction execution module 30 includes:

[0088] A system interface calling unit, configured to call the system interface using the target application through the AI ​​SDK based on the API calling information;

[0089] A voice instruction execution unit is used to execute the voice instruction based on the system interface.

[0090] The working principles of each module in the voice command execution system of this embodiment are the same as the principles of each step in the above method embodiment, and will not be repeated here.

[0091] Based on the above embodiment, the present invention further provides a terminal, the principle block diagram of the terminal can be as follows: Figure 5 The terminal may include one or more processors 100 ( Figure 5 Only one is shown in the figure), a memory 101 and a computer program 102 stored in the memory 101 and executable on one or more processors 100, for example, a voice command execution program. When one or more processors 100 execute the computer program 102, the various steps in the voice command execution method embodiment can be implemented. Alternatively, when one or more processors 100 execute the computer program 102, the functions of the various modules / units in the voice command execution method embodiment can be implemented, which is not limited here.

[0092] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0093] In one embodiment, the memory 101 may be an internal storage unit of an electronic device, such as a hard disk or memory of the electronic device. The memory 101 may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the memory 101 may also include both an internal storage unit of the electronic device and an external storage device. The memory 101 is used to store computer programs and other programs and data required by the terminal. The memory 101 may also be used to temporarily store data that has been output or is to be output.

[0094] Those skilled in the art will understand that Figure 5 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0095] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, operating database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double operational data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for executing a voice command, characterized in that: The method comprises: Receive a voice command issued by a user, and determine the user intention corresponding to the voice command based on a preset basic large model; Based on the user intent, determine the target application, and determine the API call information corresponding to the user intent based on a preset MOE model; The system interface corresponding to the API call information is called through the target application, and the voice instruction is executed based on the system interface.

2. The voice command execution method according to claim 1, characterized in that: The receiving of the voice command issued by the user and determining the user intention corresponding to the voice command based on a preset basic large model also includes: Generate a natural language text describing the functional characteristics of the system program itself according to a preset natural language description template, wherein the preset natural language description template includes a role definition of the application program, an application identifier, a skill description, a skill triggering condition, and a corresponding response instruction; The natural language text is sent as registration information to a preset basic large model, and the basic large model reports the functional characteristics of the system program according to the received registration information.

3. The voice command execution method according to claim 2, characterized in that: The receiving a voice command issued by a user and determining the user intention corresponding to the voice command based on a preset basic large model includes: Receive voice commands from users and convert the voice commands into corresponding text content; Based on the text content, searching the reported registration information for skill triggering conditions in a natural language text that matches the text content through the basic large model; Based on the skill triggering condition, a response instruction corresponding to the skill triggering condition is obtained to obtain the user intention.

4. The voice command execution method according to claim 3, characterized in that: The determining of the target application based on the user intent includes: Based on user intent, obtain system program application identifier; Based on the system program application identifier, a target application program is determined.

5. The voice command execution method according to claim 4, characterized in that: The determining the API call information corresponding to the user intention based on the preset MOE model includes: The target application calls the AI ​​SDK to upload the user intention to the control model of the preset MOE model; The control model is used to forward the user intent to the corresponding sub-model in the MOE model through the gated network and the domain name to understand the intent and obtain the API call information.

6. The voice command execution method according to claim 5, characterized in that: The user intent is forwarded to the corresponding sub-model in the MOE model for intent understanding to obtain API call information, including: Forwarding the user intent to the corresponding sub-model in the MOE model for intent understanding, and obtaining a function corresponding to the user intent and parameters of the function; If the parameter has a specified value, then obtaining API call information according to the function and the specified value of the parameter; If the parameter does not have a specified value, the API call information is obtained according to the function and the preset value of the parameter.

7. The voice command execution method according to claim 6, characterized in that: The step of calling the system interface corresponding to the API call information through the target application and executing the voice command based on the system interface includes: Based on the API call information, using the target application to call the system interface through the AI ​​SDK; Based on the system interface, the voice command is executed.

8. A voice command execution system, characterized in that: The system comprises: A user intention determination module, used to receive a voice command issued by a user and determine the user intention corresponding to the voice command based on a preset basic large model; A target application and API call information determination module, used to determine the target application based on the user's intention, and determine the API call information corresponding to the user's intention based on a preset MOE model; The voice instruction execution module is used to call the system interface corresponding to the API call information through the target application program, and execute the voice instruction based on the system interface.

9. A terminal, characterized in that: The terminal includes a memory, a processor, and a voice instruction execution program stored in the memory and executable on the processor. When the processor executes the voice instruction execution program, the steps of the voice instruction execution method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a voice instruction execution program, and when the voice instruction execution program is executed by the processor, the steps of the voice instruction execution method according to any one of claims 1 to 7 are implemented.

Images