Device wake-up word generation method and electronic device

Abstract

The present disclosure provides a device wake-up word generation method, comprising: in response to monitoring that a function module configured by a target device changes, obtaining first attribute information corresponding to the changed function module; generating a target constraint rule based on the first attribute information; and generating at least one target wake-up word satisfying the target constraint rule through a first model based on the target constraint rule, the target wake-up word being used to wake up the function module.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and more specifically, to a method for generating a device wake word and an electronic device. Background Technology

[0002] With the rapid development of artificial intelligence and the Internet of Things (IoT) technologies, voice interaction has become one of the most important human-computer interaction methods for smart terminal devices. As smart devices offer increasingly richer functions and services, the need to configure dedicated wake words for specific scenarios or functional modules to improve interaction efficiency is becoming increasingly prominent. However, the wake words of devices often require manual intervention to set them. Summary of the Invention

[0003] In view of this, the present disclosure provides a method for generating a device wake word and an electronic device.

[0004] One aspect of this disclosure provides a device wake-up word generation method, comprising: in response to detecting a change in a functional module configured in a target device, acquiring first attribute information corresponding to the changed functional module; generating target constraint rules based on the first attribute information; and generating at least one target wake-up word that satisfies the target constraint rules through a first model based on the target constraint rules, wherein the target wake-up word is used to wake up the functional module.

[0005] According to embodiments of this disclosure, generating target constraint rules based on first attribute information includes: generating target prompt words as target constraint rules based on the first attribute information, wherein the target prompt words can constrain the output content of the first model to have a semantic association with the first attribute information.

[0006] According to embodiments of this disclosure, in response to detecting a change in the functional modules configured in the target device, obtaining first attribute information corresponding to the changed functional modules includes at least one of the following: in response to detecting the addition of a first functional module to the target device, obtaining first attribute information corresponding to the first functional module; in response to detecting a function update corresponding to a second functional module configured in the target device, obtaining first attribute information corresponding to the updated second functional module.

[0007] According to embodiments of this disclosure, the device wake-up word generation method further includes at least one of the following: adding a target wake-up word corresponding to a first functional module to the wake-up word set of the target device; replacing the wake-up word corresponding to a second functional module in the wake-up word set of the target device with the target wake-up word corresponding to the second functional module; and deleting the target wake-up word corresponding to the third functional module in the wake-up word set of the target device in response to detecting that a third functional module configured in the target device is offline.

[0008] According to embodiments of this disclosure, the device wake-up word generation method further includes: obtaining second attribute information corresponding to each of the functional modules currently configured in the target device; and generating target constraint rules based on the first attribute information, including: generating target constraint rules based on the first attribute information and the second attribute information.

[0009] According to embodiments of this disclosure, generating at least one target wake word that satisfies the target constraint rules through a first model based on target constraint rules includes: generating multiple candidate wake words that satisfy the target constraint rules through a first model based on the target constraint rules; performing conflict detection between each candidate wake word; and / or performing conflict detection between each candidate wake word and each wake word of the target device to determine at least one target wake word that does not have a conflict.

[0010] According to embodiments of this disclosure, determining at least one target wake word that does not conflict includes: in response to detecting a conflict between a candidate wake word and a wake word of a target device, adjusting target constraint rules based on the content of the conflict; and based on the adjusted target constraint rules, regenerating at least one target wake word that satisfies the target constraint rules and does not conflict with any wake words of the target device through a first model.

[0011] According to embodiments of this disclosure, the target device includes a first voice acquisition module and a second voice acquisition module. The first voice acquisition module continuously acquires voice information in the environment. The power consumption of the second voice acquisition module is greater than that of the first voice acquisition module when it is working. The process of waking up the functional module includes: in response to detecting that the first voice information acquired by the first voice acquisition module matches the target wake-up word, starting the second voice acquisition module to acquire the second voice information in the environment; processing the second voice information based on a second model to obtain a target instruction; and waking up the functional module in the target device corresponding to the target wake-up word based on the target instruction.

[0012] According to an embodiment of this disclosure, the target device includes a first voice acquisition module, which continuously acquires voice information in the environment. The process of waking up the functional module includes: processing the first voice information acquired by the first voice acquisition module based on a third model to obtain a target result, wherein the third model is used to identify whether the semantics of the first voice information matches a target wake-up word; and waking up the functional module in the target device corresponding to the target wake-up word in response to the target result indicating that the semantics of the first voice information matches the target wake-up word.

[0013] Another aspect of this disclosure provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the following operations: in response to detecting a change in a functional module configured in a target device, acquiring first attribute information corresponding to the changed functional module; generating target constraint rules based on the first attribute information; and generating at least one target wake-up word satisfying the target constraint rules through a first model based on the target constraint rules, the target wake-up word being used to wake up the functional module.

[0014] Another aspect of this disclosure provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform a device wake word generation method according to any of the foregoing embodiments.

[0015] Another aspect of this disclosure provides a computer program product, including a computer program / instructions, characterized in that the computer program / instructions, when executed by a processor, implement the operation of the device wake word generation method of any of the foregoing embodiments. Attached Figure Description

[0016] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 A flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0018] Figure 2 Another flowchart of a device wake word generation method according to an embodiment of the present disclosure is illustrated schematically;

[0019] Figure 3 Another flowchart of a device wake word generation method according to an embodiment of the present disclosure is illustrated schematically;

[0020] Figure 4 Another flowchart of a device wake word generation method according to an embodiment of the present disclosure is illustrated schematically;

[0021] Figure 5 Another flowchart of a device wake word generation method according to an embodiment of the present disclosure is illustrated schematically;

[0022] Figure 6 A flowchart illustrating the generation of a target wake word in a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 7The flowchart illustrating the process of determining at least one target wake-up that does not conflict in a device wake-up word generation method according to an embodiment of the present disclosure is shown in the illustration.

[0024] Figure 8 A flowchart illustrating a wake-up function module according to an embodiment of the present disclosure is shown schematically;

[0025] Figure 9 Another flowchart of the wake-up function module according to an embodiment of the present disclosure is shown schematically;

[0026] Figure 10 A block diagram schematically illustrates a device wake word generation apparatus according to an embodiment of the present disclosure; and

[0027] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Detailed Implementation

[0028] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0029] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0031] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0032] In the embodiments disclosed herein, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0033] Embodiments of this disclosure provide a device wake-up word generation method, including: in response to detecting a change in a functional module configured in a target device, obtaining first attribute information corresponding to the changed functional module; generating target constraint rules based on the first attribute information; and generating at least one target wake-up word that satisfies the target constraint rules through a first model based on the target constraint rules, wherein the target wake-up word is used to wake up the functional module.

[0034] Figure 1 A flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0035] like Figure 1 As shown, the device wake-up word generation method may include at least operations S110 to S130.

[0036] In operation S110, in response to detecting a change in the functional modules configured on the target device, the first attribute information corresponding to the changed functional module is obtained. The target device can be a hardware carrier with voice interaction capabilities, such as a smart speaker or wireless headphones. A change in a functional module usually refers to a change in the state of the service content or application capabilities carried by the device. The first attribute information is used to characterize the specific features of this changed functional module, such as the application scenario, the target user group, and the parameters of the functional module itself.

[0037] In actual operation, the backend maintains a lightweight monitoring mechanism. Once a change event of a functional module is detected, it automatically extracts the feature tags closely related to that function as the primary attribute information. That is, it first knows exactly what function has changed, who is using this function and in what scenario, and what configuration parameters it has.

[0038] For example, when a new "children's bedtime story" service is automatically launched in the background of a smart speaker, the system will immediately detect this change and extract the corresponding primary attribute information, which is that the usage scenario is "bedroom before bedtime", the target audience is "children", and the function module's own parameter is "low volume playback".

[0039] In operation S120, target constraint rules are generated based on the first attribute information. These target constraint rules set boundary conditions for the generated wake word in terms of acoustic characteristics, vocabulary length, or semantic direction, ensuring that the final generated vocabulary conforms to human pronunciation habits and accurately matches the tone and parameter characteristics of the corresponding function. Specifically, the extracted first attribute information is logically transformed and mapped into constraints that the generation algorithm or model can understand. This essentially defines a specific range for the generation of the wake word, avoiding the output of invalid words that deviate from the functional positioning or do not meet acoustic requirements. For example, for the aforementioned "children's bedtime story" function, combining the extracted demographic and scene attributes with the "low volume playback" function parameter, the generated constraint rules could be: "vocabulary length limited to three to four syllables, pronunciation needs to be soft to match low volume output, and semantically should have a childlike or sleep-inducing connotation."

[0040] In operation S130, based on the target constraint rules, at least one target wake-up word that satisfies the target constraint rules is generated by the first model. The target wake-up word is used to wake up the functional module. The first model includes, but is not limited to, lightweight neural network models with text generation capabilities, large language models, etc., and the target wake-up word is a dedicated trigger password for the functional module that has undergone changes.

[0041] During this operation, the pre-defined target constraint rules are input into the first model. The model then automatically generates and outputs a custom wake word that meets the requirements, based on the boundary conditions constrained by the target constraint rules. This can be understood as converting functional attributes into specific customized requirements and inputting them into the model, which then generates a matching wake word accordingly. For example, after receiving the constraint rule of "three to four syllables, soft pronunciation, playful or sleep-inducing," the first model calculates and outputs "Goodnight Baby" as the target wake word. When the user utters this wake word, the corresponding children's bedtime story function is triggered.

[0042] According to the embodiments of this disclosure, since the wake word is dynamically generated following the functional module, it is equivalent to configuring a dedicated trigger entry for each specific function. This changes the mode of the device relying on a single fixed wake word globally. It can tailor the most suitable wake word according to the specific application scenario, target audience, and other characteristics of the functional module, making voice interaction more natural and in line with specific scenarios. Moreover, it provides independent and precise trigger entry for different functional modules, effectively improving the device's front-end wake-up adaptation capability during the functional iteration process.

[0043] Figure 2 Another flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0044] like Figure 2As shown, based on the aforementioned embodiments, operation S120 may include operation S210.

[0045] In operation S210, based on the first attribute information, a target prompt word is generated as a target constraint rule. The target prompt word can constrain the output content of the first model to have a semantic association with the first attribute information. The target prompt word usually refers to a piece of natural language text or a specific formatted instruction, used to guide and regulate the output direction of the generative model (such as a large language model). Semantic association means that the generated wake-up word must be directly linked to the attribute features of the functional module in its literal meaning or associative meaning, and cannot be meaningless random characters or general words.

[0046] Specifically, the extracted discrete attribute tags (such as scene, crowd, etc.) are concatenated, transformed, or fitted with preset natural language templates to construct a complete prompt word. This prompt word clearly tells the model what direction of content to generate. By using prompt word engineering to input specific customized requirements into the model, the output words are constrained to be semantically closely related to the function, so that users can intuitively perceive the corresponding service content through the wake word.

[0047] For example, suppose the extracted first attribute information is "Scene: Kitchen", "Target Audience: Housewives", and "Function: Recipe Broadcast". Filling this information into a preset template, the generated target prompt word could be: "Please generate a voice wake-up word for a recipe broadcast function in a kitchen setting, targeting housewives. The wake-up word should be semantically highly related to cooking, food, or a kitchen assistant, and should be 3-4 characters long." After receiving this prompt word, the first model might output the wake-up word "Food Assistant".

[0048] According to embodiments of this disclosure, by converting functional attribute information into target prompt words, the powerful understanding capabilities of generative models of natural language instructions can be fully utilized, eliminating the tedious process of writing complex rule codes. Furthermore, by forcibly constraining semantic relevance in the prompt words, it is ensured that the generated exclusive wake words carry "functional cues," thus establishing an intuitive logical connection between the wake words and specific functions. When users need to use a certain function, they can naturally associate it with the corresponding wake word, reducing the user's memory burden and improving the intuitiveness and ease of use of voice interaction.

[0049] Figure 3 Another flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0050] like Figure 3 As shown, based on the foregoing embodiments, operation S110 may include at least one of operations S310 to S320.

[0051] When operating S310, in response to detecting the addition of a first functional module to the target device, the first attribute information corresponding to the first functional module is obtained. Adding a first functional module typically involves the target device deploying and activating a new service or application capability that it did not previously possess through methods such as system firmware upgrades, app store downloads, or silent cloud pushes.

[0052] Specifically, the device monitors application installation or service registration events at the underlying device level in real time. Once a new functional module is confirmed to be integrated, it will proactively parse the module's configuration file or metadata to extract initial feature tags such as scenario, target audience, or configuration parameters, using these as primary attribute information to ensure that the new function has the data foundation to generate a unique wake word upon deployment.

[0053] For example, when a smart speaker first installs the "elderly medication reminder" service via cloud push, it detects the newly added first functional module, then parses its configuration to obtain the first attribute information: the target audience is "elderly people", the application scenario is "daily health", and the function parameter is "timed voice broadcast".

[0054] When operating the S320, in response to detecting a function update corresponding to the second functional module configured on the target device, the system obtains the first attribute information corresponding to the updated second functional module. The function update corresponding to the second functional module can be an update resulting from version iterations, feature expansions, or modifications to core operating parameters of services already existing and running on the device. Specifically, the device continuously tracks the version number or configuration status of existing modules. When an upgrade or change event is detected, the module is re-evaluated, and the updated feature tags are extracted. The purpose of this step is that function updates are often accompanied by shifts in applicable scenarios or audiences; therefore, it is essential to obtain the latest attribute information to prevent subsequently generated constraint rules from becoming disconnected from actual functionality.

[0055] For example, the original "normal weather broadcast" module (i.e. the second functional module) of the device has been upgraded to include the "extreme weather strong warning" function. After detecting this update, its attributes are re-extracted, and the first attribute information is obtained with the updated application scenario as "emergency avoidance" and the function parameter as "forced broadcast at the highest volume".

[0056] According to the embodiments of this disclosure, by using the "addition" and "update" of functional modules as independent triggering conditions for obtaining attribute information, it is ensured that whether it is the introduction of new capabilities or the iterative upgrade of existing capabilities, the substantial changes in functional characteristics can be automatically captured in the first time, eliminating the tediousness of manual intervention or reconfiguration by users, and further ensuring that the constraint rules and wake words generated subsequently can always remain synchronized with the current functional state of the device.

[0057] Figure 4 Another flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0058] like Figure 4 As shown, based on the foregoing embodiments, the device wake-up word generation method may further include at least one of operations S410 to S430.

[0059] When operating the S410, a target wake-up word corresponding to the first functional module is added to the wake-up word set of the target device. The wake-up word set can be a wake-up word library stored in the target device's local audio processing chip or front-end memory, used to wake up various functional modules. After the first model generates a dedicated target wake-up word for the newly added first functional module, it writes the word into the aforementioned set and establishes a call mapping relationship between the word and the new function at the underlying level. From an application perspective, the device's front-end monitoring engine has officially loaded a completely new trigger entry point and begins to have the ability to recognize and respond to new functions.

[0060] For example, following the newly added "elderly medication reminder" service in the aforementioned embodiments, once the model generates the target wake word "health assistant," it will directly add "health assistant" to the device's wake word set. Subsequently, the device can capture this word during daily monitoring and activate the medication reminder function.

[0061] In operation S420, the wake word corresponding to the second functional module in the wake word set of the target device is replaced with the target wake word corresponding to the second functional module. This replacement operation essentially overwrites or refreshes existing mapping entries in the wake word set. For the updated second functional module, due to changes in its attributes leading to the regeneration of a more suitable target wake word, the old entries in the wake word set are located and replaced with the newly generated words. In other words, old words that no longer fit the current functional tone are eliminated, preventing users from experiencing cognitive biases or triggering ambiguity when using the updated function.

[0062] For example, in the scenario described in the previous embodiment where "normal weather forecast" is upgraded to "extreme weather warning," the original "weather forecast" in the wake word set will be replaced with the newly generated target wake word "emergency weather." After the replacement, the user needs to use "emergency weather" to trigger the upgraded module, and the old word will no longer be effective.

[0063] When operating the S430, in response to the detection that a third functional module configured on the target device is offline, the target wake word corresponding to the third functional module is deleted from the wake word set of the target device. The offline status of a third functional module typically refers to a state where the application on the device is uninstalled, a time-limited service expires, or it is forcibly disabled by the cloud. Once a function deregistration event is detected, the wake word set is simultaneously cleaned up, completely removing the target wake word bound to this invalid function. This not only frees up limited front-end memory resources but also prevents meaningless false wake-ups when the user says the old wake word.

[0064] For example, when the "Christmas Special Interaction" module configured on the device automatically goes offline after the holiday (i.e., the third functional module goes offline), the corresponding target wake word "Merry Christmas" will be immediately removed from the wake word set. Afterward, the device will no longer perform feature matching or monitoring on that word.

[0065] According to the embodiments of this disclosure, by dynamically maintaining the wake word set through "addition, replacement, and deletion," it is ensured that the wake word library is always strictly aligned with the actual capabilities of the device, effectively avoiding the problem of false triggering caused by "zombie wake words" left over after the old functions are discontinued or updated; at the same time, timely cleaning up invalid words releases the memory and computing resources of the front-end audio processing chip, enabling the device to maintain efficient and accurate voice monitoring performance under limited hardware conditions.

[0066] Figure 5 Another flowchart illustrating a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0067] like Figure 5 As shown, based on the aforementioned embodiments, the device wake-up word generation method may include operation S510, and operation S120 may include operation S520.

[0068] When operating S510, the system retrieves the second attribute information corresponding to each of the currently configured functional modules on the target device. The currently configured functional modules refer to existing services or applications on the target device that are already running in the background or active, excluding the functional modules that have been changed. Similar to the first attribute information, the second attribute information is used to characterize the specific features of these existing functional modules, such as application scenarios, target users, and functional parameters. During this operation, the system registry or service list of the target device is traversed to extract the feature tags of all existing functional modules. This process is equivalent to performing a global scan and assessment of the device's overall functional ecosystem and operating environment before processing newly changed functional modules, in order to obtain the device's existing contextual information.

[0069] For example, assuming the smart speaker is currently configured with two functional modules, "music playback" and "smart home control", it will obtain their second attribute information respectively. For example, the scenario corresponding to "music playback" is "daily entertainment" and the scenario corresponding to "smart home control" is "home appliance control".

[0070] In operation S520, target constraint rules are generated based on the first attribute information and the second attribute information. After introducing the second attribute information, the logic for generating target constraint rules can include two different implementation paths.

[0071] The first approach focuses on conflict prevention through single-node generation, generating target constraint rules only for the functional module undergoing modification. However, during generation, the second attribute information is used as an exclusive reference, requiring the new rules to avoid the feature ranges of existing functions at acoustic or semantic boundaries. The second approach focuses on overall differentiation through batch generation, combining the first and second attribute information to generate corresponding target constraint rules for both the modified functional module and each existing configured functional module.

[0072] Specifically, in the first approach, the second attribute information will be transformed into a reverse constraint and added to the rule corresponding to the first attribute information to define a safe generation boundary; in the second approach, the attribute information of all modules will be compared and analyzed together, and non-overlapping constraint boundaries will be redefined or confirmed for each functional module, outputting multiple independent constraint rules so that a dedicated wake word can be generated or updated for each module in the future.

[0073] For example, for the first method: a new "children's bedtime story" (first attribute information) has been added, while the device already has "adult sleep aid white noise" (second attribute information). The constraint rules generated for "children's bedtime story" not only require the wake word to be "childlike", but also explicitly add the restriction that "it must be significantly different from the adult sleep aid features in terms of semantics and pronunciation".

[0074] For example, regarding the second approach: Suppose the device has added a "children's bedtime story" function (first attribute information: children, sleep scene), and the "smart home appliance control" function is currently configured (second attribute information: adults, whole-house command scene). By comparing the attribute information of these two, two sets of rules with significant differences are generated for them: The constraint rule generated for "children's bedtime story" is limited to "the semantics must contain small animals or fairy tale characters, the pronunciation must be fluent and soft, and the length must be 3 to 4 syllables"; at the same time, the constraint rule generated for "smart home appliance control" is limited to "the semantics must contain command words such as 'butler' or 'device,' and the length is strictly limited to 2 syllables."

[0075] According to embodiments of this disclosure, by incorporating the second attribute information of the currently configured functional modules of the device when generating constraint rules, the generation of wake words is no longer an isolated single-point operation, avoiding overlap in pronunciation or semantics between newly generated dedicated wake words and existing functional wake words on the device. Whether preventing conflicts through exclusive restrictions or generating independent constraint rules for each module to maintain overall distinctiveness, both methods reduce the cross-false wake-up rate between different functional modules within the device, ensuring the accuracy of voice interaction in multi-functional concurrent scenarios.

[0076] Figure 6 The flowchart illustrating the generation of a target wake word in a device wake word generation method according to an embodiment of the present disclosure is shown schematically.

[0077] like Figure 6 As shown, based on the aforementioned embodiments, operation S120 may include operations S610 to S620.

[0078] When operating the S610, based on the target constraint rules, the first model generates multiple candidate wake words that satisfy the target constraint rules. Candidate wake words refer to the set of alternative words initially output by the generative model that conform to the constraint rules literally or semantically, but have not yet undergone final usability and security verification. In the actual generation process, to improve the success rate of the final word selection and provide a certain degree of diversity, the model is usually not restricted to outputting only a single result. Instead, the first model is controlled to generate and output a batch of words at once according to the input target constraint rules, thereby constructing a candidate pool.

[0079] For example, regarding the constraint rules of "children's bedtime stories", the first model, through reasoning, outputs multiple candidate wake words at once, such as "goodnight baby", "baby sleep", "sleep well" and "goodnight little bear", forming a candidate set.

[0080] In operation S620, conflict detection is performed between candidate wake words, and / or between candidate wake words and wake words of the target device, to determine at least one target wake word without conflict. Conflict detection refers to the verification process of evaluating whether different words are too similar in pronunciation features, text structure, or semantic expression (leading to cross-functional triggering ambiguity), or whether there are contradictions or serious discrepancies in semantic tendency (leading to cognitive confusion under the same function). It should be clarified that if multiple candidate words with similar pronunciation or semantics point to the same functional module and have the same semantic direction, this usually does not constitute a conflict and can be retained to increase the flexibility of wake-up; however, due to the uncontrollability of the large language model generation, if the model's different understanding of the emphasis of the prompt words leads to semantically opposite or seriously deviated candidates under the same function (i.e., internal semantic conflict), they also need to be eliminated. The wake words of the target device refer to the existing wake words that are currently active and in listening state of the device.

[0081] This operation can be "internal detection," which involves comparing each candidate word pairwise within the candidate pool, focusing on detecting semantic contradictions or serious deviations (i.e., semantic conflicts) between candidate words, while also removing words whose pronunciation might cause misjudgment by the underlying engine. Alternatively, it can be "external detection," which compares candidate words across sets with existing wake words on the device, intercepting and removing those that are likely to "collide" in acoustic features or cause cognitive ambiguity for users (e.g., the meaning of a new word might easily evoke associations with old functions). After one or more layers of filtering, the remaining wake words are confirmed as target wake words. It should be noted that the final target wake word can be one or multiple non-conflicting words, thus providing a single or multiple parallel trigger entry points for the same functional module.

[0082] For example, suppose that for the newly added "Children's Bedtime Story" function (attribute requirements: low volume, sleep aid), the large model generates candidate words such as "Goodnight Baby" (emphasizing sleep), "Sleep Well" (emphasizing sleep), and "Happy Story Time" (emphasizing story, but semantically leaning towards excitement and activity) due to different focuses. Meanwhile, the device already has the existing wake word "I Want to Sleep" for the "Adult Sleep Aid" function. During internal testing, it was found that "Happy Story Time" and "Goodnight Baby" have a significant internal semantic conflict (activity contradicts sleep aid), so "Happy Story Time" was removed. ("Goodnight Baby" and "Sleep Well" are semantically consistent and not contradictory, so they were retained). Subsequently, external testing revealed that the candidate word "Sleep Well" is too similar in pronunciation to the existing "I Want to Sleep" (acoustic conflict), which could easily lead to false triggering across functions, so "Sleep Well" was also removed. Finally, "Goodnight Baby," which has no internal semantic conflicts or external cross-functional conflicts, was safely retained as the target wake word for this children's bedtime story function.

[0083] According to the embodiments of this disclosure, by introducing a screening mechanism of "first generating multiple candidate pools and then performing conflict detection", the fault tolerance rate of wake word generation is improved. More importantly, through strict internal and external dual anti-conflict filtering, on the one hand, internal semantic contradiction words caused by misunderstanding of the large model are eliminated, ensuring the tone consistency of single function trigger commands; on the other hand, the high distinguishability between newly generated wake words and existing wake words of the device is guaranteed, effectively avoiding cross-wake-up at the device level due to similar pronunciation, and eliminating user cognitive ambiguity caused by semantic ambiguity or contradiction, further improving the stability and ease of use of voice interaction in multi-functional concurrent scenarios.

[0084] Figure 7 The flowchart illustrating the process of determining at least one target wake-up that does not conflict in a device wake-up word generation method according to an embodiment of the present disclosure is shown.

[0085] like Figure 7 As shown, based on the aforementioned embodiments, operation S620 may include operations S710~S720.

[0086] When operating S710, in response to the detection of a conflict between the candidate wake word and the target device's wake word, the target constraint rules are adjusted based on the conflict content. The conflict content typically refers to the specific reasons or feature dimensions that cause confusion between the candidate wake word and the device's existing wake words, such as local high overlap at the pinyin / phoneme level or semantic expression conflict. Adjusting the target constraint rules involves transforming these specific factors causing the conflict into reverse constraints (such as "negative prompt words" or exclusive parameters), adding or updating them to the original constraint rules. This is equivalent to explicitly telling the large model "not to generate content containing specific features," thus defining a more precise and safer boundary for the next generation.

[0087] For example, suppose the candidate word "baby sleep" generated for the children's story function conflicts with the existing stock word "I want to sleep" on the device. The analysis shows that the conflict is that "the pronunciation of the word 'sleep' at the end overlaps, and the first two syllables both contain the ending sound 'oh,' indicating a high degree of pronunciation overlap." Based on this, the target constraint rules are adjusted. In addition to the original rule of "meeting the children's bedtime scenario," a clear reverse restriction is added: "The generated wake word cannot contain the word 'sleep' at the end, and cannot contain words with similar pronunciations such as the initial syllable vowel 'o' (not an English letter, but a Chinese pinyin letter, pronounced 'oh') or 'ao' (not an English letter, but a Chinese pinyin letter, pronounced 'ao')."

[0088] When operating the S720, based on the adjusted target constraint rules, at least one target wake word is regenerated through the first model that satisfies the target constraint rules and does not conflict with any wake words of the target device. Since the adjusted target constraint rules already contain explicit exclusivity instructions, the attention mechanism of the first model actively avoids acoustic or semantic pitfalls that previously caused conflicts during regeneration. This iterative generation method based on negative feedback ensures that the newly generated words, while satisfying functional attributes, naturally possess distinguishability from existing words, thus successfully passing subsequent conflict detection and becoming truly usable target wake words.

[0089] According to the embodiments of this disclosure, by accurately extracting the conflict issues and transforming them into reverse constraints, the model can be intelligently guided to bypass easily confused word regions, effectively overcoming the uncontrollability that may exist in the single generation of generative models, avoiding getting stuck in a dead loop or wasting computing power due to repeated blind retries, and thus efficiently and stably outputting highly distinguishable wake words without human intervention.

[0090] Figure 8 A flowchart illustrating a wake-up function module according to an embodiment of the present disclosure is shown schematically.

[0091] like Figure 8As shown, based on the aforementioned embodiments, the target device includes a first voice acquisition module and a second voice acquisition module. The first voice acquisition module continuously acquires voice information from the environment, while the second voice acquisition module consumes more power than the first. The first voice acquisition module typically refers to an audio monitoring component configured as low-power, always-on in the target device; the second voice acquisition module refers to an audio processing component that is in a dormant or truncated state but possesses higher sound pickup quality or data transmission capabilities when activated. In a specific hardware architecture, the first voice acquisition module is a single or a few low-power microphones on the device, primarily responsible for continuously picking up ambient noise in standby mode; while the second voice acquisition module is a complete pickup array containing multiple microphones, requiring more power to provide high-quality far-field voice acquisition capabilities such as beamforming and noise reduction.

[0092] For example, when a smart speaker is in standby mode, it only turns on one low-power microphone (the first voice acquisition module) to continuously listen to ambient sounds, while the circular array of six microphones (the second voice acquisition module) is in a power-off sleep state to save energy.

[0093] The process of waking up the functional module may include operations S810~S830.

[0094] In operation of S810, in response to the detection that the first voice information collected by the first voice acquisition module matches the target wake-up word, the second voice acquisition module is activated to collect the second voice information in the environment. The first voice information is the ambient audio stream continuously collected by the low-power microphone in a normally-on state. The second voice information is the audio data collected by the second voice acquisition module after it is triggered and activated. Since the second voice acquisition module is activated only after detecting the wake-up word, it collects the natural voice command input by the user immediately after the wake-up word. When the continuously working first voice acquisition module listens to a voice stream that matches the acoustic characteristics of the target wake-up word, it determines that the user has initiated a clear interaction intention. At this time, the target wake-up word only serves as a trigger condition for front-end wake-up, powering the second voice acquisition module, which is in a dormant state, and activating it to obtain higher quality and more complete voice data from the user.

[0095] For example, when a continuously operating single microphone hears the user say "Goodnight, baby" (target wake word), it immediately powers and activates the 6-microphone array, which then picks up the user's next words, "Tell me a story about Snow White" (second voice information).

[0096] When operating the S820, the second model processes the second voice information to obtain the target command. The second model, typically a large language model and accompanying speech recognition model deployed on the device's local main control chip or in the cloud, possesses powerful natural language processing and semantic understanding capabilities. The target command is the specific execution parameters or structured control signaling generated by the second model after parsing the user's intent, used to schedule underlying functional modules. Since the front-end wake-up matching stage only handles the underlying acoustic triggering and does not involve complex calculations such as semantic understanding and task classification, the user's subsequent complete natural speech input (the second voice information) will be transmitted to the back-end and entirely processed by the second model. The second model will transcribe the speech, recognize intent, and decompose tasks, ultimately outputting machine-readable instructions that clearly specify which function needs to be invoked and what specific task that function should perform.

[0097] When operating the S830, based on the target command, the functional module corresponding to the target wake-up word in the target device is activated. Since the front-end wake-up matching stage is only responsible for the underlying acoustic triggering and does not involve complex calculations such as semantic understanding and task classification, the user's subsequent complete natural speech (second speech information) will be transmitted to the back-end and handled entirely by the second model. Continuing the previous example, after receiving the target command, the "Children's Bedtime Story" functional module on the device is officially activated and started, and "Play Snow White" is passed to the module as a parameter. The module then begins to play the corresponding audio content for the user.

[0098] According to the embodiments of this disclosure, by continuously listening with low power consumption, a high-power voice acquisition module is activated to collect subsequent instructions after a wake-up word is detected. This significantly reduces the hardware power consumption of the device during daily standby, while fully leveraging the advantages of the large language model in natural language processing, enabling users to express their intentions in the most natural spoken language after waking up the device.

[0099] As a parallel implementation, based on the aforementioned embodiments, the target device includes a voice acquisition module and a voice transmission module. The voice acquisition module continuously acquires voice information from the environment, and the voice transmission module, upon detecting a target wake-up word, transmits the voice information acquired by the voice acquisition module from the target wake-up word to the second model. The voice acquisition module typically refers to a physical sensor such as a microphone at the front end of the target device, responsible for converting acoustic signals into electrical signals; the voice transmission module refers to the data bus, network interface, and related transmission control logic connecting the front-end sensors to the back-end main control CPU or cloud.

[0100] At this time, the front-end voice acquisition module is constantly acquiring data to monitor ambient sounds; however, the voice transmission module responsible for sending data to the back-end is either off or logically truncated in its default standby state. Only when the local front-end chip detects the target wake word will this "data pipeline" be opened, and the user's voice following the wake word will be streamed to the second model in the back-end for processing.

[0101] For example, the microphone (voice acquisition module) of a smart speaker is always receiving ambient sounds, but when in standby mode, the high-speed audio data bus (voice transmission module) leading to the main control CPU is turned off. At this time, ambient sounds only circulate in the front-end low-power DSP chip for wake-word comparison and are not reported to the CPU.

[0102] The process of waking up the functional module may include: in response to the detection that the voice information collected by the voice acquisition module matches the target wake-up word, starting the voice transmission module to transmit subsequent voice information (including the target wake-up word) collected from the target wake-up word; processing the transmitted subsequent voice information based on the second model to obtain the target instruction; and waking up the functional module in the target device corresponding to the target wake-up word based on the target instruction.

[0103] In the above process, the definition of subsequent voice information, the processing of the second model, and the specific logic of the wake-up function module are basically the same as in the aforementioned embodiment. The difference lies in that when the audio stream captured by the voice acquisition module is successfully matched at the front end, the target wake-up word acts as a "password" to release the data bus from its interception state. Once the voice transmission module is activated, the target wake-up word, the subsequent natural language parsing work, and the activation process of the corresponding function module are all the same as the aforementioned process in which the second model takes over the processing and issues the target command, and will not be repeated here.

[0104] For example, when the microphone picks up "Goodnight, baby" and the front-end DSP successfully matches it, it immediately connects the audio bus to the CPU (starting the voice transmission module). This bus then transmits the user's next words, "Tell me a Snow White story," to the back-end CPU in real time. The second model on the CPU generates the target instruction through semantic parsing and starts the "Children's Bedtime Story" process in the background to execute the playback action.

[0105] As described in the aforementioned embodiments, the dynamically generated target wake-up word serves only as the sole trigger credential for front-end wake-up. Its function is strictly limited to low-level operations such as activating high-power microphone hardware and opening the voice reception channel. Complex subsequent tasks, including semantic understanding, task classification, instruction parsing, and function control, are entirely handled by the back-end second model. This avoids the device uploading a large amount of invalid environmental audio to the cloud or main control unit when not explicitly woken up, thus saving network bandwidth and system computing power and reducing invalid responses caused by misidentification. Simultaneously, it avoids the drawbacks of indiscriminate continuous microphone pickup and full upload, eliminating the risk of long-term eavesdropping or leakage of everyday private environmental sounds.

[0106] Figure 9 Another flowchart of a wake-up function module according to an embodiment of the present disclosure is shown schematically.

[0107] like Figure 9 As shown, based on the aforementioned embodiments, the target device includes a first voice acquisition module. The first voice acquisition module continuously acquires voice information in the environment. The process of waking up the functional module may include operations S910 to S920.

[0108] When operating the S910, the first voice information acquired by the first voice acquisition module is processed based on the third model to obtain the target result. The third model is used to identify whether the semantics of the first voice information matches the target wake word. The third model typically refers to an algorithm model with natural language understanding and semantic generalization capabilities, such as a lightweight large language model, a semantic similarity calculation model, or an intent recognition model. The target result refers to the judgment conclusion output by the third model, representing whether the user's current voice expression's true intent points to a specific wake word in the device.

[0109] Unlike conventional voice wake-up, which requires users to pronounce words exactly as preset words ("hard acoustic matching"), the first voice acquisition module captures the user's pronunciation and then hands it over to a third model for deep semantic analysis. The third model extracts the core meaning of the user's expression and calculates the semantic similarity between this meaning and the target wake-up word currently set by the device. As long as the user's expression is semantically similar to the target wake-up word, the third model can determine that the match is successful and output a positive result.

[0110] For example, the target wake word dynamically generated for the children's story function is "Baby sleep". The first voice message actually spoken by the user is "Baby needs to rest" or "Turn on sleep-inducing mode". After processing these voice messages, the third model recognizes that their core semantics are highly consistent with "Baby sleep", and thus outputs the target result of "semantic matching successful".

[0111] When operating the S920, in response to the semantic matching of the first voice information representing the target result with the target wake-up word, the functional module in the target device corresponding to the target wake-up word is activated. When the third model confirms that the user's voice is semantically equivalent to saying the target wake-up word, the wake-up condition is considered to be met. At this time, the underlying functional module or service process bound to the target wake-up word will be activated and launched in the background directly according to the semantic mapping relationship.

[0112] For example, continuing from the previous example, when the third model outputs the target result that semantically matches "Baby needs to rest" with the target wake word "Baby sleeps", it directly responds by waking up and launching the "Children's Bedtime Story" function module bound to "Baby sleeps" in the background.

[0113] Based on the foregoing embodiments, the wake-up function module process may further include: in response to a change in the wake-up word set of the target device, retraining the third model based on the changed wake-up word set. A change in the wake-up word set refers to a change in the word set, such as the dynamic generation of new target wake-up words due to added features on the device, or the deletion of old wake-up words, resulting in a change in the word library. Since the third model performs semantic comparison based on the current wake-up word set of the device, when a new, specific wake-up word is dynamically generated and added to the set, the third model needs to synchronously update its cognitive "benchmark." This involves inputting the new wake-up word and its possible corresponding generalized expressions as new samples or context into the third model, enabling it to recognize various semantic variations of the new wake-up word.

[0114] For example, the device's original wake word set only contained "I want to listen to music," but later, a target wake word, "Goodnight Bear," specifically for children, was dynamically generated and added. After the wake word set changed, "Goodnight Bear" and its related semantic expressions (such as "Little Bear Sleeping" and "Soothing Little Bear") were used as incremental data to fine-tune the training of the third model or update the knowledge base. After the update, when the user says "Little Bear Sleeping," the third model can accurately identify that it matches the newly added "Goodnight Bear."

[0115] According to embodiments of this disclosure, by introducing a third model that allows users to use more natural, colloquial, or even ambiguous expressions to wake up specific functions of the device, the fault tolerance of voice interaction and user experience are improved. Simultaneously, in conjunction with a model adaptive update mechanism for a dynamic wake-up word set, a stable and flexible wake-up experience based on semantic understanding is ensured even as new wake-up words are continuously generated.

[0116] Figure 10 A block diagram of a device wake word generation apparatus according to an embodiment of the present disclosure is shown schematically.

[0117] like Figure 10As shown, the device wake-up word generation device 1000 may include an acquisition module 1010, a first generation module 1020, and a second generation module 1030.

[0118] The acquisition module 1010 is used to acquire the first attribute information corresponding to the changed functional module in response to detecting a change in the functional module configured in the target device. In some embodiments, the acquisition module 1010 can be used to perform operation S110 in the device wake-word generation method described above, which will not be elaborated here.

[0119] The first generation module 1020 is used to generate target constraint rules based on the first attribute information. In some embodiments, the first generation module 1020 can be used to perform operation S120 in the device wake word generation method described above, which will not be elaborated here.

[0120] The second generation module 1030 is used to generate at least one target wake-up word that satisfies the target constraint rules through the first model based on the target constraint rules. The target wake-up word is used to wake up the functional module. In some embodiments, the second generation module 1030 can be used to perform operation S130 in the above-described device wake-up word generation method, which will not be described in detail here.

[0121] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-a-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0122] For example, any plurality of the acquisition module 1010, the first generation module 1020, and the second generation module 1030 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits can be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of this disclosure, at least one of the acquisition module 1010, the first generation module 1020, and the second generation module 1030 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, at least one of the acquisition module 1010, the first generation module 1020, and the second generation module 1030 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0123] It should be noted that the data processing system part in the embodiments of this disclosure corresponds to the data processing method part in the embodiments of this disclosure. The specific description of the data processing system part is referred to in the data processing method part, and will not be repeated here.

[0124] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 11 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0125] like Figure 11 As shown, an electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage portion 1108 into a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0126] RAM 1103 stores various programs and data required for the operation of electronic device 1100. Processor 1101, ROM 1102, and RAM 1103 are interconnected via bus 1104. Processor 1101 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1102 and / or RAM 1103. It should be noted that the programs may also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0127] According to embodiments of this disclosure, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to a bus 1104. The electronic device 1100 may also include one or more of the following components connected to the input / output (I / O) interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1110 as needed so that computer programs read from it can be installed into the storage section 1108 as needed.

[0128] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1109, and / or installed from removable medium 1111. When the computer program is executed by processor 1101, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0129] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0130] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0131] For example, according to embodiments of this disclosure, a computer-readable storage medium may include one or more memories other than the ROM 1102 and / or RAM 1103 described above and / or ROM 1102 and RAM 1103.

[0132] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the device wake word generation method provided in the embodiments of this disclosure.

[0133] When the computer program is executed by the processor 1101, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0134] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices or magnetic storage devices. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via communication section 1109, and / or installed from removable medium 1111. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof. According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code may be executed entirely on a user computing device, partially on a user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to user computing devices via any type of network, including local area networks (LANs) or wide area networks (WANs), or they can be connected to external computing devices (e.g., via the Internet using an Internet service provider).

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0136] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for generating a device wake word, comprising: In response to the detection of a change in the functional modules configured on the target device, the first attribute information corresponding to the changed functional module is obtained; Based on the first attribute information, generate target constraint rules; Based on the target constraint rules, at least one target wake-up word that satisfies the target constraint rules is generated by the first model, and the target wake-up word is used to wake up the functional module.

2. The method according to claim 1, wherein generating the target constraint rule based on the first attribute information includes: Based on the first attribute information, target prompt words are generated as target constraint rules. The target prompt words can constrain the output content of the first model to have a semantic association with the first attribute information.

3. The method according to claim 1, wherein the step of obtaining the first attribute information corresponding to the changed functional module in response to detecting a change in the functional module configured in the target device includes at least one of the following: In response to detecting the addition of a first functional module to the target device, the first attribute information corresponding to the first functional module is obtained; In response to detecting a function update corresponding to the second functional module configured in the target device, the first attribute information corresponding to the updated second functional module is obtained.

4. The method of claim 3, further comprising at least one of the following: Add the target wake-up word corresponding to the first functional module to the wake-up word set of the target device; Replace the wake word corresponding to the second functional module in the wake word set of the target device with the target wake word corresponding to the second functional module; In response to detecting that the third functional module configured in the target device is offline, the target wake word corresponding to the third functional module is deleted from the wake word set of the target device.

5. The method according to claim 1, further comprising: Obtain the second attribute information corresponding to each of the currently configured functional modules of the target device; The step of generating target constraint rules based on the first attribute information includes: Based on the first attribute information and the second attribute information, target constraint rules are generated.

6. The method according to claim 1, wherein generating at least one target wake word satisfying the target constraint rule through a first model based on the target constraint rule comprises: Based on the target constraint rules, a first model generates multiple candidate wake words that satisfy the target constraint rules; Conflict detection is performed between the candidate wake words, and / or conflict detection is performed between the candidate wake words and the wake words of the target device to determine at least one target wake word that does not conflict.

7. The method of claim 6, wherein determining at least one target wake word that does not conflict comprises: In response to the detection of a conflict between the candidate wake word and the wake word of the target device, the target constraint rules are adjusted based on the content of the conflict; Based on the adjusted target constraint rules, at least one target wake word is regenerated through the first model that satisfies the target constraint rules and does not conflict with the wake words of the target device.

8. The method according to claim 1, wherein the target device includes a first voice acquisition module and a second voice acquisition module, the first voice acquisition module continuously acquires voice information in the environment, the power consumption of the second voice acquisition module is greater than that of the first voice acquisition module, and the process of waking up the functional module includes: In response to the detection that the first voice information collected by the first voice acquisition module matches the target wake word, the second voice acquisition module is activated to collect the second voice information in the environment; The second speech information is processed based on the second model to obtain the target instruction; Based on the target instruction, the functional module in the target device corresponding to the target wake word is activated.

9. The method according to claim 1, wherein the target device includes a first voice acquisition module, the first voice acquisition module continuously acquires voice information in the environment, and the process of waking up the functional module includes: The first speech information acquired by the first speech acquisition module is processed based on the third model to obtain the target result. The third model is used to identify whether the semantics of the first speech information matches the target wake word. In response to the target result indicating that the semantics of the first voice information matches the target wake word, the functional module in the target device corresponding to the target wake word is activated.

10. An electronic device, comprising: At least one processor; as well as The memory connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the following operations: In response to detecting a change in the functional modules configured in the target device, the system acquires first attribute information corresponding to the changed functional modules; based on the first attribute information, it generates target constraint rules; based on the target constraint rules, it generates at least one target wake-up word that satisfies the target constraint rules through a first model, and the target wake-up word is used to wake up the functional modules.

Images