Server, terminal device and user intention recognition method

Abstract

The embodiment of the invention provides a server, terminal equipment and a user intention recognition method. The terminal device can collect a user voice instruction and send the user voice instruction to the server. And the server converts the user voice instruction into an instruction text and obtains accent words in the instruction text. And the server performs semantic analysis on the instruction text to obtain an initial user intention, scores the initial user intention through the accent words, and obtains a target user intention corresponding to the instruction text based on the intention score. And calling the business service processing accent word and the instruction text corresponding to the target user intention to obtain a business processing result, and sending the business processing result to the terminal equipment. The intention of the user is accurately recognized through the accent words of the user, so that the voice instruction of the user is responded, and the user experience is improved.

Description

Technical Field

[0001] This application relates to the field of terminal device technology, and in particular to a terminal device and a method for recognizing user intent. Background Technology

[0002] Terminal devices can be electronic devices with sound acquisition capabilities, such as smart TVs, mobile phones, smart speakers, computers, and robots. Taking smart TVs as an example, smart TVs are based on Internet application technology, have an open operating system and chip, and have a voice recognition module, enabling two-way human-computer interaction. They integrate multiple functions such as audio-visual entertainment and data to meet diverse and personalized user needs.

[0003] Terminal devices can have voice interaction capabilities, allowing users to control them via voice commands. In implementing this function, a voice recognition module typically converts the user's voice commands into text. A semantic analysis module then performs lexical, syntactic, and semantic analysis on this text to determine the user's intent. Finally, the terminal device can execute corresponding actions based on the user's intent, thus achieving voice interaction.

[0004] When users interact via voice input, they may emphasize a particular angle or word, hoping the device will respond accordingly. However, if the device only recognizes the intent based on the converted text, without considering the stressed words in the audio, the accuracy of intent recognition will be low, resulting in a poor user experience. Summary of the Invention

[0005] This application provides a server, a terminal device, and a user intent recognition method to solve the problem in related technologies where the terminal device only performs intent recognition through the text converted from speech, without taking into account the stressed word information contained in the speech audio, resulting in low accuracy of intent recognition and a poor user experience.

[0006] In a first aspect, some embodiments of this application provide a server, including a communication module and a control module. The communication module is configured to establish a communication connection with a terminal device; the control module is configured to:

[0007] In response to a user voice command sent by a terminal device, the user voice command is converted into command text;

[0008] Obtain the stressed words in the instruction text;

[0009] Perform semantic analysis on the instruction text to obtain at least one initial user intent;

[0010] Each initial user intent is scored based on the stressed word to obtain an intent score corresponding to the initial user intent;

[0011] Based on the intent score, the target user intent corresponding to the instruction text is selected from the initial user intent;

[0012] Input the stressed word and the instruction text into the business service corresponding to the target user's intent;

[0013] The business processing results output by the business service are sent to the terminal device.

[0014] The above technical solution offers the following advantages: the server can obtain the accented words in the command, thereby scoring the initial user intent based on the accented words, and obtaining the target user intent based on the intent score. By accurately identifying the user's intent through the user's accented words, the server can respond to the user's voice commands, thus improving the user experience.

[0015] In some embodiments, the control module is configured to retrieve stressed words from the instruction text, specifically as follows:

[0016] The instruction text is segmented to obtain multiple text segments;

[0017] Obtain the audio features corresponding to the text segmentation;

[0018] The audio features are subjected to stress detection to obtain the stress confidence of each of the text segments;

[0019] The text segmentation words whose stress confidence reaches a preset confidence threshold are marked as stressed words.

[0020] The above technical solution has the following beneficial effects: by detecting the audio features of text segmentation, the stress probability of each segment can be accurately analyzed.

[0021] In some embodiments, the control module performs the acquisition of audio features corresponding to the text segmentation, specifically configured as follows:

[0022] The Mel frequency cepstral coefficients are obtained based on the audio corresponding to the user's voice command.

[0023] Based on the Mel frequency cepstral coefficients, the instruction text is subjected to forced alignment processing to obtain the audio timing of the text segmentation in the speech audio.

[0024] Based on the audio time, determine the audio data corresponding to the text segmentation;

[0025] Based on the audio data, extract the tone features and energy features corresponding to the text word segmentation.

[0026] The above technical solution has the following beneficial effects: by detecting Mel frequency cepstral coefficients, text segmentation and audio sequences are aligned, thereby accurately obtaining the audio features of each segmented word.

[0027] In some embodiments, the control module performs the operation of obtaining Mel-frequency cepstral coefficients based on the voice audio corresponding to the user's voice command, specifically configured as follows:

[0028] The audio corresponding to the user's voice command is pre-emphasized to obtain a first audio signal;

[0029] The first audio signal is subjected to frame segmentation and windowing processing to obtain the second audio signal;

[0030] Perform Fourier transform processing on the second audio signal to convert the second audio signal into a speech spectrum;

[0031] The harmonics of the speech spectrum are eliminated to obtain the Mel spectrum;

[0032] The Mel frequency cepstral coefficients are obtained by performing a logarithmic operation and discrete cosine transform on the Mel spectrum. In some embodiments, the control module performs semantic analysis on the instruction text, specifically configured as follows:

[0033] The instruction text is input into a preset user intent recognition model so that the user intent recognition model outputs user intent analysis results; the user intent analysis results include at least one initial user intent and the intent confidence of the initial user intent.

[0034] The above technical solution has the following beneficial effects: it can detect multiple possible initial user intentions and the probability of each initial user intention through a user intention recognition model.

[0035] In some embodiments, the control module performs a scoring of each initial user intent based on the stressed word, specifically configured as follows:

[0036] Detect stressed words in the instruction text;

[0037] If the instruction text contains accented words, then the step of scoring the at least one initial user intent based on the accented words is performed;

[0038] If the instruction text does not contain stressed words, then the initial user intent with the highest intent confidence is set as the target user intent corresponding to the instruction text.

[0039] The above technical solution has the following beneficial effects: when no stressed words are detected, the user's intent can be determined directly from the text; when stressed words are detected, the user's intent can be determined by combining the text and audio.

[0040] In some embodiments, the control module performs a scoring of each initial user intent based on the stressed word, specifically configured as follows:

[0041] The target keywords corresponding to the initial user intent are obtained based on a preset keyword matching table; the keyword matching table is used to store the correspondence between keywords and initial user intents.

[0042] Detect stressed words in the instruction text based on the target keywords;

[0043] If the target keyword contains the stressed word, then the initial user intent is scored based on the stressed word;

[0044] If the target keyword does not contain stressed words, then the intent confidence of the initial user intent is detected, and the intent confidence is set as the intent score of the initial user intent.

[0045] The above technical solution has the following beneficial effects: by detecting the matching degree between stressed words and initial user intent through the keyword matching table corresponding to each initial user intent, the initial user intent can be accurately scored.

[0046] In some embodiments, the control module performs a scoring of the initial user intent based on the stressed word, specifically configured as follows:

[0047] Detect the stress confidence of the text segmentation;

[0048] Accent weights are calculated based on the accent confidence score with the highest numerical value.

[0049] The intent confidence of the initial user intent is weighted based on the accent weights to obtain the intent score of the initial user intent.

[0050] Secondly, some embodiments of this application provide a terminal device, including a sound collector, a communication device, and a controller. The sound collector is configured to collect audio; the communication device is configured to establish a communication connection with a server; and the controller is configured to:

[0051] Acquire user voice commands collected by the sound acquisition device;

[0052] The user's voice command is sent to the server, which then performs accent detection on the command text corresponding to the user's voice command to obtain accented words in the command text, and performs semantic analysis on the command text to obtain at least one initial user intent; and, based on the accented words, obtains an intent score corresponding to the initial user intent, and filters the at least one initial user intent based on the intent score to obtain the target user intent corresponding to the command text; obtains the service type corresponding to the target user intent, and obtains the service processing result through the service corresponding to the service type;

[0053] Receive the business processing results sent by the server.

[0054] Thirdly, some embodiments of this application provide a user intent recognition method applied to the aforementioned server, the method comprising:

[0055] In response to a user voice command sent by a terminal device, the user voice command is converted into command text;

[0056] Obtain the stressed words in the instruction text;

[0057] Perform semantic analysis on the instruction text to obtain at least one initial user intent;

[0058] Each initial user intent is scored based on the stressed word to obtain an intent score corresponding to the initial user intent;

[0059] Based on the intent score, the target user intent corresponding to the instruction text is selected from the initial user intent;

[0060] Input the stressed word and the instruction text into the business service corresponding to the target user's intent;

[0061] The business processing results output by the business service are sent to the terminal device. Attached Figure Description

[0062] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 The system architecture of a voice interaction scenario for a terminal device is shown in some embodiments;

[0064] Figure 2 Hardware configuration block diagrams of terminal devices in some embodiments are shown;

[0065] Figure 3The software configuration diagram of the terminal device in some embodiments is shown;

[0066] Figure 4 A schematic diagram of the voice interaction network architecture in some embodiments is shown;

[0067] Figure 5 A schematic diagram of the application list interface is shown in some embodiments;

[0068] Figure 6 The interaction flowcharts between the terminal device and the server in some embodiments are shown;

[0069] Figure 7 Schematic diagrams of the interaction between terminal devices and servers in some embodiments are shown;

[0070] Figure 8 A flowchart illustrating the process of obtaining Mel frequency cepstral coefficients is shown in some embodiments;

[0071] Figure 9 A schematic diagram of the terminal device displaying the system settings interface is shown in some embodiments;

[0072] Figure 10 Schematic diagrams of the device control interface in some embodiments are shown;

[0073] Figure 11 A schematic diagram of the main control device settings interface is shown in some embodiments. Detailed Implementation

[0074] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0075] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0076] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0077] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0078] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0079] In some embodiments, "terminal device" can refer to any device with voice interaction capabilities. Figure 1 The system architecture of a voice interaction scenario for a terminal device is shown in some embodiments. For example... Figure 1 As shown, the voice interaction scenario system may include a server 100 and a terminal device 200. An exemplary terminal device 200 may include a smart TV 200a, a mobile device 200b, and a smart speaker 200c.

[0080] In this embodiment, server 100 and terminal device 200 communicate via various communication methods. Terminal device 200 can communicate via local area network (LAN), wireless local area network (WLAN), and other networks. Server 100 can provide various content and interactive features to terminal device 200. For example, terminal device 200 and server 100 can send and receive information, and receive software updates.

[0081] In some embodiments, server 100 may be a server that provides various services, such as a backend server that supports audio data collected by terminal device 200. The backend server may analyze and process the received audio and other data, and feed back the processing results (e.g., endpoint information) to the terminal device. Server 100 may be a server cluster or multiple server clusters, and may include one or more types of servers.

[0082] In some embodiments, the terminal device 200 can be hardware or software. When the terminal device 200 is hardware, it can be various electronic devices with sound acquisition capabilities, including but not limited to smart speakers, smartphones, televisions, tablets, e-book readers, smartwatches, media players, computers, AI devices, robots, smart vehicles, etc. When the terminal devices 200, 201, and 202 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., for providing sound acquisition services) or as a single software program or software module. No specific limitations are made here.

[0083] Figure 2Hardware configuration block diagrams of terminal devices in some embodiments are shown. For example... Figure 2 As shown, the terminal device 200 may include at least one of the following: a tuner 210, a communication device 220, a detector 230, a device interface 240, a controller 250, a display 260, an audio output device 270, a memory, a power supply, and a user input interface 280.

[0084] In some embodiments, detector 230 is used to acquire signals from the external environment or to interact with the outside world. For example, detector 230 includes a light receiver, a sensor for acquiring ambient light intensity; or, detector 230 includes an image acquisition device, such as a camera, which can be used to acquire external environmental scenes, user attributes, or user interaction gestures; or, detector 230 includes a sound acquisition device, such as a microphone, for receiving external sounds.

[0085] In some embodiments, the sound acquisition device can be a microphone, also known as a "microphone" or "voice transducer," which can be used to receive the user's voice and convert the sound signal into an electrical signal. The terminal device 200 can be equipped with at least one microphone. In other embodiments, the terminal device 200 can be equipped with two microphones, which, in addition to acquiring sound signals, can also perform noise reduction. In still other embodiments, the terminal device 200 can also be equipped with three, four, or more microphones, enabling sound signal acquisition, noise reduction, sound source identification, and directional recording functions, etc.

[0086] In some embodiments, the microphone may be built into the terminal device 200, or the microphone may be connected to the terminal device 200 via wired or wireless means. Of course, the embodiments of this application do not limit the location of the microphone on the terminal device 200. Alternatively, the terminal device 200 may not include a microphone, that is, the microphone is not provided in the terminal device 200. The terminal device 200 may connect an external microphone (also called a microphone) via an interface (such as a USB interface 130). This external microphone can be fixed to the terminal device 200 using an external fastener (such as a camera bracket with a clip).

[0087] In some embodiments, the display 260 includes display function components for presenting images and driving components for driving image display. The display 260 is used to receive and display image signals output from the controller 250. For example, the display 260 can be used to display video content, image content, menu control interface components, and user control UI interfaces, etc.

[0088] In some embodiments, the display 260 may be a liquid crystal display, an OLED display, or a projection display, and may also be a projection device and a projection screen.

[0089] In some embodiments, the communication device 220 is a component used to communicate with external devices or the server 100 according to various communication protocol types. The terminal device 200 may have multiple communication devices 220 depending on the supported communication methods. For example, when the terminal device 200 supports wireless network communication, it may have a communication device 220 with WiFi functionality. When the terminal device 200 supports Bluetooth connection communication, it needs to have a communication device 220 with Bluetooth functionality.

[0090] The communication device 220 enables the terminal device 200 to communicate with external devices or the server 100 via wireless or wired connections. Wired connections utilize data cables, interfaces, or other components to connect the terminal device 200 to external devices. Wireless connections utilize wireless signals or wireless networks. The terminal device 200 can establish a direct connection with external devices or indirectly through gateways, routers, or other connection devices.

[0091] In some embodiments, the controller 250 may include at least one of a central processing unit, a video processor, an audio processor, a graphics processor, and a power processor, and a first to an nth interface for input / output. The controller 250 controls the operation of the terminal device 200 and responds to user operations through various software control programs stored in memory. The controller 250 controls the overall operation of the terminal device 200.

[0092] In some embodiments, the controller 250 and the tuner 210 may be located in different separate devices, that is, the tuner 210 may also be located in an external device of the main device where the controller 250 is located, such as an external set-top box.

[0093] In some embodiments, a user can input user commands through a graphical user interface (GUI) displayed on a display 260, and the user input interface receives user input commands through the graphical user interface (GUI).

[0094] In some embodiments, the audio output device 270 can be the built-in speaker of the terminal device 200 or an external audio output device connected to the terminal device 200. For the external audio output device connected to the terminal device 200, the terminal device 200 may also be provided with an external audio output terminal, through which the audio output device can be connected to the terminal device 200 to output sound from the terminal device 200.

[0095] In some embodiments, the user input interface 280 can be used to receive instructions from user input.

[0096] In some embodiments, to enable user interaction, the terminal device 200 may run an operating system. The operating system is a computer program used to manage and control the hardware and software resources of the terminal device 200. The operating system can control the terminal device 200 to provide a user interface; for example, the operating system can directly control the terminal device 200 to provide a user interface, or it can provide a user interface by running applications. The operating system also allows users to interact with the terminal device 200.

[0097] It should be noted that the operating system can be a native operating system based on a specific operating platform, a third-party operating system that is deeply customized based on a specific operating platform, or an independent operating system specifically developed for the terminal device 200.

[0098] Operating systems can be divided into different modules or levels based on the functions they implement. Figure 3 A software configuration diagram of the terminal device 200 in some embodiments is shown. For example... Figure 3 As shown, in some embodiments, the system is divided into four layers, from top to bottom: the Applications layer (referred to as the "Application Layer"), the Application Framework layer (referred to as the "Framework Layer"), the System Library layer, and the Kernel layer.

[0099] In some embodiments, the application layer provides services and interfaces for applications, enabling the terminal device 200 to run applications and interact with the user based on the applications. The application layer may contain at least one application, which may be a built-in Windows program, system settings program, or clock program of the operating system; or it may be an application developed by a third-party developer. In specific implementations, the application packages in the application layer are not limited to the examples above.

[0100] The framework layer provides application programming interfaces (APIs) and a programming framework for applications. The application framework layer includes predefined functions. It acts as a central processing unit, determining the actions taken by applications within the application layer. Through the API, applications can access system resources and obtain system services during execution.

[0101] like Figure 3As shown, the application framework layer in this embodiment includes a view system, managers, and content providers. The view system designs and implements the application's interface and interactions, and includes lists, grids, text boxes, and buttons. The managers include at least one of the following modules: an activity manager for interacting with all running activities in the system; a location manager for providing system services or applications with access to system location services; a package manager for retrieving various information related to application packages currently installed on the device; a notification manager for controlling the display and clearing of notification messages; and a window manager for managing icons, windows, toolbars, wallpapers, and desktop widgets on the user interface.

[0102] In some embodiments, the Activity Manager manages the lifecycle of individual applications and common navigation and back functions, such as controlling application exit, opening, and back actions. The Window Manager manages all window programs, such as obtaining the screen size, determining if a status bar is present, locking the screen, capturing the screen, and controlling changes to the display window, such as shrinking the display window, shaking the display, or distorting the display.

[0103] In some embodiments, the system runtime library layer can provide support for the framework layer. When the framework layer is used, the operating system runs the instruction library contained in the system runtime library layer, such as the C / C++ instruction library, to implement the functions to be performed by the framework layer.

[0104] In some embodiments, the kernel layer is a functional layer situated between the hardware and software of the terminal device 200. The kernel layer can implement functions such as hardware abstraction, multitasking, and memory management. For example, ... Figure 3 As shown, hardware drivers can be configured in the kernel layer. The drivers included in the kernel layer can be at least one of the following: audio driver, display driver, Bluetooth driver, camera driver, WIFI driver, USB driver, HDMI driver, sensor driver (such as fingerprint sensor, temperature sensor, pressure sensor, etc.), and power driver, etc.

[0105] It should be noted that the above examples are merely a simple division of operating system functions and do not limit the specific form of the operating system of the terminal device 200 in this application embodiment. Depending on the functions of the terminal device 200, the type of operating system, and other factors, the number of layers and the specific type of the operating system may take other forms.

[0106] In some embodiments, the terminal device 200 has a voice interaction function, and the user can control the terminal device 200 by voice.

[0107] Figure 4 Schematic diagrams of voice interaction network architectures in some embodiments are shown. For example... Figure 4 As shown, terminal device 200 can receive input information, such as user audio data, through detector 230, and output the processing result of that information. Terminal device 200 deploys an Automatic Speech Recognition (ASR) service to recognize audio as text. Terminal device 200 performs semantic understanding on the recognized text based on the deployed Natural Language Understanding (NLU) service. For example, in response to voice interaction commands, terminal device 200 can perform noise reduction and feature extraction on the source audio data of the voice interaction commands, where noise reduction may include steps such as removing echoes and removing ambient noise; then, it performs natural language understanding on the recognized candidate text and associated context information, parsing the text into structured, machine-readable information, and obtaining executable intents through information such as business domain, intent, and slots in the readable information to determine the intent confidence score. The semantic understanding module selects one or more candidate executable intents based on the determined intent confidence score.

[0108] In some embodiments, the terminal device 200 can send a query instruction to the corresponding business management module according to a determined executable intent to obtain the query results provided by the business service and to execute the actions required by the user's final request, and then feed back the device execution instruction corresponding to the query results. For example, the terminal device 200 can send a corresponding query instruction to the Dialog Management (DM) configured on the server 100 according to the selected executable intent to obtain the query results of the voice interaction instructions.

[0109] To respond to voice interaction commands, in some embodiments, the terminal device 200 can also generate language text through a deployed Natural Language Generation (NLG) service and respond to the voice interaction commands using the generated language text. Specifically, the response types can be categorized as conversational, task-oriented, knowledge-based question-answering, and recommendation-based. In conversational responses, NLG performs intent recognition and sentiment analysis based on context to generate open-ended responses. In task-oriented responses, learned strategies are used to generate responses, typically including clarifying needs, guiding the user, asking questions, confirming, and closing remarks. In knowledge-based question-answering responses, the required knowledge (knowledge, entities, fragments, etc.) is generated based on question type identification and classification, information retrieval, or text matching. In recommendation-based systems, results are matched and candidate recommendations are ranked based on the user's input history before generating recommended content for the user.

[0110] In some embodiments, the terminal device 200 can synthesize corresponding response audio data based on the language text generated by NLG through a Text-to-Speech (TTS) service, and control the audio output device 270 to play the response audio data to output the sound formed by the response audio data to the user. For example, the generated language text can be in the form of a text string, and the terminal device 200 can convert the text string into playable response audio data.

[0111] It should be noted that, Figure 5 The architecture shown is merely an example and is not intended to limit the scope of protection of this application. Other architectures can also be used to achieve similar functions in the embodiments of this application. For example, all or part of the above process can be completed by the terminal device 200, which will not be elaborated here.

[0112] Terminal device 200 can have various functions, such as browsing web pages, playing media, playing games, and screen mirroring, thereby providing users with a wide variety of services. Users can control terminal device 200 to launch a specific application, thereby enabling the corresponding function.

[0113] In some embodiments, when the user control terminal device 200 is powered on, the controller 250 can control the display 260 to show the user interface. The user interface may be a specific target image, such as various media resources obtained from a network signal source, including videos, pictures, and other content. The user interface may also be some pages obtained by the terminal device 200, including web pages provided by a browser, some UI interfaces of the terminal device 200, such as the system main interface, and various control interfaces of applications installed on the terminal device 200.

[0114] Users can control terminal device 200 to display an application list interface, which contains the applications installed on terminal device 200, i.e., the functions supported by terminal device 200. Upon receiving an application launch command input by the user based on the application list interface, terminal device 200 runs the corresponding application, enabling terminal device 200 to support the application's functions. It should be noted that the applications installed on terminal device 200 can be system applications or third-party applications. Users control terminal device 200 to implement the corresponding functions of an application by launching it.

[0115] Figure 5 The diagram illustrates an application list interface in some embodiments. For example... Figure 5 As shown, the application list interface includes three controls: "Cable TV," "Online Media Assets," and "Browser." The terminal device 200 detects user touch operations to implement different functions. In response to a user's touch operation on the "Cable TV" control, the terminal device 200 can display a media asset channel interface, allowing the user to select a media asset channel for playback. In response to a user's touch operation on the "Online Media Assets" control, the terminal device 200 can launch an online media asset application to play various online media assets. In response to a user's touch operation on the "Browser" control, the terminal device 200 launches a browser application, allowing the user to access various web pages provided on the internet. Web pages can display various elements, such as text, images, and videos, for the user to view.

[0116] In some embodiments, users can control the terminal device 200 via voice input commands, such as searching for media assets by voice. The terminal device 200 can convert the user's voice command into command text, perform semantic analysis to identify the user's intent, and then provide a corresponding response and perform the appropriate operation based on the intent. For example, a user can voice input "I want to watch movie A". The terminal device 200 can perform semantic analysis on the user's voice command, extracting entity segmentation or tag information from the voice command through word segmentation and other methods, including determining that the user's intent is a media asset search, and recognizing the entity information of "movie A". The terminal device 200 can execute the media asset search process for "movie A", including displaying media asset recommendation results for "movie A". Furthermore, the terminal device 200 can respond to the user via voice, with a response such as "Movie A has been recommended for you, please watch it".

[0117] It's important to note that when users interact via voice input, they may emphasize a particular aspect, stressing a specific word in their speech to elicit a response from the device. For example, a user's voice command might be "Please search for TV dramas for children." The stressed word could be either "children" or "TV dramas." Different stressed words indicate different emphases on the device's response. The voice / audio command "Please search for TV dramas for [children]" emphasizes "children," meaning the user prefers the device to search for children's content. Conversely, the voice / audio command "Please search for [TV dramas] for children" emphasizes "TV dramas," meaning the user prefers the device to search for TV drama content.

[0118] Therefore, stressed words in a user's audio can characterize the type of intent the user wants the terminal device to respond to. However, terminal devices only perform intent recognition based on the text converted from speech, without considering the stressed word information contained in the audio, resulting in low accuracy of intent recognition and a poor user experience.

[0119] To address the aforementioned issues, terminal devices can be equipped with user intent recognition functionality. The terminal device can collect user voice commands and send them to a server. The server converts the user's voice commands into command text and extracts the stressed words from the command text. The server performs semantic analysis on the command text to obtain the initial user intent, scores the initial user intent based on stressed words, and obtains the target user intent corresponding to the command text based on the intent score. The server then invokes the business service corresponding to the target user intent to process the stressed words and command text, obtains the business processing result, and sends the business processing result back to the terminal device. By accurately identifying user intent through stressed words, the server can respond to user voice commands, thereby improving the user experience.

[0120] Figure 6 The following are flowcharts illustrating the interaction between the terminal device and the server in some embodiments, including the following steps:

[0121] S601. In response to a user voice command sent by a terminal device, the server converts the user voice command into command text.

[0122] S602. The server retrieves the stressed words in the instruction text;

[0123] S603. The server performs semantic analysis on the instruction text to obtain at least one initial user intent;

[0124] S604. The server scores each initial user intent based on the stressed word to obtain the intent score corresponding to the initial user intent.

[0125] S605. The server filters the target user intent corresponding to the instruction text from the initial user intents based on the intent score;

[0126] S606. The server inputs the stressed word and the instruction text into the business service corresponding to the target user's intent.

[0127] S607. The server sends the business processing result output by the business service to the terminal device.

[0128] In some embodiments, users can execute various commands via voice control terminal device 200. Different commands can correspond to different service types, such as media asset services, weather query services, and device control services. Among them, device control refers to controlling devices such as air conditioners and refrigerators, for example, adjusting the air conditioner temperature.

[0129] Terminal devices can detect user intent based on voice, thereby determining the corresponding service type and obtaining the service processing result to provide feedback to the user, thus achieving voice interaction. For example, if a user inputs "I want to watch a movie," terminal device 200 can recommend movie-related media resources to the user.

[0130] It should be noted that, considering the limited computing power of the terminal device 200, the accuracy in recognizing user intent and stressed words may be low. Therefore, the user intent recognition process can be performed by the server 100. The terminal device 200 can send the user's voice to the server 100 and obtain the business processing results of the user's voice through the server 100.

[0131] In some embodiments, the terminal device 200 may collect audio, including user voice commands, via a sound acquisition device. The terminal device 200 may then send the user voice commands to the server 100.

[0132] Terminal device 200 can communicate with server 100 to exchange data, including receiving various data sent by server 100. For example, terminal device 200 can obtain media asset data from the network or obtain web page data that the user needs to browse through server 100. Both terminal device 200 and server 100 can be connected to the Internet and transmit interactive data between terminal device 200 and server 100 through Internet Protocol (IP).

[0133] Figure 7 Schematic diagrams illustrating the interaction between terminal device 200 and server 100 in some embodiments are shown. For example... Figure 7As shown, both the terminal device 200 and the server 100 need to be equipped with components for establishing a communication connection. The terminal device 200 includes a communication device 220, and the server 100 may include a communication module 110. The communication device 220 and the communication module 110 can simultaneously support at least one of the same communication methods to establish a communication connection. For example, the communication device 220 on the terminal device 200 includes a fiber optic interface for connecting to a network; similarly, the communication module 110 on the server 100 also includes a fiber optic interface to enable communication between the terminal device 200 and the server 100.

[0134] It should be noted that the terminal device 200 and the server 100 can also establish a communication connection using other methods. For example, wired networks, wireless LANs, cellular networks, Bluetooth, infrared, radio frequency communication, etc.

[0135] In some embodiments, after the server 100 receives a user's voice command, the control module can convert the user's voice command into command text.

[0136] The control module can detect the form of the user's voice commands sent by the terminal device 200. If the terminal device 200 has already sent the command text corresponding to the user's voice command, the control module does not need to convert it again and can directly use the command text. If the terminal device 200 sends the audio corresponding to the voice command, the control module can convert the audio into command text.

[0137] In some embodiments, the control module can perform stress detection on the instruction text to obtain stressed words in the instruction text. Stressed words can be syllables or words that are significantly prominent in terms of duration, amplitude (energy), and tone, reflecting the content that the user emphasizes, i.e., the content that the user wants the terminal device 200 to recognize.

[0138] When performing accent detection, the control module can perform word segmentation on the instruction text to obtain multiple text segments.

[0139] The control module can obtain all the words contained in the instruction text and determine them as text segmentation. Text segmentation can be a single character or a word composed of multiple characters. For example, the text "Play movies that children love to watch" after segmentation will result in the text segments "play", "children", "love", "watch", "of", and "movies".

[0140] In some embodiments, considering that the voice commands input by the user may not only be in Chinese, but may also be in English or other languages, or even be in a multilingual form, such as a voice command that is a mixture of Chinese and English, where the English command could be "search for XXX movie".

[0141] Different languages ​​differ significantly in grammatical habits and text structure. For example, Chinese, Japanese, and Thai are written in characters without spaces, while English and French use words as the smallest unit with spaces between them. If a sentence contains multiple languages, such as "Has the Dow Jones index fallen?", and the sentence contains both Chinese and English, the control module can first identify the language to accurately obtain the corresponding text segmentation for the instruction.

[0142] The control module can use Unicode regular expression matching, language word lists, and the Python language detection toolkit `langdetect` to identify the language in the command text. The identification process is as follows: First, the control module uses Unicode regular expression matching to detect the original data character by character. Since different languages ​​use different Unicode encodings for their letters, Unicode encoding can roughly identify some languages, such as Chinese, Japanese, Thai, and Arabic. However, languages ​​like English and French, which share the Latin alphabet, have the same alphabetic system and cannot be distinguished using Unicode encoding. Therefore, after detecting the language using Unicode regular expression matching, if there are original data entries that do not identify the language due to shared alphabetic systems, the control module can further detect the original data word by word using language word lists. Because language word lists are incomplete and languages ​​may share common words, there are cases where complete differentiation is not possible. In such cases, if there are original data entries that do not identify the language due to shared words, the control module can use the Python language detection toolkit `langdetect` to detect the language and combine the results of the previous detections to provide the final detection result.

[0143] After detecting the language, the control module can perform word segmentation processing using word units for each language (e.g., words for English, phrases for Chinese). For example, for the text "Dow Jones' index fell?", the text is segmented as [Dow, Jones, of, index, fell, fell, did].

[0144] In some embodiments, the control module can acquire audio features corresponding to text segmentation. These audio features may include Mel-frequency cepstrum coefficients (MFCCs), tone features, and energy features. Mel-frequency cepstrum coefficients are a representation of speech signals as low-dimensional features, simulating the nonlinear perception of frequency by the human ear.

[0145] The control module can obtain the Mel frequency cepstral coefficients based on the voice audio corresponding to the user's voice command.

[0146] The controller can first pre-emphasize the audio corresponding to the user's voice command to obtain the first audio signal.

[0147] Because the sound system of a terminal device suppresses the high-frequency components of speech data, it becomes difficult to identify the high-frequency spectrum, resulting in an inability to accurately recognize the emotional features of the speech data. The control module can input the speech audio into a Gaussian filter for pre-emphasis processing, enhancing the energy of the high-frequency components. The resulting first audio signal (also known as the compensated spectrum) is flatter. Simultaneously, the Gaussian filter can compensate for the high-frequency components suppressed by the sound system, making the recognition of the sound range more accurate.

[0148] The control module can sequentially perform frame segmentation and windowing processing on the first audio signal to obtain the second audio signal. Since the audio signal is continuous, it can be segmented into shorter frames for easier analysis.

[0149] The control module uses the set of N sampling points in the compensated spectrum as a frame observation unit for framing operations, according to the time sequence. Typically, N is 256 or 512, covering a time of approximately 20-30 ms; a larger value indicates a longer frame coverage time. In some embodiments, to avoid large differences in the audio range between adjacent frames, the control module sets an overlapping region between adjacent frames. This overlapping region contains M sampling points, where M is between 1 / 2 and 1 / 3 of N, to ensure a smooth transition between adjacent frames.

[0150] After framing the compensated spectrum, the control module feeds each frame's spectrum into a window function to eliminate discontinuities in the speech data between the beginning and end of each frame, preventing spectral leakage. After the control module feeds the framed compensated spectrum into the window function, a second audio signal (also called a windowed spectrum) is obtained. In this embodiment, window functions such as square windows, Hamming windows, and Hanning windows can be used. Based on the frequency domain characteristics of window functions, Hamming windows are commonly used.

[0151] The control module can perform Fourier transform processing on the second audio signal to convert it into a speech spectrum. Fourier transform is a method for converting audio signals from the time domain to the frequency domain, which helps in analyzing the intensity of different frequency components in the audio signal. Taking short-time Fourier transform as an example, a Fourier transform is performed on each frame, resulting in a series of spectrum diagrams.

[0152] The control module can input the speech spectrum into a Mel filter to eliminate harmonics in the speech spectrum, thus obtaining the Mel spectrum. By eliminating harmonics in the speech energy spectrum and highlighting regions with relatively concentrated energy in the speech energy spectrum, the computational load for extracting audio emotional features is reduced.

[0153] The control module can perform logarithmic operations and discrete cosine transform on the Mel frequency spectrum to obtain the Mel frequency cepstral coefficients.

[0154] Figure 8 The diagram illustrates the flowchart for obtaining Mel-frequency cepstral coefficients in some embodiments. For example... Figure 8 As shown, the control module sequentially performs pre-emphasis processing, frame segmentation processing, windowing processing, Fourier transform processing, Mel filter processing, logarithmic operation, and discrete cosine transform processing on the speech audio to obtain the Mel frequency cepstral coefficients.

[0155] In some embodiments, the control module can perform forced alignment processing on the instruction text based on the Mel frequency cepstral coefficients to obtain the audio time of the text segmentation in the speech audio.

[0156] Forced alignment refers to precisely aligning the words in the instruction text with the audio sequence of the speech, thus determining the start and end times of each word (or more finely granular phonemes) in the audio stream. The time between the start and end times constitutes the audio time for each text segmentation. The audio sequence corresponding to this audio time is the audio signal for text segmentation.

[0157] Forced alignment can be achieved using the Dynamic Time Warping (DTW) algorithm. DTW is an algorithm used to measure the similarity between two sets of time-series signals or time series. It allows for matching sequences by stretching and compressing a non-linear time axis, thereby measuring their similarity.

[0158] The dynamic time warping algorithm can be used to time-align multiple text segments of the instruction text sequence with the MFCC feature sequence in the speech audio stream, so as to obtain the specific time position of each text segment in the speech audio stream.

[0159] In some embodiments, the control module can determine the audio data corresponding to the text segmentation based on audio time. The control module can then extract the tone features and energy features corresponding to the text segmentation based on the audio data.

[0160] In some embodiments, the control module can perform stress detection on each text segment to obtain the stress confidence of the text segment.

[0161] The server has a pre-configured stress detection model. The control module can input the audio features corresponding to the text segmentation into the stress detection model, which can detect stresses in the audio features and output the stress confidence score for each text segment. The stress confidence score is used to characterize the degree to which a text segment is likely a stressed word.

[0162] The stress detection model can be pre-trained on a server using a deep learning model, such as a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), or a Long Short-Term Memory (LSTM) network. Deep learning models can perform in-depth learning and analysis of audio features to achieve stress detection. Taking a CNN-based stress detection model as an example, it can extract local features from audio features through convolutional layers, reduce the dimensionality of these features through pooling layers to decrease computation, and then integrate the extracted features through fully connected layers, ultimately outputting the stress confidence score for each text segment. The server can train an initial deep learning model using a large amount of training data, including text segments corresponding to the audio and audio features, to generate the stress detection model.

[0163] The control module can input the audio features of multiple text segments into the stress detection model to obtain the stress confidence of each text segment. For example, the instruction text "I want to watch free movies" is segmented as "I_want_to_watch_free_movies", and the stress confidence results are: "I, 0.0", "want, 0.1", "watch, 0.1", "free, 0.8", "of, 0.0", "movies, 0.6".

[0164] In some embodiments, the control module can detect the stress confidence of each text segment to determine whether each text segment is a stressed word.

[0165] If the stress confidence of the text segmentation reaches the preset confidence threshold, the control module marks the text segmentation as stressed words. If the stress confidence of the text segmentation does not reach the preset confidence threshold, the control module marks the text segmentation as unstressed words. The stress confidence can be set to 0.7.

[0166] In some embodiments, the control module may perform semantic analysis on the instruction text to obtain user intent analysis results. The user intent analysis results include at least one initial user intent.

[0167] The server has a pre-configured user intent recognition model. The control module can input command text into the pre-configured user intent recognition model, so that the model can identify the user intent based on the command text and output the user intent analysis results. The user intent analysis results include at least one initial user intent and the intent confidence level of the initial user intent.

[0168] In some embodiments, the user intent recognition model can employ a multilingual training model, such as a multilingual BERT Sentence Embedding (LaBSE) model, or a transformer neural network model. The user intent recognition model can analyze the features of each text segment to obtain the feature vectors of the text segments, thereby determining the user intent corresponding to the instruction text.

[0169] Considering that user speech may contain specific nouns, such as movie titles, the control module can also obtain entity content from the text segmentation to accurately identify user intent. Entities can be nouns or words contained in the instruction text, such as movie titles or personal names; these entities are all existing nouns.

[0170] The control module can perform part-of-speech tagging and entity recognition on text segmentation, for example, by using Stanza_NLP (Stanford Natural Language Processing Toolkit). It can also use other lexical analysis tools to filter out named entities and words with noun parts of speech and treat them as text entities.

[0171] The control module can obtain the entity tag corresponding to each text entity. The control module can first query the entity tags associated with each text entity based on a preset multilingual knowledge graph. A knowledge graph refers to a knowledge base that describes various entities or concepts existing in the real world and their relationships, such as the YAGO multilingual knowledge graph. YAGO includes not only entities (such as movies, people, cities, countries, etc.) but also the relationships between these entities (who acted in which movie, which city is located in which country, etc.).

[0172] The control module can input the instruction text, text segmentation, and corresponding entity labels into the user intent recognition model, which will then analyze the user intent corresponding to the instruction text.

[0173] The user intent recognition model may identify multiple user intents, but these intents are only identified through the command text. In order to accurately obtain the user intent, the server can combine voice audio to determine the user intent.

[0174] In some embodiments, the control module can score each initial user intent based on the accented word to obtain the intent score corresponding to the initial user intent.

[0175] It should be noted that the control module can detect whether the segmented text contains stressed words based on a confidence threshold. The instruction text may or may not contain stressed words.

[0176] Therefore, the control module can detect stressed words in the instruction text.

[0177] If the instruction text does not contain stressed words, it means that the user intent cannot be determined based on the audio. The control module can set the initial user intent with the highest intent confidence as the target user intent corresponding to the instruction text.

[0178] If the instruction text contains accented words, it means that the user's intent can be determined based on the audio, and the control module can score each initial user intent based on the accented words.

[0179] In some embodiments, the server may be configured with a keyword matching table, which is used to store the correspondence between keywords and initial user intents.

[0180] Some words can be strongly associated with a user's intent. When a user's speech contains these words, it's highly likely that the user has a strongly associated intent. For example, for the user intent "media asset search," strongly associated keywords could include movies, cartoons, documentaries, etc.; for the user intent "device control," strongly associated keywords could include turn on, increase, adjust, etc.

[0181] When scoring each initial user intent, the control module can obtain a keyword matching table that matches the initial user intent and obtain the keywords corresponding to each initial user intent, which are referred to as target keywords in this embodiment.

[0182] The control module can detect accented words in the target keywords and instruction text to determine whether the target keywords contain accented words.

[0183] If the target keyword contains an accented word, the control module can score the initial user intent based on the accented word.

[0184] If the target keywords do not contain stressed words, the control module can detect the intent confidence of the initial user intent and set the intent confidence as the intent score of the initial user intent.

[0185] In some embodiments, when scoring the initial user intent based on stressed words, the control module can detect the stress confidence of text segmentation.

[0186] The control module can calculate the stress weight based on the stress confidence score with the highest value. It should be noted that each word segment in the command text corresponds to a stress confidence score, and the control module can obtain the highest stress confidence score.

[0187] The control module can calculate the product of preset weight parameters and the maximum accent confidence, and use it as the accent weight.

[0188] The control module can weight the confidence of the initial user intent based on the accent weight to obtain the intent score corresponding to the initial user intent.

[0189] The intent score can be expressed as: Intent score = Intent confidence + Accent weight × Intent confidence; where the accent weight is expressed as α × accent confidence, and α is a preset weight parameter.

[0190] In some embodiments, considering that the number of initial user intents may be large, in order to compare the scores of multiple initial user intents, the stress confidence of stressed words existing in the keyword matching table can be used when calculating stress weights.

[0191] For an initial user intent, the control module can match keywords in its keyword matching table with accented words in the instruction text to obtain the same word, i.e., the target accented word that exists in the keyword matching table. There may be one or more target accented words.

[0192] The control module can calculate the stress weight based on the highest stress confidence in the target stressed word.

[0193] In some embodiments, the control module can filter out the target user intent corresponding to the instruction text from the initial user intents based on the intent score. The control module can set the initial user intent corresponding to the highest intent score as the target user intent.

[0194] The control module can obtain the service type corresponding to the target user's intent and input the accented words and instruction text into the service corresponding to the service type. For example, if the target user's intent is media asset search, the service type is media asset service; if the target user's intent is device control, the service type is device control service.

[0195] For each business type, a separate business service can be set up on the server. This business service is used to execute user voice commands for that business type. For example, a semantic skills microservice can be used as a business service.

[0196] The control module can input stressed words and command text as search criteria into the business service corresponding to the business type. The business service can then output the business processing results; for example, the media asset business can output media asset search results.

[0197] The control module can send the business processing results output by the business service to the terminal device, and the terminal device can respond to the user's voice commands based on the business processing results.

[0198] In some embodiments, to improve the speed of user intent recognition, stressed words can be set to words of a specific part of speech. For example, stressed words can be set to nouns and verbs; or, stressed words can be set to entities.

[0199] When detecting stressed words, the control module can detect the word type of the text segmentation.

[0200] If the word type is an entity type, the control module can obtain the audio features corresponding to the text segmentation to obtain the stress confidence of the text segmentation.

[0201] If the word type is not an entity type, the control module can set the stress confidence of the text segmentation to a preset confidence level, which can be 0.

[0202] In some embodiments, considering that some electronic devices may not support voice capture, the terminal device may be equipped with a joint control function. The terminal device with the joint control function enabled can capture user voice commands and send them to the server for intent recognition. After the server provides the service processing result, the terminal device can send the control command corresponding to the service processing result to the device to be controlled.

[0203] Users can instruct terminal device 200 to enable the joint control function.

[0204] In some embodiments, a user can send a control function activation command to the terminal device 200 via a designated button on the remote control. In practical applications, a pre-defined mapping between the control function activation command and the remote control button is established; for example, a dedicated control function button can be set on the remote control. When the user touches this button, the remote control can send the control function activation command to the terminal device 200.

[0205] Users can enable the joint control function through the system settings interface of the terminal device. In response to the user's operation of initiating system settings, the terminal device 200 can display the system settings interface. The system settings interface can include multiple settings items of the terminal device 200.

[0206] Figure 9 Schematic diagrams showing the system settings interface displayed by the terminal device 200 in some embodiments are shown. For example... Figure 9 As shown, the system settings interface of terminal device 200 can include multiple settings items, such as screen settings, sound settings, more function settings, network settings, and factory reset. After detecting a user's click on the more function settings item, terminal device 200 can display controls corresponding to several supported functions, including voice input controls, power off controls, restart controls, global search controls, and virtual remote control controls. Each function control can have an enable icon, which indicates that the corresponding function is enabled. Once terminal device 200 detects a user's click on the enable icon of a function control, it can enable the function corresponding to that control.

[0207] In some embodiments, multiple electronic devices can also establish communication connections through a communicator to enable direct communication between the electronic devices for data exchange.

[0208] The terminal device can display other electronic devices that can be connected and controlled to the user. Figure 10 Schematic diagrams of device control interfaces in some embodiments are shown. For example... Figure 10 As shown, the device control interface includes controls that display other electronic devices. Each control represents an electronic device, and the control can display the name of the electronic device, including the TV in the living room on the second floor, the refrigerator in the kitchen on the first floor, the speaker in the second bedroom, and the TV in the living room on the first floor.

[0209] In some embodiments, in order to ensure that the terminal device can stably send information to the device to be controlled under the joint control function, a master control device can be set in each area to collect the user's voice. Other devices do not collect the user's voice, and the master control device sends the user's voice commands to the server.

[0210] The main control device for each area can be configured by the user. Figure 11 Schematic diagrams of the main control device settings interface are shown in some embodiments. For example... Figure 11 As shown, the main control device settings interface displays multiple zones. After the user selects a zone, the monitor can display multiple electronic devices within that zone. The user can then select a specific device as the main control device for that zone. For example, the kitchen zone includes a refrigerator, range hood, monitor, and oven. The user can select the refrigerator as the main control device for the kitchen zone.

[0211] This application also provides a user intent recognition method applied to a server, the method comprising:

[0212] Step S601: In response to the user voice command sent by the terminal device, convert the user voice command into command text;

[0213] Step S602: Obtain the stressed words in the instruction text;

[0214] Step S603: Perform semantic analysis on the instruction text to obtain at least one initial user intent;

[0215] Step S604: Score each initial user intent based on the stressed word to obtain the intent score corresponding to the initial user intent;

[0216] Step S605: Based on the intent score, filter the target user intent corresponding to the instruction text from the initial user intents;

[0217] Step S606: Input the stressed word and the instruction text into the business service corresponding to the target user's intent;

[0218] Step S607: Send the business processing result output by the business service to the terminal device.

[0219] This application also provides a user intent recognition method applied to a server, the method comprising:

[0220] In response to the interface image corresponding to the target interface sent by the terminal device, the interface image is subjected to content understanding based on a preset content understanding model in order to obtain the interface elements contained in the interface image.

[0221] Detect controllable elements in the interface image;

[0222] Obtain the element annotation information of the controllable element and generate interface image information; the interface image information includes the element annotation information of the controllable element;

[0223] The interface image information is sent to the terminal device so that the terminal device generates and displays a virtual remote control interface based on the element labeling information of the controllable elements; the virtual remote control interface is located on top of the target interface.

[0224] The same or similar parts among the various embodiments in this specification can be referred to mutually, and will not be repeated here.

[0225] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a 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 of various embodiments or certain parts of the embodiments of the present invention.

[0226] 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0227] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the above exemplary discussion is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be obtained based on the above teachings. The selection and description of the above embodiments are for the purpose of better explaining the principles and practical applications, thereby enabling those skilled in the art to better utilize the embodiments and various different variations of embodiments suitable for specific application considerations.

Claims

1. A server, characterized in that, include: The communication module is configured to establish a communication connection with the terminal device; The control module is configured as follows: In response to a user voice command sent by a terminal device, the user voice command is converted into command text; Obtain the stressed words in the instruction text; Perform semantic analysis on the instruction text to obtain at least one initial user intent; Each initial user intent is scored based on the stressed word to obtain an intent score corresponding to the initial user intent; Based on the intent score, the target user intent corresponding to the instruction text is selected from the initial user intent; Input the stressed word and the instruction text into the business service corresponding to the target user's intent; The business processing results output by the business service are sent to the terminal device.

2. The server according to claim 1, characterized in that, The control module is configured to retrieve stressed words from the instruction text, specifically as follows: The instruction text is segmented to obtain multiple text segments; Obtain the audio features corresponding to the text segmentation; The audio features are subjected to stress detection to obtain the stress confidence of each of the text segments; The text segmentation words whose stress confidence reaches a preset confidence threshold are marked as stressed words.

3. The server according to claim 2, characterized in that, The control module is configured to acquire the audio features corresponding to the text segmentation as follows: The Mel frequency cepstral coefficients are obtained based on the audio corresponding to the user's voice command. Based on the Mel frequency cepstral coefficients, the instruction text is subjected to forced alignment processing to obtain the audio timing of the text segmentation in the speech audio. Based on the audio time, determine the audio data corresponding to the text segmentation; Based on the audio data, extract the tone features and energy features corresponding to the text word segmentation.

4. The server according to claim 3, characterized in that, The control module executes the process of obtaining Mel-frequency cepstral coefficients based on the voice audio corresponding to the user's voice command, specifically configured as follows: The audio corresponding to the user's voice command is pre-emphasized to obtain a first audio signal; The first audio signal is subjected to frame segmentation and windowing processing to obtain the second audio signal; Perform Fourier transform processing on the second audio signal to convert the second audio signal into a speech spectrum; The harmonics of the speech spectrum are eliminated to obtain the Mel spectrum; The Mel frequency cepstral coefficients are obtained by performing logarithmic and discrete cosine transform operations on the Mel spectrum.

5. The server according to claim 1, characterized in that, The control module performs semantic analysis on the instruction text, specifically configured as follows: The instruction text is input into a preset user intent recognition model so that the user intent recognition model outputs user intent analysis results; the user intent analysis results include at least one initial user intent and the intent confidence of the initial user intent.

6. The server according to claim 5, characterized in that, The control module performs a scoring of each initial user intent based on the stressed word, specifically configured as follows: Detect stressed words in the instruction text; If the instruction text contains accented words, then the step of scoring the at least one initial user intent based on the accented words is performed; If the instruction text does not contain stressed words, then the initial user intent with the highest intent confidence is set as the target user intent corresponding to the instruction text.

7. The server according to claim 1, characterized in that, The control module performs a scoring of each initial user intent based on the stressed word, specifically configured as follows: The target keywords corresponding to the initial user intent are obtained based on a preset keyword matching table; the keyword matching table is used to store the correspondence between keywords and initial user intents. Detect stressed words in the instruction text based on the target keywords; If the target keyword contains the stressed word, then the initial user intent is scored based on the stressed word; If the target keyword does not contain stressed words, then the intent confidence of the initial user intent is detected, and the intent confidence is set as the intent score of the initial user intent.

8. The server according to claim 7, characterized in that, The control module performs a scoring of the initial user intent based on the stressed words, specifically configured as follows: Detect the stress confidence of the text segmentation; Accent weights are calculated based on the accent confidence score with the highest numerical value. The intent confidence of the initial user intent is weighted based on the accent weights to obtain the intent score of the initial user intent.

9. A terminal device, characterized in that, include: The sound acquisition device is configured to acquire audio. A communication device configured to establish a communication connection with a server; The controller is configured as follows: Acquire user voice commands collected by the sound acquisition device; The user's voice command is sent to the server, so that the server performs stress detection on the command text corresponding to the user's voice command to obtain stressed words in the command text, and performs semantic analysis on the command text to obtain at least one initial user intent. Furthermore, based on the accented words, an intent score corresponding to the initial user intent is obtained, and based on the intent score, the at least one initial user intent is filtered to obtain the target user intent corresponding to the instruction text; Obtain the business type corresponding to the target user's intent, and obtain the business processing result through the business service corresponding to the business type; Receive the business processing results sent by the server.

10. A method for recognizing user intent, characterized in that, Applied to the server according to any one of claims 1-8, the method comprises: In response to a user voice command sent by a terminal device, the user voice command is converted into command text; Obtain the stressed words in the instruction text; Perform semantic analysis on the instruction text to obtain at least one initial user intent; Each initial user intent is scored based on the stressed word to obtain an intent score corresponding to the initial user intent; Based on the intent score, the target user intent corresponding to the instruction text is selected from the initial user intent; Input the stressed word and the instruction text into the business service corresponding to the target user's intent; The business processing results output by the business service are sent to the terminal device.

Images