Illuminating lamp voice interaction method and system used in different scenes

Abstract

The invention relates to the technical field of intelligent lighting control, and discloses a lighting lamp voice interaction method and system used in different scenes, and the method comprises the steps: constructing a multi-scene voice recognition model, collecting command words related to the different scenes, carrying out the preprocessing of the collected command words, constructing a training data set, and carrying out the recognition of the multi-scene voice recognition model. Training the multi-scene speech recognition model by using the training data set; embedding the trained multi-scene speech recognition model into an SDK (Software Development Kit), distinguishing and combining the collected command words according to scenes to generate command word combinations, performing scene labeling on each command word combination, and storing the command word combinations and labels thereof in the SDK in a data structure of a hash table; and recognizing a command word combination to which a currently triggered command word belongs through the multi-scene voice recognition model, binding a scene according to a label of the combination, filtering and intercepting command words which do not belong to the command word combination in newly input command words, and realizing that a plurality of lighting lamp devices do not interfere with each other in different scenes.

Description

Technical Field

[0001] The present invention relates to the field of intelligent lighting control technology, and in particular to a method and system for voice interaction of lighting lamps used in different scenarios. Background Art

[0002] In our daily life, home lighting has a variety of styles, and its form and appearance structure have developed very maturely. Now we are in the era of big data, intelligent manufacturing has attracted much attention and popularity, and home lighting is also developing rapidly.

[0003] Generally, smart lighting equipment will pre-train all command words into the model according to the customer's product solution requirements to support all command word control. However, in this case, the voice effect will not be ideal, and there will be confusion and misrecognition. In the actual experience of self-learning function products, it was found through testing that the self-learning function may fail to learn due to unclear human voice or noisy environment, and the effect of voice interaction after learning will be relatively poor.

[0004] In addition, in general home situations, there are multiple scenes, such as living room, study and bedroom, etc., which are triggered by only common command words, and multiple lighting devices interfere with each other. Generally, smart devices can realize scene-by-scene control by inputting scene parameters in advance, but the function usage process is cumbersome.

[0005] In order to be compatible with command control in more scenarios and take into account the actual effect of voice interaction, a solution that supports "scene binding" of user terminals is very necessary. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the above-mentioned prior art and provide a method and system for voice interaction of lighting lamps used in different scenarios. Even when the actual installation location of the lighting fixture is unknown, the current scene can be identified according to the command. At the same time, multiple lighting fixtures do not interfere with each other (voice interaction), and a relatively good voice recognition effect can be guaranteed to avoid misoperation (turning off the light) and unnecessary energy waste (turning on the light by mistake).

[0007] On the one hand, a method for voice interaction of a lighting lamp used in different scenarios is provided, comprising the following steps: S1: construct a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; S2: embed the trained multi-scenario speech recognition model into the SDK, distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure inside the SDK; S3: Establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

[0008] Furthermore, in step S1, the command words involved in the different scenarios include general command words, special command words and targeted command words, and the command words are processed using a natural language processing basic algorithm, and the preprocessing of the collected command words further includes: Remove repeated and ambiguous parts, and group the command words with the same meaning but different expressions into one group for multi-scenario speech recognition model learning; Label each command word, including its corresponding function and scene information, so that the multi-scene speech recognition model can be accurately associated during training; The labeled command words are organized into a unified data format to adapt them to the input specifications of multi-scenario speech recognition models.

[0009] Preferably, in step S1, constructing a multi-scenario speech recognition model further includes: A supervised learning algorithm is used to use the labeled training data set as a training sample for the multi-scenario speech recognition model, and the mapping relationship between the corresponding functions and scenarios of different command words is learned. By repeatedly adjusting the parameters of the multi-scenario speech recognition model, the model can accurately predict the corresponding operation and applicable scenario based on the input command word; A portion of the labeled data is reserved as a test set. During the training process, the test set is used regularly to evaluate the multi-scenario speech recognition model, and the accuracy and recall rate indicators are calculated. According to the evaluation results, the hyperparameters of the multi-scenario speech recognition model are adjusted to improve the model structure.

[0010] Furthermore, in step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: According to the programming language and interface specifications supported by the SDK, the code of the multi-scenario speech recognition model is encapsulated and adapted to ensure that the multi-scenario speech recognition model can be called normally in the operating environment of the SDK and provide a unified voice interaction interface function to the outside world.

[0011] Preferably, the function of providing a unified voice interaction interface to the outside world further includes: After receiving the external input voice signal, the voice interaction interface provided by the SDK calls the integrated multi-scene speech recognition model for recognition, converts it into a command word in text form, and transmits the command word to the command word matching module inside the model. The module uses a string matching algorithm to traverse the command word combination for matching, and after determining the grouping, reads the scene marked by the command word combination.

[0012] Furthermore, in step S3, the establishing mode state management mechanism further includes: The mode states include scene control mode and general mode. Within a preset time after power-on, If a specific voice command that meets the conditions for entering a mode state switching is received, the corresponding mode state switching process is started; If no specific voice command that meets the conditions for entering the mode state switching is received, the general mode is entered by default; After successfully triggering the mode state switch, the user is informed of the current mode state of the device.

[0013] Preferably, the mode state switching further includes: If the current mode state is the scene control mode, the scene binding process in step S3 is started; After binding the scene, if you need to switch the mode status, power on again, receive the specific voice command for scene reset within the preset time after power-on, start the scene reset process, clear the currently bound scene status identifier and the command word set during the scene binding period, and restore the lighting fixture to the initial general mode.

[0014] Preferably, in step S3, a complete closed-loop logic of voice interaction in the scene is constructed by continuously maintaining the scene state identification and updating the parameters, and the light state is continuously updated according to the logical relationship between the current light state and the adjustment command words, so that the subsequent voice command words in the same scene are reasonably responded to based on the latest state, further comprising: After the scene is bound, the lighting fixture continuously maintains the scene status identifier, and while executing the operation corresponding to the command word, updates and records the relevant parameters of the lighting status of the current scene, so as to accurately and coherently respond to other continuous lighting voice commands in the same scene; Taking into account the unexpected power outage, long-term voice commands are set, including incoming call memory, incoming call always on and incoming call always off. After the scene is successfully bound, the long-term voice command is selected according to the specific situation to control the incoming call of the light. At the same time, the lighting parameters are recorded according to the usage habits, and the recorded lighting parameters are quickly applied through the incoming call memory voice command.

[0015] For voice interaction operations bound to the current scene, the lighting fixtures provide exclusive voice feedback and prompts to help users understand the current scene and the scope of valid voice commands under it.

[0016] Furthermore, in step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: Configure the firmware according to the characteristics and requirements of the target lighting fixture hardware equipment, including configuring voice interaction parameters, audio encoding format, and hardware resource allocation; Use the compiler corresponding to the target lighting fixture hardware to compile the firmware engineering source code including the voice interaction SDK and other underlying hardware drivers and system control codes to generate an executable binary firmware file; The compiled secondary firmware file is written into the storage chip of the target lighting fixture hardware device through a special burning tool.

[0017] On the other hand, a lighting voice interaction system for use in different scenarios is disclosed, comprising: A model building module is used to build a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; The model integration and grouping module is used to embed the trained multi-scenario speech recognition model into the SDK, distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure within the SDK; The state management module is used to establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the command words in the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention distinguishes and combines the required command words that are sorted and summarized according to the scene, and realizes scene distinction by recognizing the voice command words, so that the voice interactions of multiple lighting devices do not interfere with each other; The present invention achieves scene matching based on recognition of command words by constructing a multi-scene speech recognition model and conducting learning and training; The present invention achieves a relatively good speech recognition effect by integrating a multi-scenario speech recognition model into the SDK. Otherwise, misoperation (turning off the light) and unnecessary energy waste (turning on the light by mistake) may occur. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a voice interaction method for lighting lamps used in different scenarios of the present invention; Figure 2 This is a structural block diagram of a lighting voice interaction system used in different scenarios of the present invention; Figure 3 The present invention is a user use flow diagram. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0022] The specific implementation of the present invention is described below with reference to the accompanying drawings and embodiments.

[0023] Example 1 like Figure 1 As shown, a technical solution of a lighting voice interaction method used in different scenarios provided by this embodiment includes the following steps: S1: construct a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; S2: embed the trained multi-scenario speech recognition model into the SDK (software development kit), distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure inside the SDK; S3: Establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

[0024] In step S1, we first conduct in-depth research on different usage scenarios. For example, in the living room scenario, we may need to involve command words such as turning on and off lights, adjusting brightness, and switching light colors; in the bedroom scenario, we may also have special command words such as setting a timer to turn off lights and turning on night light mode; and in commercial places such as conference rooms, we may have targeted command words such as adjusting lights in projection mode. Through communication with potential users, relevant industry personnel, and actual observation, we can comprehensively collect the voice interaction needs that may be used in various scenarios.

[0025] Furthermore, the command words involved in the different scenarios include general command words, special command words and targeted command words. The command words are processed using the basic natural language processing algorithm to decompose the command words into vocabulary units to help the multi-scenario speech recognition model understand the word structure. At the same time, the grammatical analysis algorithm is used to analyze the grammatical relationship between each word, so as to construct the understanding logic of the command words at the grammatical level. Among them, the pre-processing of the collected command words includes: Remove repeated and ambiguous parts, and group the command words with the same meaning but different expressions into one group for multi-scenario speech recognition model learning. For example, for brightness adjustment, you can group a series of command words with different expressions but the same meaning, such as "brighter", "darker", "brightness to 50%", etc., to facilitate subsequent model learning; Label each command word, including its corresponding function and scene information, so that the multi-scenario speech recognition model can accurately associate it during training. For example, "turn on the light" is labeled as a control switch function, applicable to all indoor lighting scenes, etc., so that the model can accurately understand and associate it during training; The annotated command words are organized into a unified data format to adapt to the input specifications of the multi-scenario speech recognition model. In this embodiment, the CSV format is uniformly adopted, and each line contains a command word and corresponding annotation information.

[0026] In step S1, building a multi-scenario speech recognition model further includes: Using supervised learning algorithms, such as Naive Bayes algorithm, support vector machine (SVM) or neural networks in deep learning (such as recurrent neural network RNN ​​and its variants LSTM, GRU, etc., considering that voice command words have sequence characteristics), using the labeled training data set as training samples of the multi-scenario speech recognition model, learning the mapping relationship between the corresponding functions and scenes of different command words, and repeatedly adjusting the parameters of the multi-scenario speech recognition model (such as weights in the neural network, etc.), so that it can accurately predict the corresponding operation and applicable scene according to the input command words; A portion of the labeled data is reserved as a test set. During the training process, the test set is used regularly to evaluate the multi-scenario speech recognition model, and indicators such as accuracy, recall, and F1 value are calculated. According to the results of the evaluation, the hyperparameters of the multi-scenario speech recognition model (such as learning rate, number of iterations, etc.) are adjusted to improve the model structure.

[0027] In step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: According to the programming language and interface specifications supported by the SDK, the code of the multi-scenario speech recognition model is encapsulated and adapted to ensure that the multi-scenario speech recognition model can be called normally in the operating environment of the SDK and provide a unified voice interaction interface function to the outside world.

[0028] The command words that have been sorted out and summarized are grouped according to different usage scenarios determined in previous surveys, such as home, office, and commercial places. For example, relevant command words such as "turn on the living room light", "turn off the bedroom light", and "adjust the brightness of the kitchen light" are put into the home scene group; command words suitable for office environments such as "turn on the conference room light" and "adjust the color of the workstation light" are included in the office scene group, forming a set of command words corresponding to each scene.

[0029] Appropriate data structures are used within the SDK to store these command word combinations that are differentiated by scene. Common structures may be hash tables or maps, with the scene name as the key and the corresponding command word set as the value, so as to facilitate subsequent rapid search and matching of corresponding command words based on the currently recognized scene.

[0030] Among them, in this embodiment, the function of providing a unified voice interaction interface to the outside world further includes: After receiving the external input voice signal, the voice interaction interface provided by the SDK calls the integrated multi-scene speech recognition model for recognition, converts it into a command word in text form, and transmits the command word to the command word matching module inside the model. The module uses a string matching algorithm to traverse the command word combination for matching, and after determining the grouping, reads the scene marked by the command word combination.

[0031] Specifically, in this embodiment, a string matching algorithm (such as simple direct comparison, data structure search based on a dictionary tree (Trie), etc.) is used. After the recognized command word text is obtained, the command word set corresponding to the scene is traversed, and through efficient string matching operations, it is quickly determined whether the command word is in a certain scene group, thereby achieving logical verification of command words distinguished by scene, and ensuring the rationality of the scene of subsequent voice interaction operations.

[0032] In step S3, the establishing mode state management mechanism further comprises: The mode states include scene control mode and general mode. Within 30 seconds after power-on, If a specific voice command that meets the conditions for entering a mode state switching is received, the corresponding mode state switching process is started; If no specific voice command that meets the conditions for entering the mode state switching is received, the general mode is entered by default; After successfully triggering the mode state switch, the user is informed of the current mode state of the device.

[0033] Wherein, the mode state switching further includes: If the current mode state is the scene control mode, the scene binding process in step S3 is started; After binding the scene, if you need to switch the mode status, power on again, receive the specific voice command for scene reset within 30 seconds after power-on, start the scene reset process, clear the currently bound scene status identifier and the command word set during the scene binding period, and restore the lighting fixture to the initial general mode.

[0034] The 30-second timer limit ensures that the state switching operation can be triggered within a specific time period after the device is powered on, avoiding unnecessary function switching confusion caused by user misoperation or long waiting. The logic of voice triggering relies on accurate voice recognition and matching judgment of specific triggering instructions. Only when the recognized voice instruction is completely consistent with the preset instruction to enter the state switching, the subsequent switching process will be executed to ensure the accuracy and seriousness of the state switching operation.

[0035] Specifically, in this embodiment, we use the state machine algorithm idea to treat different scene combinations as different states, and each state has a corresponding set of legal input command words as state transition conditions. When a command word triggers a state transition (entering a certain scene combination state), only subsequent inputs that meet the transition conditions in the state (i.e., command words belonging to the scene combination) can keep the state or transfer to other related sub-states in the scene combination (such as waiting for the next valid operation command word in the same scene after executing brightness adjustment). If it does not meet the conditions, the original state is maintained and the input is ignored, so as to implement strict command word constraint logic within the scene.

[0036] In addition, for each command word received, a comparison algorithm with the set of legal command words in the current state (such as a set member judgment algorithm) is used to determine whether to execute the operation. If it is an illegal command word, the feedback algorithm is triggered to provide the user with corresponding prompt voice or text information (such as "This command is invalid in the current scenario, please change the command word") to ensure that the user can clearly understand the effectiveness of the voice interaction and improve the interactive experience.

[0037] In step S3, a complete closed-loop voice interaction logic is constructed in the scene by continuously maintaining the scene status identifier and updating the parameters, and the light status is continuously updated according to the logical relationship between the current light status and the adjustment command words, so that the subsequent voice command words in the same scene respond reasonably based on the latest status, further including: After the scene is bound, the lighting fixture continuously maintains the scene status identifier, and while executing the operation corresponding to the command word, updates and records the relevant parameters of the lighting status of the current scene, so as to accurately and coherently respond to other continuous lighting voice commands in the same scene; Taking into account the unexpected power outage, long-term voice commands are set, including incoming call memory, incoming call always on and incoming call always off. After the scene is successfully bound, the long-term voice command is selected according to the specific situation to control the incoming call of the light. At the same time, the lighting parameters are recorded according to the usage habits, and the recorded lighting parameters are quickly applied through the incoming call memory voice command.

[0038] For voice interaction operations bound to the current scene, the lighting fixtures provide exclusive voice feedback and prompts to help users understand the current scene and the scope of valid voice commands under it.

[0039] In addition, in step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: Configure the firmware according to the characteristics and requirements of the target lighting fixture hardware equipment, including configuring voice interaction parameters, audio encoding format, and hardware resource allocation; Use the compiler corresponding to the target lighting fixture hardware to compile the firmware engineering source code including the voice interaction SDK and other underlying hardware drivers and system control codes to generate an executable binary firmware file; The compiled secondary firmware file is written into the storage chip of the target lighting fixture hardware device through a special burning tool.

[0040] This embodiment also provides a lighting voice interaction system for use in different scenarios, such as Figure 2 As shown, including: A model building module is used to build a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; The model integration and grouping module is used to embed the trained multi-scenario speech recognition model into the SDK, distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure within the SDK; The state management module is used to establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the command words in the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

[0041] in, Figure 2 The functional implementation of each module and unit in the lighting lamp voice interaction system used in different scenarios corresponds to the steps in the above-mentioned lighting lamp voice interaction method used in different scenarios, and its functions and implementation processes will not be repeated here one by one.

[0042] like Figure 3 As shown, the process of using this system based on the above method is as follows: Device power-on: hardware connection and power supply preparation, boot guidance and self-test, and initialization of voice interaction environment; A. Use the default general command words to control a single device. The user speaks the default general command words, such as "turn on the light", "turn off the light", "brighter", and other simple and common words used to control the basic functions of the light. After the microphone on the device receives the voice signal, it transmits it to the voice processing module for subsequent processing. Once the legal and valid command words and corresponding operation instructions are parsed, the control module of the device implements the corresponding control action of the light according to the instructions through the control circuit (such as adjusting the PWM signal of the light driver chip to control the brightness, etc.); B. Within 30 seconds after power-on, the device is triggered by voice commands and is ready to switch mode according to the received triggering voice commands, which involves switching from the default general command word control state to the scene-based control state. After the state switch is successfully triggered, the user is informed of the current state change of the device through voice response; B. (2) Binding scenes, such as turning on / off the living room light, then the device will only support voice commands for the living room light, such as turning on / off the living room light, making the living room light brighter / dimmer, etc. For voice interaction operations under scene binding, the device provides exclusive voice feedback and prompt information.

[0043] (3) Just like the mode state switching trigger mechanism in step 3, trigger the scene reset through voice commands within 30 seconds of power-on. After the reset, only the default general command words are supported when powering on again.

[0044] Finally, it should be noted that the above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

[0045] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A lighting voice interaction method used in different scenarios, characterized in that: The steps include: S1: construct a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; S2: embed the trained multi-scenario speech recognition model into the SDK, distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure inside the SDK; S3: Establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

2. The voice interaction method for lighting lamps used in different scenarios according to claim 1, characterized in that: In step S1, the command words involved in the different scenarios include general command words, special command words and targeted command words. The command words are processed using a natural language processing basic algorithm. The preprocessing of the collected command words further includes: Remove repeated and ambiguous parts, and group the command words with the same meaning but different expressions into one group for multi-scenario speech recognition model learning; Label each command word, including its corresponding function and scene information, so that the multi-scene speech recognition model can be accurately associated during training; The labeled command words are organized into a unified data format to adapt them to the input specifications of multi-scenario speech recognition models.

3. The lighting lamp voice interaction method used in different scenarios according to claim 2 is characterized in that: In step S1, building a multi-scenario speech recognition model further includes: A supervised learning algorithm is used to use the labeled training data set as a training sample for the multi-scenario speech recognition model, and the mapping relationship between the corresponding functions and scenarios of different command words is learned. By repeatedly adjusting the parameters of the multi-scenario speech recognition model, the model can accurately predict the corresponding operation and applicable scenario based on the input command word; A portion of the labeled data is reserved as a test set. During the training process, the test set is used regularly to evaluate the multi-scenario speech recognition model, and the accuracy and recall rate indicators are calculated. According to the evaluation results, the hyperparameters of the multi-scenario speech recognition model are adjusted to improve the model structure.

4. The lighting lamp voice interaction method used in different scenarios according to claim 1, characterized in that: In step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: According to the programming language and interface specifications supported by the SDK, the code of the multi-scenario speech recognition model is encapsulated and adapted to ensure that the multi-scenario speech recognition model can be called normally in the operating environment of the SDK and provide a unified voice interaction interface function to the outside world.

5. The method for voice interaction of lighting lamps used in different scenarios according to claim 4, characterized in that: The function of providing a unified voice interaction interface to the outside world further includes: After receiving the external input voice signal, the voice interaction interface provided by the SDK calls the integrated multi-scenario speech recognition model for recognition, converts it into a command word in text form, and transmits the command word to the command word matching module inside the model. The module uses a string matching algorithm to traverse the command word combination for matching, and after determining the grouping, reads the scene marked by the command word combination.

6. The lighting lamp voice interaction method used in different scenarios according to claim 1, characterized in that: In step S3, the establishing mode state management mechanism further comprises: The mode states include scene control mode and general mode. Within a preset time after power-on, If a specific voice command that meets the conditions for entering a mode state switching is received, the corresponding mode state switching process is started; If no specific voice command that meets the conditions for entering the mode state switching is received, the general mode is entered by default; After successfully triggering the mode state switch, the user is informed of the current mode state of the device.

7. The method for voice interaction of lighting lamps used in different scenarios according to claim 6, characterized in that: The mode state switching further includes: If the current mode state is the scene control mode, the scene binding process in step S3 is started; After binding the scene, if you need to switch the mode status, power on again, receive the specific voice command for scene reset within the preset time after power-on, start the scene reset process, clear the currently bound scene status identifier and the command word set during the scene binding period, and restore the lighting fixture to the initial general mode.

8. The lighting lamp voice interaction method used in different scenarios according to claim 1, characterized in that: In step S3, a complete closed-loop voice interaction logic is constructed in the scene by continuously maintaining the scene status identifier and updating the parameters, and the light status is continuously updated according to the logical relationship between the current light status and the adjustment command words, so that the subsequent voice command words in the same scene respond reasonably based on the latest status, further including: After the scene is bound, the lighting fixture continuously maintains the scene status identifier, and while executing the operation corresponding to the command word, updates and records the relevant parameters of the lighting status of the current scene, so as to accurately and coherently respond to other continuous lighting voice commands in the same scene; Taking into account the unexpected power outage, long-term voice commands are set, including incoming call memory, incoming call always on and incoming call always off. After the scene is successfully bound, the long-term voice command is selected according to the specific situation to control the incoming call of the light. At the same time, the lighting parameters are recorded according to the usage habits, and the recorded lighting parameters are quickly applied through the incoming call memory voice command.

9. For the voice interaction operations bound to the current scene, the lighting fixtures provide exclusive voice feedback and prompts to help users understand the current scene and the effective voice command range under it.

10. The lighting lamp voice interaction method used in different scenarios according to claim 1, characterized in that: In step S2, embedding the trained multi-scenario speech recognition model into the SDK further includes: Configure the firmware according to the characteristics and requirements of the target lighting fixture hardware equipment, including configuring voice interaction parameters, audio encoding format, and hardware resource allocation; Use the compiler corresponding to the target lighting fixture hardware to compile the firmware engineering source code including the voice interaction SDK and other underlying hardware drivers and system control codes to generate an executable binary firmware file; The compiled secondary firmware file is written into the storage chip of the target lighting fixture hardware device through a special burning tool.

11. A lighting voice interaction system used in different scenarios, characterized in that: include: A model building module is used to build a multi-scenario speech recognition model, collect command words involved in different scenarios, construct a training data set after pre-processing the collected command words, and use the training data set to train the multi-scenario speech recognition model; The model integration and grouping module is used to embed the trained multi-scenario speech recognition model into the SDK, distinguish and combine the collected command words according to the scene to generate command word combinations, annotate each command word combination with the scene, and store the command word combinations and their annotations in a hash table data structure within the SDK; The state management module is used to establish a mode state management mechanism, identify the command word combination to which the currently triggered command word belongs through the multi-scenario speech recognition model, and filter and intercept the command words in the newly input command words that do not belong to the command word combination according to the annotated binding scene of the combination.

Images