Method and device for identifying operation intention, storage medium and electronic device

Abstract

The present application discloses a method and device for identifying operation intentions, a storage medium, and an electronic device, relating to the field of smart home technology. The method comprises: identifying device functions and multiple target keywords carried in a target voice; combining the device functions and the multiple target keywords to construct a reference operation intention; and screening out an operation intention with a credibility higher than a target threshold from the reference operation intentions as a target operation intention corresponding to the target voice. The above technical solution solves the problem of low efficiency in identifying operation intentions carried in voice in related technologies.

Description

Technical Field

[0001] The present application relates to the field of smart home technology, and more specifically, to a method and device for identifying operation intentions, a storage medium, and an electronic device. Background Art

[0002] With the advancement of technology, home appliances are becoming increasingly intelligent. By controlling home appliances, they can help people complete their operations. Voice control, as a common method of device interaction, plays a key role in home appliance control. By interacting with home appliances through voice, control commands are sent to them. The appliances recognize the commands and respond to them by performing the corresponding operations, thus achieving control of the home appliances.

[0003] Currently, human-computer voice interaction control in the industry only supports control with a single intent, that is, when an intelligent device is instructed to perform an operation through voice at one time, in order to ensure the accuracy of a single intent, an overall domain classifier is usually first established, and then each sub-domain is accurately classified. For example, we have three major categories: air conditioners, refrigerators, washing machines, and water heaters. Air conditioners have four device functions: power on, power off, temperature, and mode setting. Refrigerators have four device functions: power on, power off, setting cabin temperature, and food management. Washing machines have five device functions, and water heaters have six device functions. Therefore, the recognition process is to first perform domain classification to distinguish whether it is an air conditioner, refrigerator, washing machine, or water heater. Assuming that it is classified as an air conditioner, the air conditioner model is then used to accurately match the device functions to distinguish whether it is power on, power off, temperature, or mode. However, with people's demand for smart home appliance operations, sometimes it is necessary to control multiple home appliances at the same time, so multi-intent control of home appliances will be performed through a voice carrying multiple intentions. Obviously, the current human-computer voice interaction control method is not applicable to this multi-intent control machine. Therefore, how to achieve multi-intent control of home appliances through a voice carrying multiple intentions is a technical problem that people in the industry urgently need to solve.

[0004] There is no effective solution to the problem that related technologies have low efficiency in recognizing the operation intentions carried in voice. Summary of the Invention

[0005] The embodiments of the present application provide a method and device for identifying operation intentions, a storage medium, and an electronic device, so as to at least solve the problem of low efficiency in identifying operation intentions carried in voice in related technologies.

[0006] According to one embodiment of the embodiments of the present application, a method for identifying operation intentions is provided, including: a method for identifying operation intentions, characterized in that it includes: identifying a device function and multiple target keywords carried in a target voice, wherein the device function is used to indicate the operation type of the operation to be performed by the smart device, and the target keyword is used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled; combining the device function and the multiple target keywords to construct a reference operation intention, wherein the reference operation intention includes the operation intention formed by combining the device function and the multiple target keywords; screening out operation intentions with a credibility higher than a target threshold from the reference operation intentions as the target operation intention corresponding to the target voice, wherein the credibility is the degree of credibility corresponding to each of the reference operation intentions detected based on the expression information of the device function and the multiple target keywords in the target voice.

[0007] According to another embodiment of the embodiments of the present application, an operation intention recognition device is also provided, including: an identification module, used to recognize a device function and multiple target keywords carried in a target voice, wherein the device function is used to indicate the operation type of the operation to be performed by the smart device, and the target keyword is used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled; a combination module, used to combine the device function and the multiple target keywords to construct a reference operation intention, wherein the reference operation intention includes the operation intention formed by combining the device function and the multiple target keywords; a screening module, used to screen out operation intentions with a credibility higher than a target threshold from the reference operation intentions as the target operation intention corresponding to the target voice, wherein the credibility is the degree of credibility corresponding to each of the reference operation intentions detected based on the expression information of the device function and the multiple target keywords in the target voice.

[0008] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for identifying the operation intention during operation.

[0009] According to another aspect of an embodiment of the present application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method for identifying the operation intention through the computer program.

[0010] In an embodiment of the present application, the target speech carries target keywords for indicating the operation to be performed by the device to be controlled and the smart device to be controlled, as well as device functions for indicating the operation type of the operation to be performed by the smart device. By recognizing the target speech, the device functions and target keywords in the target speech are obtained. Then, by combining the device functions and multiple target keywords in the target speech, the reference operation intentions obtained by the combination can cover all operation intentions that can be combined with the device functions and multiple target keywords. Then, based on the credibility of each reference operation intention that can be detected based on the information expressed by the device functions and multiple functions in the target speech, the reference operation intentions are filtered according to the credibility of each reference operation intention, thereby filtering out target operation intentions with a credibility higher than a target threshold, thereby achieving rapid and accurate recognition of the target operation intentions included in the target speech. The above technical solution solves the problem of low recognition efficiency of operation intentions carried in speech in the related art, and achieves the technical effect of improving the recognition efficiency of operation intentions carried in speech. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0013] Figure 1 1 is a schematic diagram of a hardware environment for a method for identifying an operation intention according to an embodiment of the present application;

[0014] Figure 2 is a flow chart of a method for identifying operation intention according to an embodiment of the present application;

[0015] Figure 3 is an optional target speech recognition schematic diagram according to an embodiment of the present application;

[0016] Figure 4 is a schematic diagram of an optional initial recognition model according to an embodiment of the present application;

[0017] Figure 5 is a schematic diagram of an optional feature extraction layer structure according to an embodiment of the present application;

[0018] Figure 6 is a schematic diagram of constructing an optional reference operation intention according to an embodiment of the present application;

[0019] Figure 7 This is an optional target operation intention screening flow chart according to an embodiment of the present application;

[0020] Figure 8 This is a structural block diagram of a device for identifying operation intention according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] According to one aspect of the embodiment of the present application, a method for identifying an operation intention is provided. The method is widely used in smart home (Smart Home), smart home, smart home device ecology, smart home (IntelligenceHouse) ecology and other whole-house intelligent digital control application scenarios. Optionally, in this embodiment, Figure 1 This is a hardware environment diagram of a method for identifying an operation intention according to an embodiment of the present application. The above method can be applied to Figure 1 In the hardware environment shown in FIG. 1 , which is composed of a terminal device 102 and a server 104. Figure 1 As shown, the server 104 is connected to the terminal device 102 via a network, and can be used to provide services (such as application services, etc.) for the terminal or the client installed on the terminal. A database can be set up on the server or independently of the server to provide data storage services for the server 104. Cloud computing and / or edge computing services can be configured on the server or independently of the server to provide data computing services for the server 104.

[0024] The aforementioned network may include, but is not limited to, at least one of the following: a wired network and a wireless network. The aforementioned wired network may include, but is not limited to, at least one of the following: a wide area network, a metropolitan area network, and a local area network. The aforementioned wireless network may include, but is not limited to, at least one of the following: Wi-Fi (Wireless Fidelity) and Bluetooth. The terminal device 102 may be, but is not limited to, a PC, a mobile phone, a tablet computer, a smart air conditioner, a smart range hood, a smart refrigerator, a smart oven, a smart stove, a smart washing machine, a smart water heater, a smart washing machine, a smart dishwasher, a smart projection device, a smart TV, a smart clothes drying rack, smart curtains, smart audio and video, a smart socket, a smart speaker, a smart fresh air device, smart kitchen and bathroom equipment, smart bathroom equipment, a smart sweeping robot, a smart window cleaning robot, a smart mopping robot, a smart air purifier, a smart steamer, a smart microwave oven, a smart kitchen treasure, a smart purifier, a smart water dispenser, a smart door lock, etc.

[0025] In this embodiment, a method for identifying operation intention is provided, which is applied to the above-mentioned device terminal. Figure 2 is a flow chart of a method for identifying operation intention according to an embodiment of the present application, such as Figure 2 As shown, the process includes the following steps:

[0026] Step S202: identifying a device function and a plurality of target keywords carried in the target speech, wherein the device function is used to indicate the type of operation to be performed by the smart device, and the target keywords are used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled;

[0027] Step S204: Combining the device function and the multiple target keywords to construct a reference operation intention, wherein the reference operation intention includes the operation intention formed by combining the device function and the multiple target keywords;

[0028] Step S206, filter out the operation intentions with a credibility higher than the target threshold from the reference operation intentions as the target operation intentions corresponding to the target voice, wherein the credibility is the credibility level corresponding to each of the reference operation intentions detected based on the device function and the expression information of the multiple target keywords in the target voice.

[0029] Through the above steps, the target speech will carry target keywords for indicating the operation performed by the device to be controlled and the smart device to be controlled, as well as device functions for indicating the operation type of the operation to be performed by the smart device. By recognizing the target speech, the device functions and target keywords in the target speech are obtained. Then, by combining the device functions and multiple target keywords in the target speech, the reference operation intentions obtained by the combination can cover all operation intentions that can be combined with the device functions and multiple target keywords. Then, according to the credibility of each reference operation intention that can be detected based on the information expressed by the device functions and multiple functions in the target speech, the reference operation intentions are filtered according to the credibility of each reference operation intention, thereby filtering out target operation intentions with a credibility higher than the target threshold, thereby achieving rapid and accurate recognition of the target operation intentions included in the target speech. The above technical solution solves the problem of low recognition efficiency of operation intentions carried in speech in the related art, and achieves the technical effect of improving the recognition efficiency of operation intentions carried in speech.

[0030] In the technical solution provided in the above step S202, the recognition of the device function and multiple target keywords carried in the target speech can be obtained by using a target recognition model to recognize the target speech. The target recognition model can be obtained by training an initial recognition model using training samples labeled with device functions and keywords. Figure 3 is an optional target speech recognition schematic diagram according to an embodiment of the present application, such as Figure 3 As shown, it at least includes the following steps:

[0031] S301 streamlines the functions of household appliances. For example, there are three major categories: air conditioners, refrigerators, washing machines, and water heaters. Air conditioners have six functions: power on, power off, set temperature, increase temperature, decrease temperature, and set mode. Refrigerators have four functions: power on, power off, set compartment temperature, and food management. Washing machines have five functions: power on, power off, start washing, end washing, and set mode. Water heaters have five functions: power on, power off, increase water temperature, decrease water temperature, and set mode. Dishwashers have five functions: power on, power off, start washing dishes, end washing dishes, and set mode. , then the functions of the obtained devices will be merged to obtain 14 device functions, including power on, power off, setting temperature, raising temperature, lowering temperature, setting mode, setting cabin temperature, food management, starting laundry, ending laundry, raising water temperature, lowering water temperature, starting dishwashing, and ending dishwashing. Expanding to more devices will inevitably cause an explosive increase in device functions, so the device functions need to be streamlined. For example, starting dishwashing and starting laundry can be merged into starting; ending laundry and ending dishwashing can be merged into ending; setting temperature, setting cabin temperature, and setting water temperature can be merged into setting temperature, and cabin and water temperature become slots. After the merger, there are only power on, power off, start, end, raising temperature, and lowering temperature. Among them, food management is a mode unique to refrigerators, so it is removed;

[0032] S302: Obtain streamlined device functions, label training samples with the streamlined device functions, and label target keywords in the training samples, thereby obtaining labeled training samples. The labeled training samples are used to train an initial recognition model to obtain a target recognition model. The initial recognition model may include, but is not limited to, two network models: an initial device function recognition network and an initial keyword recognition network. The target recognition model is obtained by training these two network models using the training samples.

[0033] S303, obtaining a target voice, where the target voice is used to instruct one or more smart devices to perform one or more operations;

[0034] S304, inputting the target speech into the trained target recognition model to obtain the device functions and target keywords included in the target speech;

[0035] S305, sorting and storing the device functions and target keywords according to the function order of the device functions output by the target recognition model and the keyword order of the target keywords output;

[0036] S306 , combining the device function and multiple target keywords to construct a reference operation intention.

[0037] Figure 4 is a schematic diagram of an optional initial recognition model according to an embodiment of the present application, such as Figure 4As shown, the initial recognition model includes the device input layer, feature extraction layer, device function device network, keyword recognition network, and loss function layer in sequence;

[0038] The input layer is used to encode the text in the target speech and positionally encode the positional relationship of the text in the target speech. Since each word is encoded using one-hot encoding, which is a discrete encoding, sparse vectors are very unfavorable for deep model learning and also waste a lot of storage space. Therefore, word embedding is needed to map the sparse one-hot encoding to a dense vector. Since the model uses a bidirectional autoencoding structure, all the text in a sentence is input to the model at the same time. The model needs to know the position information between words, so PositionEmbedding is added. Its formula is as follows:

[0039] PE t，2i =sin(t / 10000 2i / d ), PE t，2i+1 =cos(t / 10000 2i / d )

[0040] Where t is the absolute position of the word, d represents the vector dimension of each word, and i represents the index of the dimension. We add the word embedding and the position embedding as the input of the model;

[0041] The feature extraction layer is used to extract features between characters or keywords in the target speech, which can be but not limited to a transform model;

[0042] The device function recognition network is used to identify the device functions involved in the target speech based on the output of the feature extraction layer. It can be, but is not limited to, a GLS (generalized linear regression) network;

[0043] The keyword recognition network is used to identify the keywords in the target speech according to the results output by the feature extraction layer. During the recognition process, the target keywords used to indicate the smart devices to be controlled and the operations performed by the smart devices to be controlled are determined by identifying the operation information of each keyword slot in the speech. For example, the target speech "turn on the air conditioner and dishwasher" contains a total of 8 characters. The keyword recognition network outputs 8 identifiers "B-ACTION, I-ACTION, B-DEVICE, I-DEVICE, O, B-DEVICE, I-DEVICE, I-DEVICE". Each identifier corresponds to a keyword slot in the target speech, where O represents its It is not a slot. B- represents the beginning of a slot, I- represents the content of the slot. B to the last I or a single B is a slot. B-, I- are followed by the name of the slot (DEVICE (device) and ACTION (action) to indicate the meaning of the keyword). The keyword recognition network can be but is not limited to a CRF (conditional random field) network. In practical applications, we will encounter a large number of OOV (out-of-vocab) problems. For example, "5:1" in "Team A defeated Team B 5:1" is an unregistered word. To solve this problem, we introduced [MASK] to randomly replace some words. In the final model output, we extract the features corresponding to [MASK].

[0044] The Loss function layer is used to calculate the loss function based on the model output results, so as to determine whether the model has converged. During training, for each training sample, we added an identifier [CLS] to represent the sentence. We brought the model output features corresponding to [CLS] into a layer of FeedforWard (feedforward neural network), and then connected multiple (number of categories) SigMoid layers, and brought the final output into the Cross Entropy Loss function (cross loss function) to obtain Loss1. For each training sample, the keyword slot is marked. The loss function Loss2 is calculated based on the output results of the keyword recognition network and the labeled content. The loss function Loss3 is calculated based on the MASK mask result. The loss obtained by the Mask mask is a value between 0 and 1, and CRF is the accumulation of sequence losses. The gap between the two is too large, so some hyperparameters are needed to make adjustments. Therefore, the loss function of this model is Loss total =αLoss1+βLoss2+γLoss3, where αβγ are hyperparameters that need to be adjusted. Different adjustments are required for different tasks. The sum of αβγ is required to be 1.

[0045] Figure 5is a schematic diagram of an optional feature extraction layer structure according to an embodiment of the present application, such as Figure 5 As shown, we use the Self-Attention mechanism to define three matrices K, Q, and V respectively. We multiply the input vector by the three vectors to obtain three corresponding eigenvectors. The three eigenvectors are then put into the formula This yields the required attention weights. We group these three matrices together. Because features can have multiple cases, we can define multiple groups. In this experiment, we defined 12 groups: 12 K, 12 Q, and 12 V, to better extract features. This step is repeated once, which we call a layer. This model has four layers. Between each layer, we also incorporate a residual network to compensate for information loss. This adds the input of the previous layer to the output of the current layer. Before inputting the model, we need to normalize the batch of data to a uniform length. Shorter lengths are padded with zeros, and longer lengths are deleted. This introduces errors in the BatchNormalize stage. We introduce LayerNormalize. Unlike BatchNormalize, LayerNormalize focuses on each data point, rather than a batch of data points. This can significantly reduce the data errors caused by the need to standardize sentence lengths.

[0046] Optionally, in this embodiment, the target voice is a voice carrying an operation intention for instructing a smart device to perform an operation. The target voice may, but is not limited to, carry one or more operation intentions, each operation intention is used to instruct a smart device to perform an operation. When the target voice carries multiple operation intentions, the multiple operation intentions can be used to instruct multiple smart devices to perform different operations respectively, or can also be used to instruct multiple smart devices to perform the same operation, or can also be used to instruct one smart device to perform multiple operations. This solution does not limit this.

[0047] Optionally, in this embodiment, the target keywords include target keywords belonging to the device type and target keywords belonging to the operation type. For example, for the target voice "Help me set the water temperature of the water heater to 26 degrees and the air conditioner to the post-bath mode", the target keywords of the device type identified therefrom are "water heater" and "air conditioner", and the target keywords of the operation type identified therefrom include "water temperature" and the specific numerical value of the water temperature "26", as well as "post-bath model".

[0048] In the technical solution provided in the above step S204, the operation intention is used to instruct the smart device to perform an operation. Therefore, each operation intention needs to include at least the device information of the smart device corresponding to the operation intention (that is, the target keyword belonging to the device type) and the device function that needs to be triggered. For example, for the voice of "Help me turn on the air conditioner", the device function that needs to be triggered is the start function, and the device that executes the triggering function is the air conditioner. Therefore, the operation intention corresponding to the voice needs to include the start function that needs to be triggered and the keyword corresponding to the device that needs to trigger the start function (that is, air conditioner). For example, for the voice of "Set the air conditioner temperature to 26 degrees", the device function that needs to be triggered is the temperature setting function, and the device that triggers the function is the air conditioner. Therefore, the operation intention corresponding to the voice needs to include the temperature setting function that needs to be triggered and the keyword corresponding to the device that needs to trigger the function (that is, air conditioner), as well as the keyword corresponding to the numerical value for executing the temperature setting function, that is, "26".

[0049] Optionally, in this embodiment, the method of combining device functions and target keywords can be a random combination method (that is, combining any one or more target keywords with each device function), or each target keyword can be combined with each device function in turn to obtain a first operation intention, and the operation intentions that are not completed in the bid in the combined first operation intention are merged.

[0050] In the technical solution provided in the above-mentioned step S206, the expression information of the device function and multiple target keywords in the target voice can be the semantic relationship between the device function and the target keyword, or it can also be the keyword distance between the target keywords. When determining the credibility of the reference operation intention, it can be judged based on the semantic relationship between the device function and the target keyword, and / or the keyword distance between the target keywords. For example, when the device function in the operation intention and the target keyword belonging to the operation type match consistently, the credibility of the operation intention is higher than the target threshold, and / or, when the distance between the target keyword belonging to the device type in the reference operation intention and the target keyword belonging to the operation type corresponding to the device function is less than the target distance, it is determined that the credibility of the reference operation intention is higher than the target credibility.

[0051] As an optional embodiment, the combining the device function and the multiple target keywords to construct a reference operation intention includes:

[0052] Determining a function order and a keyword order, wherein the function order is used to indicate the order of each of the device functions in the target speech, and the keyword order is used to indicate the order of each of the target keywords in the target speech;

[0053] The device function and the plurality of target keywords are combined according to the function sequence and the keyword sequence to obtain the reference operation intention.

[0054] Optionally, in this embodiment, the device functions and target keywords are combined according to the function order and the keyword order by extracting the device functions in sequence according to the function order, and for each device function, traversing multiple target keywords in sequence according to the keyword order to combine the operation intentions. During the combination process, when the operation intention is fully expressed, an operation intention is obtained; when the combined operation intention is incomplete, the incomplete operation intention is suspended, and the next target keyword is traversed based on the suspended operation intention until a complete reference operation intention is obtained. In this process, when the device function traverses to a target keyword belonging to an operation type, and the operation corresponding to the target keyword conflicts with the device function, the operation intention is ignored, and the reference operation intention is finally traversed. For example, for the target speech "Help me turn on the TV and turn off the speaker", the device functions involved include "device on" and "device off", and the identified target keywords are "TV", "power on", "speaker", and "shut down". Therefore, the device function "device on" is first used to traverse multiple target keywords in sequence, combine the operation intentions, first traverse to the target keyword "TV", and then traverse to the target keyword "power on", and then obtain the reference operation intention for instructing the TV to trigger the device to turn on this device function. Then, the device function "device on" is used to continue traversing the remaining target keywords, that is, traversing to the keyword "speaker", and then traversing to the keyword "shut down", it is found that the target keyword "shut down" conflicts with the device function "device on", and then this operation intention is ignored. Similarly, the device function "device off" is used to traverse the target keywords again in keyword order, thereby obtaining the reference operation intention for instructing the speaker to trigger the device to turn off this device function.

[0055] Optionally, in this embodiment, the method of combining device functions and target keywords according to the function order and keyword order can also be to extract target keywords from multiple target keywords in sequence according to the keyword order, and combine them with the device functions arranged in the function order in sequence to obtain a first operation intention, determine the complete operation intention in the first operation intention as a reference operation intention, and merge the incomplete operation intentions in all the combined first operation intentions to obtain a complete operation intention as a reference operation intention.

[0056] As an optional embodiment, combining the device function and the plurality of target keywords according to the function order and the keyword order to obtain the reference operation intention includes:

[0057] Combining each of the target keywords included in the plurality of target keywords with the device function in sequence according to the keyword order to obtain a plurality of first operation intentions, wherein each of the target keywords is combined with the device function in the function order;

[0058] determining a complete operation intention among the plurality of first operation intentions as the reference operation intention;

[0059] The incomplete operation intentions among the multiple first operation intentions are merged to obtain a complete operation intention as the reference operation intention.

[0060] Optionally, in this embodiment, the operation intention is used to instruct the smart device to perform an operation, so the operation intention needs to include at least the smart device information (target keyword belonging to the device type) and information related to the device function of the smart device triggered (target keyword belonging to the operation type). Therefore, a complete expression of the operation intention is to include the smart device information and information related to the device function of the smart device triggered in the operation intention. For example, for the operation intention of setting the temperature of the air conditioner to 26 degrees, it can be (air conditioner, temperature setting function, 26 degrees), that is, the operation intention includes the smart device information "air conditioner", and information related to the device function of the smart device triggered "temperature setting function" and the parameter value corresponding to the function "26 degrees".

[0061] Figure 6 is a schematic diagram of constructing an optional reference operation intention according to an embodiment of the present application, such as Figure 6As shown, taking the target speech as "Raise the water temperature of the water heater by 10 degrees and lower the air conditioner temperature by 11 degrees" as an example, the device functions included in this sentence are increaseTemperatur to increase the temperature and decreaseTemperature to lower the temperature. The target keywords included in the target speech are "water heater", "raise", "10", "air conditioner", "lower", and "11". The first operation intention is constructed by traversing the target keywords and device functions in the order of keywords, and the reference operation intention is obtained by merging the complete operation intention expressed in the first operation intention. That is, the first operation intention is constructed for the first keyword "water heater" and increaseTemperatur. Since the keyword "water heater" is not an operation keyword, it is inconsistent with the action corresponding to the device function, and the intention score is 0, that is, the first operation intention is (water heater, increaseTemperatur, 0 (intent score), []). In the first operation intention, the first position is the target keyword corresponding to the device name, the second position is the device function, the third position is the intention score, and the fourth position is the specific value corresponding to the device function, here is Vacant, since the operation information in the first operation intention is vacant, it is temporarily suspended, and then according to the same principle, the first keyword "water heater" and decreaseTemperature are used to construct the first operation intention, and (water heater, decreaseTemperature, 0 (intent score), []) is obtained; then the second target keyword "raise", since the action performed by "raise" is consistent with the action of increaseTemperatur, the intention score of the first operation intention obtained is 1, so the first operation intention constructed is ([], increaseTemperatur, 1 (intent score), []), since the action performed by "raise" is consistent with the action of decreaseTemperatur, the first operation intention constructed by "raise" and decreaseTemperatur is ignored; then, when traversing the third keyword "10", according to the above principle, ([], increaseTemperature, 0 (intent score), 10) and ([], decreaseTemperature, 0 (intent score), 10) can be constructed;Similarly, the remaining target keywords are traversed in the same manner to obtain all first operation intents. Incomplete first operation intents are then merged in the order in which they were constructed. The two first operation intents are not mutually exclusive when merged. Thus, the reference operation intents for "increase the water heater temperature by 10 degrees and decrease the air conditioner temperature by 11 degrees" are (water heater, increaseTemperature, 1 (intent score), 10), (water heater, decreaseTemperature, 0 (intent score), 10), (air conditioner, increaseTemperature, 0 (intent score), 11), and (air conditioner, decreaseTemperature, 1 (intent score), 11).

[0062] As an optional embodiment, the step of merging incomplete operation intentions from a plurality of the first operation intentions to obtain a complete operation intention as the reference operation intention includes:

[0063] Merging the second operation intentions in the incomplete operation intentions in sequence according to the target order to obtain a third operation intention, wherein the second operation intentions are at least two operation intentions whose operation information is non-mutually exclusive, the operation information is information instructing the smart device to perform an operation, and the target order is the order in which the first operation intentions are obtained;

[0064] When the third operation intention is completely expressed, the third operation intention is determined as the reference operation intention.

[0065] Optionally, in this embodiment, in the process of merging the second operation intention in the incomplete operation intention according to the target order, when the operation intention obtained after the two operation intentions are merged is complete, a reference operation intention is obtained; when the operation intention obtained after the operation intention is merged is incomplete, it continues to be merged with the subsequent second operation intention according to the target order, wherein, in the entire process of merging the second operation intention, the second operation intention that has participated in the merging of the reference operation intention will not be reused.

[0066] As an optional embodiment, after merging the second operation intentions in the incomplete operation intentions in the target order to obtain the third operation intention, the method further includes:

[0067] Determining target operation information missing from the incomplete third operation intention;

[0068] generating a confirmation message, wherein the confirmation message is used to confirm the target operation information;

[0069] In response to the target voice, play the confirmation message;

[0070] The third operation intention that is fully expressed is constructed according to the reply message of the received confirmation message as the reference operation intention.

[0071] Optionally, in this embodiment, the operation information is the device class information used to indicate the device to be controlled in the operation intention and the operation class information used to indicate the operation performed by the smart device to be controlled, wherein the device class information may be but is not limited to the device identification of the smart device, the device name of the smart device, or it may also be a target keyword belonging to the device type included in multiple target keywords, and the operation class information may not be limited to including the device function and the specific numerical value of the device function triggered by the smart device (such as the temperature value adjusted when the temperature adjustment function is triggered).

[0072] Optionally, in this embodiment, the operation information in the operation intention is arranged in a certain order, that is, different operation information is arranged in different positions in the operation intention. When the operation intention is referred to according to the device function and multiple target keyword combinations, the corresponding operation information (device function and target keyword) is also arranged in this order. Then, when the operation information is missing at a certain position in the operation intention, the operation information type of the missing target operation information can be determined according to the position of the missing operation information. For example, the operation intention can be in the following form (target keyword belonging to the device type, device function, target keyword belonging to the device function value), that is, in the operation intention, the first position is set with the target keyword belonging to the device type, the device function and the target keyword belonging to the device function value (for example, the device function is to set the temperature, and the target keyword corresponding to the device function value is temperature). degree value, such as 26 degrees), if a third operation intention with missing target operation information is (air conditioning, [], 26 degrees), it can be seen that the operation information in the second position is missing, that is, the target operation information is the operation information used to indicate the function of the device. Therefore, and there is a specific temperature value in the third operation intention, we can determine that the device functions with temperature values ​​include setting temperature, raising temperature, and lowering temperature. Therefore, it generates "Do you want the air conditioning to be raised to 26 degrees, lowered to 26 degrees, or set to 26 degrees?" At this time, the user replies "Set to 26 degrees", so it can be determined that the target operation information replied by the user is the temperature setting function. Therefore, the temperature setting function is added to the second position in the third operation intention to obtain (air conditioning, temperature setting function, 26 degrees). Therefore, it can be known that the operation intention is used to trigger the temperature setting function of the air conditioner and the value of the triggered temperature setting function is 26 degrees.

[0073] As an optional embodiment, the step of selecting an operation intention having a credibility higher than a target threshold from the reference operation intentions as the target operation intention corresponding to the target voice includes:

[0074] If there are no multiple fourth operation intentions in the reference operation intention, determining the reference operation intention as the target operation intention, wherein the multiple fourth operation intentions respectively instruct the same smart device to perform an operation and the indicated operations are different from each other;

[0075] When there are multiple fourth operation intentions in the reference operation intentions, the target operation intention is filtered from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions, wherein the intention score is a score determined based on the matching relationship between the target keyword belonging to the operation type and the device function when each target keyword is combined with each device function respectively, and the credibility includes the intention score; when the target operation intention is not filtered out from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions, a sixth operation intention corresponding to each fourth operation intention is filtered out from the fifth operation intentions included in the reference operation intentions, wherein the fifth operation intention is an operation intention other than the fourth operation intention in the reference operation intention, and the sixth operation intention is an operation intention that performs the same operation as the fourth operation intention; the target operation intention is filtered out from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance, wherein the keyword distance is used to indicate the distance between the target keyword belonging to the operation type and the target keyword belonging to the device type included in each operation intention in the target voice, and the credibility includes the keyword distance.

[0076] Optionally, in this embodiment, the intention score is used to indicate whether the target keyword belonging to the operation type matches the device function. When they match, the intention score is set to 1. When they do not match or the target keyword belonging to the operation type is missing, the intention score is set to 0. For example, for the target voice "Set the water temperature of the water heater to 26 degrees", the device function included in this voice is the temperature setting function, and the target keywords included are "water heater", "temperature setting", and "26 degrees" in sequence. Then, when combining the first operation intention according to the keyword order and the function order, first combine the device function and the target keyword "water heater" to obtain the first operation intention as (water heater, temperature setting function, 0, []). In this first operation intention, they are the target keyword belonging to the device type, the device function, the intention score, and the specific value for executing the device function in sequence. Since the first target keyword "water heater" is not a keyword belonging to the operation type, the intention score in the third position is 0, and the specific value for executing the device function in the fourth position is vacant. Then, combine the device function and the target keyword "temperature setting" again to obtain the next first operation intention as ([], temperature setting function, 1, []). Since the device function and the target keyword "temperature setting" match, the intention score in the third position of this first operation intention is 1, and the first and second positions are vacant.

[0077] Optionally, in the embodiment, the calculation of the keyword distance is unidirectional, that is, it is calculated from front to back or from back to front in the target voice. For example, for the voice "Turn on the water heater and then set the air conditioner to the post-bath mode", D(air conditioner|post-bath) represents the distance from post-bath to the air conditioner, and D is unidirectional and can only go forward. It is found that D(air conditioner|post-bath) < D(water heater|post-bath), so we determine that the post-bath mode belongs to the air conditioner and does not belong to the water heater.

[0078] As an optional embodiment, screening the target operation intention from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions includes:

[0079] When there is only one operation intention with the intention score being the target intention score in the fourth operation intention, determining the fourth operation intention with the target intention score as the target operation intention, where the target intention score is used to indicate that when each target keyword is combined with each device function respectively, the target keyword belonging to the operation type matches the device function;

[0080] When all the operation intentions in the fourth operation intention are operation intentions with the target intention score or there is no operation intention with the target intention score, it is determined that the target operation intention is not screened out from the fourth operation intention.

[0081] Optionally, in this embodiment, in the process of combining the device function and multiple target keywords to form the first operation intention, when the target keyword that matches the device function and belongs to the operation type is consistent with the device function, the score of the operation intention is set to the target intention score, and the score of the reference operation intention obtained by combining the operation intentions is the target operation intention, and then when the smart devices to be controlled indicated in the two operation intentions are the same smart device, and the operations performed are different operations, the reference operation intention with the intention score of the target intention score is used as the target operation intention. For example, the reference operation intentions obtained by the combination are (air conditioning, temperature setting function, 1 (intention score), 26 degrees), (air conditioning, temperature increase function, 0 (intention score), 26 degrees). Since the intention score in the operation intention of (air conditioning, temperature setting function, 1 (intention score), 26 degrees) is the target intention score, it is considered that the credibility of the operation intention is high, and the operation intention is determined as the target operation intention.

[0082] As an optional embodiment, the step of filtering out the target operation intention from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention based on the keyword distance includes:

[0083] In a case where all of the plurality of fourth operation intentions are operation intentions with a target intention score, determining whether the operations indicated by the plurality of fourth operation intentions to be performed by the same smart device are mutually exclusive operations, wherein the target intention score is used to indicate whether the target keyword belonging to the operation type matches the device function when each target keyword is combined with each device function respectively;

[0084] In a case where the operations performed by the same smart device indicated by the multiple fourth operation intentions are non-mutually exclusive operations, determining the fourth operation intention as the target operation intention;

[0085] In the case that the operations executed by the same smart device indicated by multiple fourth operation intentions are mutually exclusive operations, the target operation intention is filtered out from the fourth operation intentions and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance.

[0086] Optionally, in this embodiment, when the intention scores of multiple fourth operation intentions are all target intention scores, and the executed operations are non-mutually exclusive operations, then these operation intentions are considered to be credible operation intentions and can be executed. For example, for the target voice "turn on the air conditioner and set it to 26 degrees", the device functions included therein are the device opening function openDevice and the temperature setting function setTemperature, respectively, and the recognized target keywords are "open", "air conditioner", "set", and "26 degrees". When combining, the first target keyword "open" is first combined with the device opening function openDevice to obtain the first operation intention ([ ], openDevice, 0 ), since the target keyword "open" and the device opening function openDevice match each other, the intention score of the first reference intention should be set to 1, that is, it becomes ([], openDevice, 1), and since the operation information in the operation intention is incomplete, it is suspended, and the combination of the first target keyword "open" and the temperature setting function setTemperature is continued. Since the actions performed by the keyword open and the temperature setting function are different, and based on the principle that an operation intention can only include one operation, the operation intention obtained by combining the first target keyword "open" and the temperature setting function setTemperature is ignored, and the second target keyword is continued to be traversed. "Air conditioning", combine it with the device opening function openDevice to get the first operation intention (air conditioning, openDevice, 0), find that (air conditioning, openDevice, 0) can be merged with ([ ], openDevice, 1), so the reference operation intention (air conditioning, openDevice, 1) is obtained, and then continue to combine the target keyword with the temperature setting function setTemperature to get the first operation intention (air conditioning, setTemperature, 0), continue to traverse the target keyword "setting", find that this target keyword and the device opening function openDevice have different execution actions, ignore its combination operation Intent, and the target keyword "setting" is combined with the temperature setting function setTemperature. Since the device operation matches, the intent score is the target intent score 1, and the first operation intention is ([], setTemperature, 1, []). Then, the keyword "26 degrees" is traversed and it is found that it conflicts with the device opening function openDevice. Then, according to the target keyword "26 degrees" and "temperature setting function setTemperature", the first operation intention is ([], setTemperature, 0, 26 degrees). In addition, due to the first operation intention (air conditioning, setTemperature, 0), ([],setTemperature, 1, []) and ([], setTemperature, 0, 26 degrees) can be merged, so the reference operation intention is (air conditioning, setTemperature, 1, 26 degrees). In summary, the reference operation intentions obtained by combining the device function and the target keyword are (air conditioning, setTemperature, 1, 26 degrees) and (air conditioning, openDevice, 1). Since both of these intention scores are the target intention score of 1, and the operations performed in the two operation intentions are non-conflicting (non-mutually exclusive) operations, these two reference operation intentions are determined as the target operation intentions.

[0087] As an optional embodiment, the step of filtering out the target operation intention from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention based on the keyword distance includes:

[0088] The operation intention with the smallest keyword distance is selected from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention as the target operation intention, wherein the smaller the keyword distance is, the higher the credibility is.

[0089] Optionally, in this embodiment, when calculating the keyword distance, the one-way principle needs to be followed, that is, calculate from front to back or from back to front in the target speech. For example, for the target speech "Open the water heater and then set the air conditioner to the post-bath mode", using the above method to obtain the reference operation intentions: (water heater, openDevice, 0 (intention score is 0)), (air conditioner, openDevice, 0 (intention score is 0)), (water heater, setMode, 0 (intention score is 0), post-bath), (air conditioner, setMode, 0 (intention score is 0), post-bath). It is found that the water heater needs to perform two operations: (water heater, openDevice, 0 (intention score is 0)) and (water heater, setMode, 0 (intention score is 0), post-bath), and the air conditioner also needs to perform two operations: (air conditioner, openDevice, 0 (intention score is 0)) and (air conditioner, setMode, 0 (intention score is 0)), and the intention scores of all operations are 0. Therefore, it is suspected that there is an error. Therefore, the correct result can be screened out by comparing the keyword distances. That is, D(air conditioner|post-bath) represents the distance from post-bath to the air conditioner, and D is one-way and can only go forward. It is found that D(air conditioner|post-bath) < D(water heater|post-bath). So we judge that the post-bath mode belongs to the air conditioner, not the water heater. So delete the multi-tuple (water heater, setMode, 0 (intention score is 0), post-bath). Finally, the target operation intention obtained is (air conditioner, openDevice, 0 (intention score is 0)), (air conditioner, setMode, 0 (intention score is 0)), (water heater, openDevice, 0 (intention score is 0)). Although there is one more (air conditioner, openDevice, 0 (intention score is 0)) than in the target speech, it does not conflict and does not prevent the normal function of the device. So no excessive processing is required.

[0090] Figure 7 It is a flowchart for screening the target operation intention according to an optional embodiment of the present application, as Figure 7 shown:

[0091] S701, in the case where there are multiple reference operation intentions, screen out multiple fourth operation intentions from the reference operation intentions. The multiple fourth operation intentions respectively indicate that the same intelligent device performs an operation and the indicated operations are different from each other;

[0092] S702, determine the intention score of each fourth operation intention;

[0093] S703, when it is determined that the intention scores of multiple fourth operation intentions are all 0, execute step S704;

[0094] S704, calculating the keyword distance between the target keyword corresponding to the device function included in each operation intention and the target keyword belonging to the device type in the target speech, and determining the operation intention with the smallest keyword distance as the target operation intention;

[0095] S705: When it is determined that the intention scores of the plurality of fourth operation intentions are all 1, executing step S706;

[0096] S706, determining whether the operations executed by the device in the plurality of fourth operation intentions conflict, for example, increaseTemperatur and decreaseTemperatur are conflicting operations. If the executed operations conflict, executing step S704; if the executed operations do not conflict, executing step S709;

[0097] S707, when it is determined that the intention scores of the plurality of fourth operation intentions are 0 and 1, executing step S708;

[0098] S708: Filter out the operation intention with an intention score of 1 from the fourth operation intention as the target operation intention.

[0099] S709: Obtain the target operation intention included in the target speech.

[0100] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0101] Figure 8 is a structural block diagram of an operation intention recognition device according to an embodiment of the present application; Figure 8 As shown, it includes: a recognition module 82, which is used to recognize the device function and multiple target keywords carried in the target voice, wherein the device function is used to indicate the operation type of the operation to be performed by the smart device, and the target keyword is used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled;

[0102] a combining module 84, configured to combine the device function and the plurality of target keywords to construct a reference operation intention, wherein the reference operation intention includes the operation intention formed by combining the device function and the plurality of target keywords;

[0103] The screening module 86 is used to screen out the operation intentions with a credibility higher than the target threshold from the reference operation intentions as the target operation intentions corresponding to the target voice, wherein the credibility is the credibility level corresponding to each of the reference operation intentions detected based on the device function and the expression information of the multiple target keywords in the target voice.

[0104] Through the above embodiment, the target speech carries target keywords for indicating the operation to be performed by the device to be controlled and the smart device to be controlled, as well as device functions for indicating the operation type of the operation to be performed by the smart device. By recognizing the target speech, the device functions and target keywords in the target speech are obtained. Then, by combining the device functions and multiple target keywords in the target speech, the reference operation intentions obtained by the combination can cover all operation intentions that can be combined with the device functions and multiple target keywords. Then, according to the credibility of each reference operation intention that can be detected based on the information expressed by the device functions and multiple functions in the target speech, the reference operation intentions are filtered according to the credibility of each reference operation intention, thereby filtering out target operation intentions with a credibility higher than the target threshold, thereby achieving rapid and accurate recognition of the target operation intentions included in the target speech. The above technical solution solves the problem of low recognition efficiency of operation intentions carried in speech in the related art, and achieves the technical effect of improving the recognition efficiency of operation intentions carried in speech.

[0105] Optionally, the combination module includes: a first determination unit, used to determine the function order and the keyword order, wherein the function order is used to indicate the order of each of the device functions in the target voice, and the keyword order is used to indicate the order of each of the target keywords in the target voice; a combination unit, used to combine the device functions and multiple target keywords according to the function order and the keyword order to obtain the reference operation intention.

[0106] Optionally, the combination unit is used to: combine each of the target keywords included in the multiple target keywords with the device function in sequence according to the keyword order to obtain multiple first operation intentions, wherein each target keyword is combined with the device function according to the function order; determine the complete operation intention among the multiple first operation intentions as the reference operation intention; merge the incomplete operation intentions among the multiple first operation intentions to form a complete operation intention as the reference operation intention.

[0107] Optionally, the combination unit is used to: merge the second operation intentions in the incomplete operation intentions in sequence according to the target order to obtain a third operation intention, wherein the second operation intentions are at least two operation intentions whose operation information is non-mutually exclusive, the operation information is information instructing the smart device to perform an operation, and the target order is the order in which the first operation intentions are obtained; when the third operation intention is completely expressed, the third operation intention is determined as the reference operation intention.

[0108] Optionally, the device also includes: a determination module, used to merge the second operation intentions in the incomplete operation intentions in the target order in sequence to obtain the third operation intention, and then determine the target operation information missing from the incomplete third operation intention; a generation module, used to generate a confirmation message, wherein the confirmation message is used to confirm the target operation information; a playback module, used to respond to the target voice and play the confirmation message; a construction module, used to construct the complete third operation intention as the reference operation intention based on the reply message received to the confirmation message.

[0109] Optionally, the screening module includes: a second determination unit for determining the reference operation intention as the target operation intention when there are no multiple fourth operation intentions in the reference operation intention, wherein the multiple fourth operation intentions respectively indicate that the same smart device performs an operation and the indicated operations are different from each other; a processing unit for screening the target operation intention from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions when there are multiple fourth operation intentions in the reference operation intention, wherein the intention score is a score determined according to the matching relationship between the target keyword belonging to the operation type and the device function when each target keyword is respectively combined with each device function, and the credibility includes the intention score; When the intention scores of the four operation intentions fail to filter out the target operation intention from the multiple fourth operation intentions, a sixth operation intention corresponding to each of the fourth operation intentions is filtered out from the fifth operation intention included in the reference operation intention, wherein the fifth operation intention is an operation intention other than the fourth operation intention in the reference operation intention, and the sixth operation intention is an operation intention that performs the same operation as the fourth operation intention; the target operation intention is filtered out from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention based on the keyword distance, wherein the keyword distance is used to indicate the distance between the target keyword belonging to the operation type and the target keyword belonging to the device type included in each operation intention in the target voice, and the credibility includes the keyword distance.

[0110] Optionally, the processing unit is used to: when the fourth operation intention includes only one operation intention with the intention score being the target intention score, determine the fourth operation intention with the target intention score as the target operation intention, wherein the target intention score is used to indicate that when each target keyword is combined with each device function respectively, the target keyword belonging to the operation type matches the device function; when all the fourth operation intentions are operation intentions with the target intention score, or do not include an operation intention with the target intention score, determine that the target operation intention has not been filtered out from the fourth operation intention.

[0111] Optionally, the processing unit is used to: in a case where all of the multiple fourth operation intentions are operation intentions with target intention scores, determine whether the operations performed by the same smart device indicated by the multiple fourth operation intentions are mutually exclusive operations, wherein the target intention score is used to indicate that when each of the target keywords is combined with each of the device functions respectively, the target keyword belonging to the operation type matches the device function; in a case where the operations performed by the same smart device indicated by the multiple fourth operation intentions are non-mutually exclusive operations, determine the fourth operation intention as the target operation intention; in a case where the operations performed by the same smart device indicated by the multiple fourth operation intentions are mutually exclusive operations, filter out the target operation intention from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance.

[0112] Optionally, the processing unit is used to: filter out the operation intention with the smallest keyword distance from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention as the target operation intention, wherein the smaller the keyword distance is, the higher the credibility is.

[0113] An embodiment of the present application further provides a storage medium, which includes a stored program, wherein when the program is run, the method for identifying any of the above-mentioned operation intentions is executed.

[0114] An embodiment of the present application further provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned operation intention recognition method embodiments.

[0115] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0116] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0117] Optionally, specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0118] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Alternatively, they can be implemented using program code executable by the computing device, so that they can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be made into separate integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0119] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying operation intention, characterized in that: include: Identify a device function and multiple target keywords carried in the target voice, wherein the device function is used to indicate the type of operation to be performed by the smart device, and the target keywords are used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled; Combining the device function and the multiple target keywords to construct a reference operation intention, wherein the reference operation intention includes the operation intention formed by combining the device function and the multiple target keywords; Filtering out the operation intentions with a credibility higher than a target threshold from the reference operation intentions as the target operation intention corresponding to the target speech, wherein the credibility is the credibility of each reference operation intention detected based on the device function and the expression information of the multiple target keywords in the target speech; Among them, the combining of the device functions and the multiple target keywords to construct a reference operation intention includes: determining a function order and a keyword order, wherein the function order is used to indicate the order of each of the device functions in the target voice, and the keyword order is used to indicate the order of each of the target keywords in the target voice; combining the device functions and the multiple target keywords according to the function order and the keyword order to obtain the reference operation intention.

2. The method according to claim 1, characterized in that The combining the device function and the plurality of target keywords according to the function sequence and the keyword sequence to obtain the reference operation intention includes: Combining each of the target keywords included in the plurality of target keywords with the device function in sequence according to the keyword order to obtain a plurality of first operation intentions, wherein each of the target keywords is combined with the device function in the function order; determining a complete operation intention among the plurality of first operation intentions as the reference operation intention; The incomplete operation intentions among the multiple first operation intentions are merged to obtain a complete operation intention as the reference operation intention.

3. The method according to claim 2, characterized in that The step of merging the incomplete operation intentions from the plurality of the first operation intentions to obtain a complete operation intention as the reference operation intention includes: Merging the second operation intentions in the incomplete operation intentions in sequence according to the target order to obtain a third operation intention, wherein the second operation intentions are at least two operation intentions whose operation information is non-mutually exclusive, the operation information is information instructing the smart device to perform an operation, and the target order is the order in which the first operation intentions are obtained; When the third operation intention is completely expressed, the third operation intention is determined as the reference operation intention.

4. The method according to claim 3, characterized in that After merging the second operation intentions in the incomplete operation intentions in the target order to obtain the third operation intention, the method further includes: Determining target operation information missing from the incomplete third operation intention; generating a confirmation message, wherein the confirmation message is used to confirm the target operation information; In response to the target voice, play the confirmation message; The third operation intention that is fully expressed is constructed according to the reply message of the received confirmation message as the reference operation intention.

5. The method according to claim 1, wherein The step of selecting an operation intention with a credibility higher than a target threshold from the reference operation intentions as the target operation intention corresponding to the target voice includes: If there are no multiple fourth operation intentions in the reference operation intention, determining the reference operation intention as the target operation intention, wherein the multiple fourth operation intentions respectively instruct the same smart device to perform an operation and the indicated operations are different from each other; When there are multiple fourth operation intentions in the reference operation intentions, the target operation intention is filtered from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions, wherein the intention score is a score determined based on the matching relationship between the target keyword belonging to the operation type and the device function when each target keyword is combined with each device function respectively, and the credibility includes the intention score; when the target operation intention is not filtered out from the multiple fourth operation intentions according to the intention scores of the multiple fourth operation intentions, a sixth operation intention corresponding to each fourth operation intention is filtered out from the fifth operation intentions included in the reference operation intentions, wherein the fifth operation intention is an operation intention other than the fourth operation intention in the reference operation intention, and the sixth operation intention is an operation intention that performs the same operation as the fourth operation intention; the target operation intention is filtered out from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance, wherein the keyword distance is used to indicate the distance between the target keyword belonging to the operation type and the target keyword belonging to the device type included in each operation intention in the target voice, and the credibility includes the keyword distance.

6. The method according to claim 5, characterized in that The step of screening the target operation intention from the plurality of fourth operation intentions according to the intention scores of the plurality of fourth operation intentions includes: In a case where the fourth operation intentions include only one operation intention with the intention score being a target intention score, determining the fourth operation intention with the target intention score as the target operation intention, wherein the target intention score is used to indicate that when each target keyword is combined with each device function, the target keyword belonging to the operation type matches the device function; In a case where all of the fourth operation intentions are operation intentions with the target intention score, or do not include any operation intention with the target intention score, it is determined that the target operation intention has not been filtered out from the fourth operation intentions.

7. The method according to claim 5, characterized in that The step of selecting the target operation intention from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance includes: In a case where all of the plurality of fourth operation intentions are operation intentions with a target intention score, determining whether the operations indicated by the plurality of fourth operation intentions to be performed by the same smart device are mutually exclusive operations, wherein the target intention score is used to indicate whether the target keyword belonging to the operation type matches the device function when each target keyword is combined with each device function respectively; In a case where the operations performed by the same smart device indicated by the multiple fourth operation intentions are non-mutually exclusive operations, determining the fourth operation intention as the target operation intention; In the case that the operations executed by the same smart device indicated by multiple fourth operation intentions are mutually exclusive operations, the target operation intention is filtered out from the fourth operation intentions and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance.

8. The method according to claim 5, characterized in that The step of selecting the target operation intention from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention according to the keyword distance includes: The operation intention with the smallest keyword distance is selected from the fourth operation intention and the sixth operation intention corresponding to the fourth operation intention as the target operation intention, wherein the smaller the keyword distance is, the higher the credibility is.

9. A device for identifying operation intention, characterized in that: include: a recognition module, configured to recognize a device function and a plurality of target keywords carried in the target speech, wherein the device function is used to indicate the type of operation to be performed by the smart device, and the target keywords are used to indicate the smart device to be controlled and the operation performed by the smart device to be controlled; a combining module, configured to combine the device function and the plurality of target keywords to construct a reference operation intention, wherein the reference operation intention includes an operation intention formed by combining the device function and the plurality of target keywords; a screening module, configured to screen out, from the reference operation intentions, operation intentions with a credibility higher than a target threshold as target operation intentions corresponding to the target speech, wherein the credibility is a degree of credibility corresponding to each reference operation intention detected based on the device function and the expression information of the multiple target keywords in the target speech; Among them, the combination module includes: a first determination unit, used to determine the function order and the keyword order, wherein the function order is used to indicate the order of each of the device functions in the target voice, and the keyword order is used to indicate the order of each of the target keywords in the target voice; a combination unit, used to combine the device function and multiple target keywords according to the function order and the keyword order to obtain the reference operation intention.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method according to any one of claims 1 to 8 when executed.

11. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 8 through the computer program.

Images