Air conditioner control method and device based on continuous voice recognition and computer storage medium

Abstract

The invention discloses an air conditioner control method and device for continuous voice recognition and a computer storage medium, and relates to the technical field of air conditioner control, the air conditioner control method comprises the steps that a user voice data set uploaded by a voice recognition unit is obtained, and the user voice data set comprises voice command data for a target air conditioner; inputting the data into a pre-trained user voice recognition model to recognize command words of the user; according to the recognized command word and the current state of the air conditioner, whether the corresponding air conditioner use condition needs to be triggered or not is judged; if not, a trigger command is sent to the air conditioner to enable the air conditioner to execute a program corresponding to the user command word, and reply voice is broadcasted through a loudspeaker. According to the method, the convenience and the intelligent level of air conditioner control are improved, a user can interact with the air conditioner through continuous voice recognition, and more natural and efficient control experience is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning control, and in particular to an air conditioning control method and device for continuous speech recognition and a computer storage medium. Background Art

[0002] With the rapid development of science and technology, smart home has become an important part of modern life. Among them, air conditioners, as key equipment in home appliances, have attracted particular attention for their intelligent control. Traditional air conditioner control methods mostly rely on remote controls or manual operations. Although this method is simple and direct, it has certain limitations in user experience. For example, users need to find the remote control, and the operation interface may not be intuitive enough. In recent years, with the continuous advancement of voice recognition technology, voice control of air conditioners has gradually become a new trend.

[0003] However, the existing voice-controlled air-conditioning technology is mostly based on simple voice command recognition. Users need to say specific command words before the air-conditioning can respond accordingly. Although this method has improved the user experience to a certain extent, there are still some problems. For example, the user's voice command may be affected by various factors such as environmental noise, speaking speed, accent, etc., resulting in low recognition accuracy. In addition, the current voice air-conditioning equipment mainly controls a single voice command when performing voice control. When performing continuous voice recognition, it is also necessary to reply to each command separately before the next voice can be recognized. If multiple voice commands are entered continuously, there are two possibilities. One is that only one command can be recognized and other commands cannot be recognized; the other is that it can be processed continuously, but the current command reply will interrupt the user and cannot make intelligent adjustments based on the user's habits or the current environment.

[0004] Therefore, an air conditioning control method with continuous speech recognition is needed, which can accurately recognize the user's continuous voice commands, understand the user's intentions, and make reasonable control decisions based on actual conditions, providing users with a more comfortable and convenient use experience. Summary of the invention

[0005] The purpose of the present invention is to provide an air conditioning control method, device and computer storage medium for continuous voice recognition, which are used to solve the problem that the user's continuous voice commands cannot be accurately recognized in the prior art and it is difficult to intelligently and reasonably control the air conditioner.

[0006] The present invention solves the above problems through the following technical solutions:

[0007] A method for controlling an air conditioner by continuous speech recognition is applied to a speech control unit, wherein the speech control unit is communicatively connected with a speech recognition unit, a speaker, and a target air conditioner, and the method comprises:

[0008] Acquire a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner;

[0009] Inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model;

[0010] According to the user's command word and the current usage status of the target air conditioner, determining whether the target air conditioner triggers the usage condition corresponding to the user's command word;

[0011] When the target air conditioner does not trigger the usage condition corresponding to the user's command word, a trigger command is sent to the target air conditioner, and the trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

[0012] Furthermore, the training process of the user speech recognition model includes:

[0013] Acquire multiple training sample sets, wherein the training sample sets are user voice data sample sets, and the user voice data sample sets include sample voice data of sample users for sample air conditioners, wherein each of the training sample sets carries a corresponding command word label;

[0014] Calling an initial user speech recognition model to divide the multiple training sample sets into multiple training batches;

[0015] Inputting the plurality of training batches into the initial user speech recognition model in sequence to obtain a predicted command word corresponding to each training sample set;

[0016] Based on the predicted command words and the command word labels corresponding to each training sample set, determining the loss value of the initial user speech recognition model according to a preset loss function;

[0017] Adjusting the model parameters of the initial user speech recognition model according to the loss value to obtain an adjusted initial user speech recognition model;

[0018] Return to the step of inputting the multiple training batches into the initial user speech recognition model in sequence, and iteratively execute the step of obtaining the predicted command word corresponding to each training sample set until the initial user speech recognition model reaches a preset training stop condition, thereby obtaining the trained user speech recognition model.

[0019] Furthermore, the initial user speech recognition model includes an input module, a speech feature extraction and processing module, a time prediction model, a pattern prediction model, and a numerical prediction model; the multiple training batches are sequentially input into the initial user speech recognition model to obtain the predicted command words corresponding to each training sample set, including:

[0020] Inputting the plurality of training batches into the speech feature extraction and processing module in sequence through the input module;

[0021] Based on the speech feature extraction and processing module, feature extraction and processing are performed to obtain time series features, usage mode features and temperature setting features;

[0022] Inputting the time series features into the time prediction model module to obtain the prediction result of the tendency setting time of the air conditioner;

[0023] Inputting the time series characteristics and the usage pattern characteristics into the pattern prediction model to obtain a prediction result of the tendency usage pattern of the air conditioner;

[0024] Inputting the time series feature, the usage pattern feature and the temperature setting feature into the numerical prediction model to obtain a prediction result of the tendency setting data of the air conditioner;

[0025] The tendency setting time prediction result, the tendency use mode prediction result and the tendency setting data prediction result constitute the prediction command word.

[0026] Furthermore, the speech feature extraction and processing module includes:

[0027] A user preprocessing submodule, used for preprocessing the training sample sets in the input multiple training batches;

[0028] A time series feature extraction submodule, used to extract time series features of the time series data in the training sample set;

[0029] A usage pattern feature extraction submodule, used to extract usage pattern features of usage pattern data in the training sample set;

[0030] The temperature setting feature extraction submodule is used to extract the water boiling temperature features of the water boiling temperature data in the training sample set.

[0031] The conflict handling submodule is used to determine whether there is a mode conflict or a temperature conflict in the extracted features, so as to select the corresponding mode or temperature according to the preset rules;

[0032] The speech rate waiting feedback processing submodule is used to set the waiting time according to the user's speech rate and system performance in order to process and analyze the extracted features.

[0033] Furthermore, the time prediction model includes a time series feature extraction module and a time preference prediction module, the pattern prediction model includes a pattern classifier, and the numerical prediction model includes a regression analysis module; the time series feature is input into the time prediction model module to obtain the prediction result of the tendency setting time of the air conditioner, including:

[0034] Inputting the time series features into the time series feature extraction module to obtain periodic features, trend features and seasonal features, and using the time preference prediction module to predict the preferred setting time of the air conditioner based on the periodic features, trend features and seasonal features;

[0035] The step of inputting the time series features and the usage pattern features into the pattern prediction model to obtain the prediction result of the tendency usage pattern of the air conditioner includes:

[0036] Inputting the time series features and the usage pattern features into the pattern classifier, and combining them with the tendency setting time prediction result, to obtain the tendency usage pattern prediction result of the air conditioner;

[0037] The step of inputting the time series feature, the usage mode feature and the temperature setting feature into the numerical prediction model to obtain the prediction result of the tendency setting data of the air conditioner includes:

[0038] The temperature setting characteristics, time series characteristics and usage pattern characteristics are input into the regression analysis module, and the tendency setting time prediction result and the tendency usage pattern prediction result are combined to obtain the tendency setting data prediction result of the air conditioner.

[0039] Further, the command word label includes a tendency setting time label, a tendency usage mode label, and a tendency setting data label; the loss value of the initial user behavior prediction model is determined according to a preset loss function based on the predicted user preference corresponding to each training sample set and the user preference label, including:

[0040] Obtaining a tendency setting time loss based on the tendency setting time label and the tendency setting time prediction result;

[0041] Obtaining a tendency usage pattern loss based on the tendency usage pattern label and the tendency usage pattern prediction result;

[0042] Obtaining a tendency setting data loss based on the tendency setting data label and the tendency setting data prediction result;

[0043] The tendency setting time loss, the tendency usage mode loss, and the tendency setting data loss are integrated to obtain a total loss, and the total loss is determined as the loss value of the initial user speech recognition model.

[0044] Furthermore, it also includes: when the target air conditioner triggers the use condition corresponding to the user's command word, maintaining the current working state of the target air conditioner and broadcasting a confirmation voice through the speaker.

[0045] An air conditioning control device with continuous speech recognition, comprising:

[0046] an acquisition module, for acquiring a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner;

[0047] A training module, inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model;

[0048] a judgment module, for judging whether the target air conditioner triggers a usage condition corresponding to the user's command word according to the user's command word and the current usage status of the target air conditioner;

[0049] The execution module sends a trigger command to the target air conditioner when the target air conditioner does not trigger the usage condition corresponding to the user's command word. The trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

[0050] A computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the aforementioned air conditioning control method for continuous speech recognition.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] The present invention can accurately understand user instructions through the trained user voice recognition model, ensure the accuracy of operation, and automatically adjust according to user habits and preferences. The conflict handling mechanism enhances the reliability and robustness of the system, ensuring that appropriate choices can be made even in complex situations. At the same time, the two-way interactive function allows users to clearly understand the air conditioning status through voice broadcast, further improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flow chart of the method of the present invention;

[0054] Figure 2 This is a schematic diagram of the continuous speech recognition process of the present invention;

[0055] Figure 3 It is a schematic diagram of the structure of the device of the present invention. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0057] Example:

[0058] Combined with Figure 1 and attached Figure 2 As shown, a method for controlling an air conditioner by continuous speech recognition is applied to a speech control unit, wherein the speech control unit is communicatively connected with a speech recognition unit, a speaker, and a target air conditioner, and the method comprises:

[0059] Step S100: acquiring a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner;

[0060] Step S200: inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model;

[0061] Step S300: judging whether the target air conditioner triggers the use condition corresponding to the user's command word according to the user's command word and the current use state of the target air conditioner;

[0062] Step S400: When the target air conditioner does not trigger the usage condition corresponding to the user's command word, a trigger command is sent to the target air conditioner, and the trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

[0063] Principle of the invention: When the user speaks a voice command to the air conditioner to the device, the voice recognition unit captures the sound signal and converts it into a user voice data set in digital format. This data set contains all the voice command data issued by the user to the target air conditioner and is the basis for subsequent processing. These user voice data sets will be transmitted to the voice control unit and input into a pre-trained user voice recognition model. This model is trained based on a large number of user voice data samples and has a high degree of accuracy and recognition ability. It can parse the command words in the user's voice, that is, the specific operations that the user wants to perform, such as "turn on the air conditioner", "raise the temperature", etc. Once the user's command words are recognized, the system will make intelligent judgments based on these command words and the current usage status of the target air conditioner. This process involves the evaluation of the current state of the air conditioner (such as whether it is turned on, the current temperature, wind speed, etc.), and the analysis of the operating conditions indicated by the user's command words (such as set temperature, wind speed level, etc.). If the system determines that the current state of the target air conditioner does not meet the conditions indicated by the user's command words, then the next operation will be triggered. The system will send a trigger command to the target air conditioner. These commands include controlling the air conditioner to execute the program corresponding to the user's command words, such as turning on the air conditioner, adjusting the temperature, changing the wind speed, etc. At the same time, in order to enhance the user experience, the system will also broadcast a reply voice to the user through the speaker to confirm that the command has been received and executed, so that the user knows that the air conditioner is operating according to their instructions.

[0064] For example, let's say a user walks into a room on a hot summer day and wants to turn on the air conditioner and lower the temperature through voice control. They simply say a voice command to the device like "Turn on the air conditioner and adjust it to 22 degrees." The voice recognition unit captures these sound signals and converts them into user voice datasets. These datasets are then input into the user voice recognition model, which recognizes that the user's command words are "turn on the air conditioner" and "adjust to 22 degrees." Next, the system checks the current status of the air conditioner. If it finds that the air conditioner is not turned on or the current temperature is above 22 degrees, it will send a trigger command to the air conditioner to turn it on and adjust it to 22 degrees. At the same time, the speaker will broadcast "The air conditioner is turned on and the temperature has been adjusted to 22 degrees" to let the user know that their command has been executed.

[0065] In general, this continuous voice recognition air conditioning control method achieves accurate and convenient control of the air conditioner by integrating multiple links such as voice recognition, intelligent judgment and execution control. It not only improves the quality of life of users, but also demonstrates the broad application prospects of smart home technology.

[0066] As an implementation method, in combination with the attached Figure 1 As shown, the training process of the user speech recognition model includes:

[0067] Acquire multiple training sample sets, wherein the training sample sets are user voice data sample sets, and the user voice data sample sets include sample voice data of sample users for sample air conditioners, wherein each of the training sample sets carries a corresponding command word label;

[0068] Calling an initial user speech recognition model to divide the multiple training sample sets into multiple training batches;

[0069] Inputting the plurality of training batches into the initial user speech recognition model in sequence to obtain a predicted command word corresponding to each training sample set;

[0070] Based on the predicted command words and the command word labels corresponding to each training sample set, determining the loss value of the initial user speech recognition model according to a preset loss function;

[0071] Adjusting the model parameters of the initial user speech recognition model according to the loss value to obtain an adjusted initial user speech recognition model;

[0072] Return to the step of inputting the multiple training batches into the initial user speech recognition model in sequence, and iteratively execute the step of obtaining the predicted command word corresponding to each training sample set until the initial user speech recognition model reaches a preset training stop condition, thereby obtaining the trained user speech recognition model.

[0073] Specifically, training the user speech recognition model requires obtaining multiple training sample sets. These training sample sets are actually a collection of user speech data samples, which contain various sample speech data spoken by sample users for sample air conditioners. Each training sample set carries a corresponding command word label, which clearly indicates the specific command words contained in the sample speech data, such as "turn on the air conditioner", "turn up the temperature", etc. The training process calls an initial user speech recognition model. This initial model is usually an untrained model or a model that has only undergone a small amount of pre-training, and it does not have the ability to accurately recognize user commands. Then, the training process divides multiple training sample sets into multiple training batches, each of which contains a part of the training sample sets so that training can be performed in batches. These training batches are input into the initial user speech recognition model in sequence. For each training sample set, the model will try to predict its corresponding command word and output a prediction result. This prediction result will be compared with the command word label carried by the training sample set to evaluate the prediction accuracy of the model. In order to quantify the prediction accuracy of the model, the training process uses a preset loss function. This function calculates a loss value based on the difference between the predicted command word and the command word label. The smaller the loss value, the more accurate the model's prediction. Based on this loss value, the training process will adjust the model parameters of the initial user speech recognition model to optimize the performance of the model. The preset loss function is the cross entropy loss function, and the specific formula is:

[0074] L=-Σ[y i ×log(p i )+(1-y i )×log(1-p i )]

[0075] in:

[0076] L is the loss value;

[0077] y i is the i-th value of the actual label (0 or 1);

[0078] p i is the i-th value predicted by the model (probability between 0 and 1);

[0079] After one or more parameter adjustments, the model will become more accurate. Then, the training process returns to the step of inputting multiple training batches into the model in sequence for iterative execution. Each iteration uses the updated model parameters and recalculates the loss value. This process continues until the model reaches the preset training stop condition. This includes the loss value falling below a certain threshold, the number of iterations reaching a preset upper limit, etc. When the model meets the training stop condition, the training process ends, and the model obtained at this time is the trained user speech recognition model, which can realize the functions of continuous speech recognition and control.

[0080] For example, suppose we have an initial user speech recognition model and a set of training samples containing command words such as "turn on the air conditioner", "turn off the air conditioner", and "turn up the temperature". During the training process, the model will gradually learn the sound characteristics and semantic meanings of these command words. Through continuous iterative training, the model will recognize these command words more and more accurately and be able to accurately respond to user voice commands in actual air conditioning control scenarios.

[0081] As an implementation method, in combination with the attached Figure 3 As shown, the initial user speech recognition model includes an input module, a speech feature extraction and processing module, a time prediction model, a pattern prediction model, and a numerical prediction model; the multiple training batches are sequentially input into the initial user speech recognition model to obtain the predicted command words corresponding to each training sample set, including:

[0082] Inputting the plurality of training batches into the speech feature extraction and processing module in sequence through the input module;

[0083] Based on the speech feature extraction and processing module, feature extraction and processing are performed to obtain time series features, usage mode features and temperature setting features;

[0084] Inputting the time series features into the time prediction model module to obtain the prediction result of the tendency setting time of the air conditioner;

[0085] Inputting the time series characteristics and the usage pattern characteristics into the pattern prediction model to obtain a prediction result of the tendency usage pattern of the air conditioner;

[0086] Inputting the time series feature, the usage pattern feature and the temperature setting feature into the numerical prediction model to obtain a prediction result of the tendency setting data of the air conditioner;

[0087] The tendency setting time prediction result, the tendency use mode prediction result and the tendency setting data prediction result constitute the prediction command word.

[0088] Specifically, the initial user speech recognition model includes multiple modules: input module, speech feature extraction and processing module, time prediction model, pattern prediction model and numerical prediction model. These modules work together to extract valuable information from the user's voice command and generate predicted command words based on it. When multiple training batches are input into the model in sequence, they first enter the speech feature extraction and processing module through the input module. The task of this module is to process the original speech data and extract the features that are crucial for recognizing command words. These features include the frequency, intensity, duration of the sound, and the unique patterns in the sound. Through complex algorithms, the module can convert these raw data into a series of more abstract and useful feature representations, such as time series features, usage pattern features and temperature setting features. Time series features focus on the time-varying characteristics of voice commands, such as changes in the rhythm and tone of the user's pronunciation. Usage pattern features reflect the habits and preferences that users may have when using air conditioners, such as frequently turning on or adjusting the air conditioner during a certain period of time. The temperature setting feature is directly related to the specific temperature value that the user wants the air conditioner to reach. These features are input into three different prediction models: time prediction model, pattern prediction model and numerical prediction model. The time prediction model uses time series features to predict the time when users may want to set the air conditioner, such as predicting whether users tend to lower the air conditioner temperature before going to bed at night. The pattern prediction model predicts the user's usage pattern based on time series features and usage pattern features, such as whether the user often chooses automatic mode or energy-saving mode. The numerical prediction model combines all features to predict the specific temperature value or other numerical parameters that users may want to set.

[0089] Ultimately, the outputs of these three prediction models - the prediction results of the tendency setting time, the prediction results of the tendency usage mode, and the prediction results of the tendency setting data - together constitute the prediction command word. This prediction command word is a complex instruction that combines time, mode, and specific values, which reflects the model's best understanding of the user's intention.

[0090] For example, if a user often says "adjust the air conditioner to 26 degrees and turn it to automatic mode" after getting off work, then after enough training, the model can extract the corresponding features from the user's voice, and use the time prediction model to predict that the user may issue this command at a certain time after get off work, the pattern prediction model to predict that the user tends to use automatic mode, and the numerical prediction model to predict that the user wants to set the temperature to 26 degrees. In this way, when the user says a similar command again, the model can accurately recognize and execute the user's intention.

[0091] As an implementation method, in combination with the attached Figure 3 As shown, the speech feature extraction and processing module includes:

[0092] A user preprocessing submodule, used for preprocessing the training sample sets in the input multiple training batches;

[0093] A time series feature extraction submodule, used to extract time series features of the time series data in the training sample set;

[0094] A usage pattern feature extraction submodule, used to extract usage pattern features of usage pattern data in the training sample set;

[0095] The temperature setting feature extraction submodule is used to extract the water boiling temperature features of the water boiling temperature data in the training sample set.

[0096] The conflict handling submodule is used to determine whether there is a mode conflict or a temperature conflict in the extracted features, so as to select the corresponding mode or temperature according to the preset rules;

[0097] The speech rate waiting feedback processing submodule is used to set the waiting time according to the user's speech rate and system performance in order to process and analyze the extracted features.

[0098] Specifically, the user preprocessing submodule is responsible for preprocessing the training sample sets in the input multiple training batches. Preprocessing may include steps such as removing background noise, enhancing speech signals, and segmenting speech segments to ensure that subsequent feature extraction can be performed more accurately. For example, when a user says "turn on the air conditioner" in a noisy environment, the preprocessing submodule will try to filter out the background noise and retain only clear speech signals. The time series feature extraction submodule extracts the time series features of the time series data in the training sample set. Time series features refer to features related to the time and duration of voice commands. These features are very important for understanding the user's habits and behavior patterns. For example, if the user often says "lower the air conditioner temperature" at a certain time period in the evening, the time series features will reflect this temporal regularity. The usage pattern feature extraction submodule focuses on extracting usage pattern features of usage pattern data in the training sample set. Usage pattern features reflect the user's preferences and habits when using the air conditioner, such as commonly used modes (cooling, heating, automatic, etc.), commonly used temperature settings, etc. By extracting these features, the model can better predict the user's future behavior. However, in the process of extracting features, mode conflicts or temperature conflicts may be encountered. At this time, the conflict handling submodule will come into play. It will determine which mode or temperature should be selected based on preset rules. For example, if the user says "cooling mode" and "28 degrees" at the same time, and these two settings may conflict in actual use (because 28 degrees usually does not require cooling), then the conflict handling submodule will select a more reasonable setting based on the preset rules. In addition, the speech rate waiting feedback processing submodule is also an indispensable part of this module. Since the user's speech rate and system performance may vary, this submodule will set a reasonable waiting time based on the user's speech rate and system performance. During this waiting time, the system will wait for the user to complete the voice command and perform necessary processing and analysis. This ensures that the system can accurately recognize and understand the user's voice commands.

[0099] For example, suppose a user says "8pm, set the air conditioner to cooling mode, temperature 20 degrees". In this example, the speech feature extraction and processing module will first pre-process the speech through the user pre-processing sub-module to remove interference factors such as noise. Then, the time series feature extraction sub-module will recognize the time information "8pm", the usage mode feature extraction sub-module will recognize the usage mode information "cooling mode", and the temperature setting feature extraction sub-module (the "water boiling temperature feature extraction sub-module" in the original text may be a typo, it should be "temperature setting feature extraction sub-module") will recognize the temperature information "20 degrees". Finally, the conflict handling sub-module will ensure that there is no conflict between these settings, and the speech rate waiting feedback processing sub-module will ensure that the system has enough time to process and parse this information. Finally, this information will be integrated into a predicted command word to control the air conditioner settings.

[0100] As an implementation method, in combination with the attached Figure 3 As shown, the time prediction model includes a time series feature extraction module and a time preference prediction module, the pattern prediction model includes a pattern classifier, and the numerical prediction model includes a regression analysis module; the time series feature is input into the time prediction model module to obtain the tendency setting time prediction result of the air conditioner, including:

[0101] Inputting the time series features into the time series feature extraction module to obtain periodic features, trend features and seasonal features, and using the time preference prediction module to predict the preferred setting time of the air conditioner based on the periodic features, trend features and seasonal features;

[0102] The step of inputting the time series features and the usage pattern features into the pattern prediction model to obtain the prediction result of the tendency usage pattern of the air conditioner includes:

[0103] Inputting the time series features and the usage pattern features into the pattern classifier, and combining them with the tendency setting time prediction result, to obtain the tendency usage pattern prediction result of the air conditioner;

[0104] The step of inputting the time series feature, the usage mode feature and the temperature setting feature into the numerical prediction model to obtain the prediction result of the tendency setting data of the air conditioner includes:

[0105] The temperature setting characteristics, time series characteristics and usage pattern characteristics are input into the regression analysis module, and the tendency setting time prediction result and the tendency usage pattern prediction result are combined to obtain the tendency setting data prediction result of the air conditioner.

[0106] Specifically, the time prediction model consists of a time series feature extraction module and a time preference prediction module. The time series feature extraction module is responsible for receiving time series features, which may include the time, frequency, and duration of the user's past use of air conditioning. Through the time series feature extraction module, we can obtain three important features: periodicity, trend, and seasonality. Periodicity reflects the repetitiveness of the time when users use air conditioning, such as a fixed time period every day or every week; trend characteristics show the long-term trend of the time when users use air conditioning, such as gradually increasing or decreasing with the change of seasons; seasonal characteristics reflect the characteristics of users using air conditioning in different seasons, such as being more inclined to use cooling mode in summer and more inclined to use heating mode in winter. The time preference prediction module predicts the time when users may set air conditioning in the future based on the above three features. For example, if the user often turned on the air conditioning around 8 pm in the past, and this behavior was particularly obvious in summer, then the time preference prediction module may predict that the user will also have a high probability of turning on the air conditioning around 8 pm in the future summer. The pattern prediction model is mainly composed of pattern classifiers. The pattern classifier receives time series features and usage pattern features, which together reflect the user's habits and preferences when using the air conditioner. Combined with the prediction results of the tendency setting time obtained by the previous time prediction model, the pattern classifier can predict the usage mode that the user may choose in the future, such as cooling, heating, automatic or dehumidification. Finally, the numerical prediction model predicts the tendency setting data of the air conditioner, such as temperature, wind speed, etc., through the regression analysis module. The regression analysis module receives the temperature setting features, time series features and usage pattern features, and combines the output results of the time prediction model and the pattern prediction model to predict the air conditioning parameters that the user may set in the future. For example, if the user usually sets the temperature to around 25 degrees when turning on the air conditioner at around 8 pm in the summer in the past, then the numerical prediction model may predict that the user will also set the temperature to around 25 degrees when turning on the air conditioner at around 8 pm in the future summer. Through the collaborative work of these three models, we can achieve accurate prediction of air conditioning settings, thereby providing users with a more intelligent and personalized air conditioning control experience. For example, when the user says "turn on the air conditioner at 8 o'clock in the evening", the system can automatically set the air conditioner to the mode and temperature that the user is most likely to prefer based on the user's habits and preferences, without the user having to perform additional operations or settings.

[0107] As an implementation manner, the command word label includes a tendency setting time label, a tendency usage mode label, and a tendency setting data label; the loss value of the initial user behavior prediction model is determined according to a preset loss function based on the predicted user preference corresponding to each training sample set and the user preference label, including:

[0108] Obtaining a tendency setting time loss based on the tendency setting time label and the tendency setting time prediction result;

[0109] Obtaining a tendency usage pattern loss based on the tendency usage pattern label and the tendency usage pattern prediction result;

[0110] Obtaining a tendency setting data loss based on the tendency setting data label and the tendency setting data prediction result;

[0111] The tendency setting time loss, the tendency usage mode loss, and the tendency setting data loss are integrated to obtain a total loss, and the total loss is determined as the loss value of the initial user speech recognition model.

[0112] Specifically, the system will label each sample with a corresponding user preference label based on the collected training sample set. These labels include the tendency setting time label, the tendency usage pattern label, and the tendency setting data label. For example, if a user often sets the air conditioner to cooling mode and adjusts the temperature to 24 degrees at a certain time point, then the user's tendency setting time label may be the time point, the tendency usage pattern label is cooling, and the tendency setting data label is 24 degrees. The system will use the initial user behavior prediction model to predict these training samples and obtain the tendency setting time prediction results, the tendency usage pattern prediction results, and the tendency setting data prediction results. These prediction results represent the user preferences inferred by the model based on the input features (such as time series features, usage pattern features, temperature setting features, etc.). In order to evaluate the prediction accuracy of the model, the system will calculate the loss value based on the preset loss function. Specifically, the system will calculate the tendency setting time loss, tendency usage pattern loss, and tendency setting data loss respectively. The tendency setting time loss measures the difference between the air conditioner setting time predicted by the model and the user's actual preferred time; the tendency usage pattern loss reflects the inconsistency between the air conditioner usage pattern predicted by the model and the user's actual preferred pattern; the tendency setting data loss evaluates the gap between the specific air conditioner setting data (such as temperature) predicted by the model and the user's actual preferred data. When calculating these losses, the system may use different loss functions, such as mean square error (MSE) or cross entropy loss, depending on the nature and data type of the prediction task. For example, for time prediction, MSE may be used to measure the gap between the predicted time and the actual time; for pattern prediction and data prediction, cross entropy loss may be used to evaluate the inconsistency between the prediction result and the actual label. Finally, the system will fuse these three loss values ​​to obtain a total loss value. This total loss value reflects the overall performance of the model in predicting the user's air conditioner setting preference. In order to optimize the model, the system will try to adjust the model's parameters to minimize this total loss value. Through continuous iterative training and parameter adjustment, the model can gradually learn the user's true preferences, so as to more accurately reflect the user's air conditioner setting needs in future predictions.

[0113] For example, suppose a user often sets the air conditioner to cooling mode and the temperature to 24 degrees at 8 pm. During the training process, the system will collect multiple training samples of this user and label each sample accordingly (such as the tendency to set the time label to 8 pm, the tendency to use the mode label to cool, and the tendency to set the data label to 24 degrees). Then, the system will use the initial user behavior prediction model to predict these samples and get the prediction results. By calculating the loss value, the system can find the accuracy of the model in predicting that the user sets the air conditioner to cooling mode and the temperature to 24 degrees at 8 pm. If there is a large difference between the predicted result and the actual label, the system will adjust the parameters of the model to improve the prediction accuracy. After multiple iterations of training, the model will be able to more accurately predict the user's air conditioning setting preferences.

[0114] As an implementation mode, it also includes: when the target air conditioner triggers the use condition corresponding to the user's command word, the current working state of the target air conditioner is maintained, and a confirmation voice is broadcast through the speaker.

[0115] Specifically, the description of "when the target air conditioner triggers the use condition corresponding to the user's command word, the current working state of the target air conditioner is maintained, and a confirmation voice is broadcast through the speaker" means that after the user issues a specific voice command, the system will determine whether these commands correspond to the scene where the current working state of the air conditioner needs to be maintained. If so, the system will not make any changes to the working state of the air conditioner, but maintain the status quo, and broadcast a confirmation voice to the user through the speaker, informing the user that the air conditioner has maintained the current working state as required.

[0116] Let's take an example to illustrate this function: suppose the user is using an air conditioner, and the current air conditioner is set to cooling mode and the temperature is 24 degrees. The user suddenly has to leave the room for a while, but he hopes that the air conditioner will still maintain the current cooling mode and temperature settings when he returns. Therefore, the user can use voice command to the system: "After I leave, please keep the current settings of the air conditioner." When the system recognizes this command and determines that it meets the use condition of "maintaining the current working state", the system will not make any adjustments to the air conditioner, but keep the cooling mode and the temperature setting of 24 degrees unchanged. At the same time, the system will broadcast a confirmation voice through the speaker, such as: "OK, the air conditioner will maintain the current settings unchanged."

[0117] This design not only improves the intelligence level of the system, but also allows users to control the air conditioner more conveniently without having to worry about the air conditioner settings being changed after leaving the room. At the same time, by broadcasting the confirmation voice, the system also gives users clear feedback, letting them know that their commands have been correctly understood and executed, thereby enhancing the user experience.

[0118] As another embodiment of the present application, an air conditioning control device for continuous speech recognition includes:

[0119] an acquisition module, for acquiring a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner;

[0120] A training module, inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model;

[0121] a judgment module, for judging whether the target air conditioner triggers a usage condition corresponding to the user's command word according to the user's command word and the current usage status of the target air conditioner;

[0122] The execution module sends a trigger command to the target air conditioner when the target air conditioner does not trigger the usage condition corresponding to the user's command word. The trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

[0123] As another embodiment of the present application, a computer storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the aforementioned air conditioning control method of continuous speech recognition are implemented.

[0124] Although the present invention is described herein with reference to the illustrative embodiments of the present invention, the above embodiments are only preferred embodiments of the present invention, and the embodiments of the present invention are not limited to the above embodiments. It should be understood that those skilled in the art can design many other modifications and embodiments, which will fall within the scope and spirit of the principles disclosed in this application.

Claims

1. A method for controlling air conditioners by continuous speech recognition, characterized in that: Applied to a voice control unit, the voice control unit is communicatively connected with a voice recognition unit, a speaker, and a target air conditioner, the method comprising: Acquire a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner; Inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model; According to the user's command word and the current usage status of the target air conditioner, determining whether the target air conditioner triggers the usage condition corresponding to the user's command word; When the target air conditioner does not trigger the usage condition corresponding to the user's command word, a trigger command is sent to the target air conditioner, and the trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

2. The air conditioning control method based on continuous speech recognition according to claim 1, characterized in that: The training process of the user speech recognition model includes: Acquire multiple training sample sets, wherein the training sample sets are user voice data sample sets, and the user voice data sample sets include sample voice data of sample users for sample air conditioners, wherein each of the training sample sets carries a corresponding command word label; Calling an initial user speech recognition model to divide the multiple training sample sets into multiple training batches; Inputting the plurality of training batches into the initial user speech recognition model in sequence to obtain a predicted command word corresponding to each training sample set; Based on the predicted command words and the command word labels corresponding to each training sample set, determining the loss value of the initial user speech recognition model according to a preset loss function; Adjusting the model parameters of the initial user speech recognition model according to the loss value to obtain an adjusted initial user speech recognition model; Return to the step of inputting the multiple training batches into the initial user speech recognition model in sequence, and iteratively execute the step of obtaining the predicted command word corresponding to each training sample set until the initial user speech recognition model reaches a preset training stop condition, thereby obtaining the trained user speech recognition model.

3. The air conditioning control method based on continuous speech recognition according to claim 2, characterized in that: The initial user speech recognition model includes an input module, a speech feature extraction and processing module, a time prediction model, a pattern prediction model, and a numerical prediction model; the multiple training batches are sequentially input into the initial user speech recognition model to obtain a predicted command word corresponding to each training sample set, including: Inputting the plurality of training batches into the speech feature extraction and processing module in sequence through the input module; Based on the speech feature extraction and processing module, feature extraction and processing are performed to obtain time series features, usage mode features and temperature setting features; Inputting the time series features into the time prediction model module to obtain the prediction result of the tendency setting time of the air conditioner; Inputting the time series characteristics and the usage pattern characteristics into the pattern prediction model to obtain a prediction result of the tendency usage pattern of the air conditioner; Inputting the time series feature, the usage pattern feature and the temperature setting feature into the numerical prediction model to obtain a prediction result of the tendency setting data of the air conditioner; The tendency setting time prediction result, the tendency use mode prediction result and the tendency setting data prediction result constitute the prediction command word.

4. The air conditioning control method based on continuous speech recognition according to claim 3, characterized in that: The speech feature extraction and processing module comprises: A user preprocessing submodule, used for preprocessing the training sample sets in the input multiple training batches; A time series feature extraction submodule, used to extract time series features of the time series data in the training sample set; A usage pattern feature extraction submodule, used to extract usage pattern features of usage pattern data in the training sample set; The temperature setting feature extraction submodule is used to extract the water boiling temperature features of the water boiling temperature data in the training sample set. The conflict handling submodule is used to determine whether there is a mode conflict or a temperature conflict in the extracted features, so as to select the corresponding mode or temperature according to the preset rules; The speech rate waiting feedback processing submodule is used to set the waiting time according to the user's speech rate and system performance in order to process and analyze the extracted features.

5. The air conditioning control method based on continuous speech recognition according to claim 3, characterized in that: The time prediction model includes a time series feature extraction module and a time preference prediction module, the mode prediction model includes a mode classifier, and the numerical prediction model includes a regression analysis module; the time series feature is input into the time prediction model module to obtain the tendency setting time prediction result of the air conditioner, including: Inputting the time series features into the time series feature extraction module to obtain periodic features, trend features and seasonal features, and using the time preference prediction module to predict the preferred setting time of the air conditioner based on the periodic features, trend features and seasonal features; The step of inputting the time series features and the usage pattern features into the pattern prediction model to obtain the prediction result of the tendency usage pattern of the air conditioner includes: Inputting the time series features and the usage pattern features into the pattern classifier, and combining them with the tendency setting time prediction result, to obtain the tendency usage pattern prediction result of the air conditioner; The step of inputting the time series feature, the usage mode feature and the temperature setting feature into the numerical prediction model to obtain the prediction result of the tendency setting data of the air conditioner includes: The temperature setting characteristics, time series characteristics and usage pattern characteristics are input into the regression analysis module, and the tendency setting time prediction result and the tendency usage pattern prediction result are combined to obtain the tendency setting data prediction result of the air conditioner.

6. The air conditioning control method based on continuous speech recognition according to claim 3, characterized in that: The command word label includes a tendency setting time label, a tendency usage mode label, and a tendency setting data label; the loss value of the initial user behavior prediction model is determined according to a preset loss function based on the predicted user preference corresponding to each training sample set and the user preference label, including: Obtaining a tendency setting time loss based on the tendency setting time label and the tendency setting time prediction result; Obtaining a tendency usage pattern loss based on the tendency usage pattern label and the tendency usage pattern prediction result; Obtaining a tendency setting data loss based on the tendency setting data label and the tendency setting data prediction result; The tendency setting time loss, the tendency usage mode loss, and the tendency setting data loss are integrated to obtain a total loss, and the total loss is determined as the loss value of the initial user speech recognition model.

7. The air conditioning control method based on continuous speech recognition according to claim 1, characterized in that: Also includes: When the target air conditioner triggers the use condition corresponding to the user's command word, the current working state of the target air conditioner is maintained, and a confirmation voice is broadcast through the speaker.

8. An air conditioning control device with continuous speech recognition, characterized in that: include: an acquisition module, for acquiring a user voice data set uploaded by the voice recognition unit, wherein the user voice data set includes voice command data of the user for the target air conditioner; A training module, inputting the user voice data set into a pre-trained user voice recognition model, and recognizing the user's command words based on the user voice recognition model; a judgment module, for judging whether the target air conditioner triggers a usage condition corresponding to the user's command word according to the user's command word and the current usage status of the target air conditioner; The execution module sends a trigger command to the target air conditioner when the target air conditioner does not trigger the usage condition corresponding to the user's command word. The trigger command includes controlling the target air conditioner to execute the program corresponding to the user's command word and broadcasting a reply voice through the speaker.

9. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the air conditioning control method based on continuous speech recognition described in any one of claims 1 to 7 are implemented.

Images