An intelligent voice recognition control system and method based on the Internet of Things

Abstract

The present invention relates to an intelligent voice recognition control system and method based on the Internet of Things, belonging to the field of voice recognition control technology. It includes a user intention feedback generation module and an execution control module. The user intention feedback generation module is used to predict the user intention by using the Internet of Things instruction library through the collected Internet of Things perception data, and actively feed back the predicted user intention to the user; the execution control module is used to judge the user intention feedback and execute the operation. If the generated user intention feedback is correct, the device execution instruction is generated, and the instruction execution record and the corresponding Internet of Things perception data are used to generate user control rules and add them to the Internet of Things instruction library. The present invention simultaneously collects the real-time information of smart home devices and the user voice instruction information, generates user intention feedback according to the Internet of Things instruction library, enhances the system context perception, and then performs voice recognition in a progressive manner to obtain the user's control intention, thereby improving the accuracy, diversity and real-time performance of voice recognition control.

Description

Technical Field

[0001] The present invention belongs to the technical field of speech recognition control, and in particular relates to an intelligent speech recognition control system and method based on the Internet of Things. Background Art

[0002] The smart voice home system uses voice recognition to control various smart devices in residents' homes, so as to achieve the purpose of controlling various smart devices with voice. As an important human-computer interaction method, voice recognition technology is widely used in smart speakers, voice assistants and other fields. However, there are still some problems with voice control in smart home control systems: low recognition accuracy: because voice signals are affected by environmental noise, inaccurate voice pronunciation and other factors, the accuracy of the voice recognition system in controlling devices is not high; insufficient command voice diversity: the current voice recognition system does not support the diversity of command voices, the voice command library is not perfect, and it cannot fit the user's behavior habits well; insufficient real-time performance: the processing time of the voice recognition system is long, and the interaction process is cumbersome, which cannot meet the needs of real-time control of smart home systems.

[0003] Therefore, there is an urgent need to propose an intelligent speech recognition control system and method based on the Internet of Things to improve the accuracy, diversity and real-time performance of speech recognition control in response to the problems existing in the above-mentioned speech control. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides an intelligent voice recognition control system and method based on the Internet of Things, which simultaneously collects real-time information of smart home devices and user voice command information, generates user intention feedback according to the Internet of Things command library, enhances the system context perception, and then performs voice recognition in a progressive manner to obtain the user's control intention, thereby improving the accuracy, diversity and real-time performance of voice recognition control.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A first aspect of the present disclosure provides an intelligent speech recognition control system based on the Internet of Things, including a user intention feedback generation module and an execution control module;

[0007] The user intention feedback generation module is used to predict the user intention by using the IoT instruction library through the collected IoT perception data, and after receiving the wake-up instruction, actively feed back the predicted user intention to the user in the form of voice;

[0008] The execution control module is used to judge the user intention feedback and execute the operation. If the generated user intention feedback is correct, the device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library;

[0009] If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online;

[0010] The device execution action after online voice recognition and the corresponding IoT perception data are used to correct the user control rules with incorrect user intention feedback, and the corrected user control rules are matched with the recognized voice features and stored in the voice rule library.

[0011] As a preferred technical solution of the present invention, the construction of the IoT instruction library includes the following steps:

[0012] Frequent item set mining: Use association rule mining algorithm to mine frequent item sets in the IoT perception database by setting the minimum support threshold;

[0013] Association rule generation: Based on the results of frequent item set mining, all possible association rules are generated, and the final association rules are screened by calculating support and confidence;

[0014] Evaluate rules: Test the effectiveness of rules in actual applications, check their accuracy and response time, and adjust algorithm parameters through actual usage feedback;

[0015] Establish an IoT instruction library: Build an IoT instruction library through empirical rules and association rules, where the empirical rules include automated tasks created based on set trigger conditions and actuator operation behaviors.

[0016] As a preferred technical solution of the present invention, it also includes an IoT sensing data acquisition module, which is used for the smart device and microphone device of the smart home control system to collect real-time information of the home environment and the user's voice command information;

[0017] And formulate the IoT perception data collection format based on time series, including time, voice command, device number, device name, device data, device status and device category;

[0018] Among them, real-time information includes various types of information such as switch quantity, analog quantity, internal quantity and parameter; voice commands include wake-up commands and control commands.

[0019] As a preferred technical solution of the present invention, it also includes an IoT sensing data processing module, which is used to process the collected IoT sensing data, complete sensor data standardization and voice data feature extraction, and the execution steps include:

[0020] Smart device data processing: By filtering out outliers and noise in the data, converting the data into a unified format and range, and filling in missing data;

[0021] Voice data processing: preprocessing and feature extraction of collected voice command data;

[0022] Data integration: The pre-processed smart device data and voice feature data are merged into item sets according to timestamps, and the IoT perception database is built by integrating them into the same data framework.

[0023] As a preferred technical solution of the present invention, it also includes a voice online recognition module, which is used to online recognize the collected user voice control instructions, and specifically includes the following steps:

[0024] The feature vector is obtained by extracting the speech features, and the speech feature vector is input into the constructed deep convolutional neural network recognition model to obtain relatively rough speech acoustic features through the convolution layer;

[0025] After the output, it enters the second convolution layer, and after being mapped by a nonlinear activation function, fine speech features are extracted. The convolutional speech features are sent to the pooling layer, and the maximization method is used to reduce the convolution parameter error.

[0026] The convolutional layer and the pooling layer output the speech feature map and then send it to the fully connected layer, and the corresponding posterior probability is obtained after the convolution operation. The CTC of the softmax layer is used to classify and optimize the speech features extracted by the above layers, and the decoding operation is used to recognize the speech features and then output them.

[0027] As a preferred technical solution of the present invention, the construction of the deep convolutional neural network recognition model includes:

[0028] The extracted features are input into the model for classification training, and the output result is the final recognition result; the deep neural network is set up with four layers, namely convolution layer, pooling layer, fully connected layer and softmax layer.

[0029] A second aspect of the present disclosure provides an intelligent voice recognition control method based on the Internet of Things, which is applied to an intelligent voice recognition control system based on the Internet of Things as described above, and includes the following steps:

[0030] IoT sensing data collection: collect real-time information about the home environment and user voice command information through smart devices and microphone devices of the smart home control system;

[0031] Formulate a time-series-based IoT sensing data collection format, including time, voice command, device number, device name, device data, device status, and device category;

[0032] IoT perception data processing: By processing the collected IoT perception data, sensor data standardization and voice data feature extraction are completed;

[0033] User intention feedback generation: The user intention is predicted by using the collected IoT perception data and the IoT command library. After receiving the wake-up command, the predicted user intention is actively fed back to the user in the form of voice.

[0034] Execution instruction judgment: If the generated user intention feedback is correct, a device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library;

[0035] If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online;

[0036] The device execution action after online voice recognition and the corresponding IoT perception data are used to correct the user control rules with incorrect user intention feedback, and the modified user control rules are matched with the voice features and stored in the voice rule library.

[0037] Furthermore, the voice data feature extraction comprises the following steps:

[0038] Perform preprocessing operations on the collected voice command data, including signal filtering and endpoint detection;

[0039] The wavelet transform method is used to perform time-frequency domain conversion on the recorded speech signal to obtain the wavelet coefficients of each signal. The wavelet coefficient spectrum is obtained by calculating the wavelet coefficients, and the wavelet coefficients are arranged from low to high and then spliced. The speech feature sequence is established based on the splicing results.

[0040] Calculate the recorded speech feature sequence, and the result is the power spectrum calculation result. The calculation formula is as follows:

[0041] ;

[0042] In the formula, P ( ω ) represents the power spectrum, ω represents the wavelet coefficient spectrum of the initial speech signal, FT ( ω ) represents the value generated after the time-frequency conversion of the initial speech signal, and T represents the total number of speech frames;

[0043] The filter is used to filter the speech signal. The time domain impulse response of the filter is expressed by the following equation:

[0044] ;

[0045] In the formula, g ( t ) means in t The filter time domain impulse response at time is b is the filter bandwidth, a is the coefficient, n is the filter order, e In order to respond to the impulse, is the filter function;

[0046] Compensate for the background noise generated by speech and calculate the long-time frame power of speech;

[0047] After adjusting the power, the time-frequency and power of the multi-channel speech are normalized. The process of normalizing the speech power is as follows: the average power of the recorded speech sample is estimated, and the estimated result is subtracted from the current power;

[0048] Performing power function nonlinear processing on the power normalized value, and transforming the speech nonlinear processing signal sequence by discrete cosine transform to obtain speech feature parameters;

[0049] The first-order and second-order differences of the acquired speech feature parameters are then concatenated with the original values ​​as the final speech features.

[0050] Furthermore, the voice long-term frame power is calculated by the following formula:

[0051] ;

[0052] In the formula, Indicates the long-term frame power of speech. P ( s , c ) represents the power spectrum of a frame, s represents a frame of speech, c Indicates the number of channels in the filter.

[0053] Furthermore, the power is adjusted and the calculation formula is as follows:

[0054] ;

[0055] In the formula, is the transfer function, c represents the number of channels in the filter, c 1 and c 2 respectively represent the number of channels before and after normalization, represents the background noise coefficient, Represents the adjusted power spectrum value.

[0056] The beneficial effects of the present invention are:

[0057] The present invention simultaneously collects real-time information of smart home devices and user voice command information, first generates user intention feedback through the Internet of Things command library based on the real-time information, enhances context perception to grasp user needs and actively controls the device, then judges the user intention feedback and performs voice recognition in a progressive manner to obtain the user's control intention, quickly and accurately controls the device, and improves operating efficiency; at the same time, during the operation of the system, the accuracy of user intention feedback generation is improved by continuously updating the Internet of Things command library, and a voice rule library is established to quickly identify user control commands, reduce the frequency of online voice recognition, and improve the responsiveness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0059] Figure 1 A schematic diagram of the structure of an intelligent voice recognition control system based on the Internet of Things provided by an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the steps for constructing an IoT instruction library provided by an embodiment of the present invention;

[0061] Figure 3 A schematic diagram of the steps of an intelligent voice recognition control method based on the Internet of Things provided by an embodiment of the present invention;

[0062] Figure 4 A schematic diagram of the steps of extracting features from speech data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, characteristics and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0064] At least one embodiment of the present disclosure provides an intelligent voice recognition control system based on the Internet of Things, such as Figure 1 As shown, it includes an IoT perception data acquisition module, an IoT perception data processing module, a user intention feedback generation module, an execution control module and a voice online recognition module.

[0065] The IoT perception data collection module is used for the smart devices and microphones of the smart home control system to collect real-time information of the home environment and the user's voice command information; and formulates a time series-based IoT perception data collection format, including time, voice command, device number, device name, device data, device status and device category.

[0066] Among them, real-time information includes various types of information such as switch quantity, analog quantity, internal quantity, parameters, etc.; voice commands include wake-up commands and control commands.

[0067] The IoT sensing data processing module is used to process the collected IoT sensing data, complete sensor data standardization and voice data feature extraction, and the execution steps include:

[0068] Smart device data processing: By filtering out outliers and noise in the data, converting the data into a unified format and range, and filling in missing data.

[0069] Voice data processing: preprocessing and feature extraction of collected voice command data;

[0070] Data integration: The pre-processed smart device data and voice feature data are merged into item sets according to timestamps, and the IoT perception database is built by integrating them into the same data framework.

[0071] The user intention feedback generation module is used to predict the user intention by using the collected IoT perception data and the IoT instruction library, and after receiving the wake-up instruction, actively feed back the predicted user intention to the user in the form of voice.

[0072] Among them, Figure 2 As shown, the construction of the IoT instruction library includes the following steps:

[0073] Frequent item set mining: Use the Apriori algorithm or FP-growth algorithm to mine frequent item sets in the IoT perception database by setting a minimum support threshold;

[0074] Association rule generation: Based on the results of frequent item set mining, all possible association rules are generated, and the final association rules are screened by calculating support and confidence.

[0075] Evaluate rules: Test the effectiveness of rules in real-world applications, check their accuracy and response time, and adjust algorithm parameters through actual usage feedback.

[0076] Establish an IoT instruction library: Build an IoT instruction library through empirical rules and association rules. The empirical rules include automated tasks created based on set trigger conditions and actuator operation behaviors, so as to improve the IoT instruction library and enhance user experience and satisfaction.

[0077] The execution control module is used to judge the user intention feedback and execute the operation. If the generated user intention feedback is correct, the device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library;

[0078] If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online;

[0079] The user control rules adopted by the user intention feedback are modified by using the voice features and device actions of the control instructions and the corresponding IoT perception data. The modified user control rules are matched with the voice features and stored in the voice rule library.

[0080] The voice online recognition module is used to online recognize the collected user voice control instructions, and includes the following steps:

[0081] The feature vector is obtained by extracting the speech features, and the speech feature vector is input into the constructed deep convolutional neural network recognition model to obtain relatively rough speech acoustic features through the convolution layer;

[0082] Construct a deep convolutional neural network recognition model, input the extracted features into the model for classification training, and the output is the final recognition result to achieve speech recognition. A deep neural network contains four layers, namely convolution layer, pooling layer, fully connected layer and softmax layer.

[0083] After the output, it enters the second convolution layer, and after being mapped by a nonlinear activation function, fine speech features are extracted. The convolved speech features are sent to the pooling layer, and the maximization method is used to reduce the convolution parameter error, so as to further obtain more accurate speech features.

[0084] The convolutional layer and the pooling layer output the speech feature map and then send it to the fully connected layer, and the corresponding posterior probability is obtained after the convolution operation. The CTC of the softmax layer is used to classify and optimize the speech features extracted by the above layers, and the decoding operation is used to recognize the speech features and then output them.

[0085] An intelligent speech recognition control method based on the Internet of Things, such as Figure 3 As shown, the following steps are included:

[0086] S1. IoT sensing data collection: collect real-time information of the home environment and user's voice command information through smart devices and microphones of the smart home control system;

[0087] Among them, real-time information includes various types of information such as switch quantity, analog quantity, internal quantity, parameters, etc.; voice commands include wake-up commands and control commands.

[0088] Develop a time series-based IoT perception data collection format, including time, voice command, device number, device name, device data, device status, and device category.

[0089] It is understandable that the switch quantity is also called discrete quantity, which is mainly used to indicate the switch state. For example, whether the circuit is connected, whether the smart device is started, etc., which are used for simple logic control and status indication. The analog quantity is used to represent the range value, which can continuously change between two limits. Common analog quantities include temperature, humidity, pressure, flow, light intensity, sound intensity and gas concentration, etc. These data are measured by the corresponding sensors and converted into digital signals by analog / digital converters for processing. The internal quantity is used to assist logic control through the parameters obtained from the inside of the smart device.

[0090] In this embodiment, not only voice data is collected, but also environmental condition data and device data are collected with the help of existing smart home control system sensors for subsequent user intention feedback prediction, enhancing context perception so that the system can better understand user needs, while improving feedback accuracy and more accurately predicting user needs.

[0091] S2. IoT perception data processing: The collected IoT perception data is processed to complete sensor data standardization and voice data feature extraction, including the following steps:

[0092] S21. Smart device data processing: Filter out outliers and noise in the data, convert the data to a uniform format and range, fill in missing data, and use methods such as interpolation, averaging, or nearest neighbor data.

[0093] It is understandable that the smart device data includes the collected data of the sensor and the action parameters of the actuator.

[0094] S22, voice data processing: pre-processing and feature extraction of the collected voice command data, such as Figure 4 As shown, the following steps are included:

[0095] S221, performing preprocessing operations on the collected voice command data, including signal filtering, endpoint detection, etc.;

[0096] S222, using a wavelet transform method to perform time-frequency domain conversion on the recorded speech signal, obtaining the wavelet coefficients of each signal, obtaining the wavelet coefficient spectrum by calculating the wavelet coefficients, arranging them from low to high and then splicing them, and establishing a speech feature sequence based on the splicing result;

[0097] S223, calculating the recorded speech feature sequence, the calculation result is the power spectrum calculation result, the calculation formula is as follows:

[0098] ;

[0099] In the formula, P ( ω ) represents the power spectrum, ω represents the wavelet coefficient spectrum of the initial speech signal, F T ( ω ) represents the value generated after the time-frequency conversion of the initial speech signal, and T represents the total number of speech frames.

[0100] S224, using a filter to perform filtering analysis on the speech signal, the time domain impulse response generated by the filter is expressed by the following equation:

[0101] ;

[0102] In the formula, g ( t ) means in t The filter time domain impulse response at time is b is the filter bandwidth, a is the coefficient, n is the filter order, e In order to respond to the impulse, is the filter function.

[0103] S225. In order to obtain the signal spectrum sequence of speech, it is necessary to compensate for the background noise generated by the speech. Then, the long-time frame power of the speech must be calculated, and the asymmetric noise that appears must be suppressed. The equation is defined as follows:

[0104] ;

[0105] In the formula, Indicates the long-term frame power of speech. P ( s , c ) represents the power spectrum of a frame, s Represents a frame of speech.

[0106] S226, after adjusting the power, normalize the time-frequency and power of the multi-channel voice, and the calculation formula is as follows:

[0107] ;

[0108] In the formula, is the transfer function, c represents the number of channels in the filter, c 1 and c 2 respectively represent the number of channels before and after normalization, represents the background noise coefficient, Represents the adjusted power spectrum value.

[0109] The process of normalizing the speech power is as follows: estimating the average power of the recorded speech sample and subtracting the estimated result from the current power.

[0110] S227, performing power function nonlinear processing on the power normalized value, and transforming the speech nonlinear processing signal sequence by discrete cosine transform to obtain speech feature parameters;

[0111] S228. In order to better reflect the dynamic continuity of the sound signal and improve the recognition accuracy, the first-order and second-order differences of the acquired speech feature parameters are taken and then concatenated with the original values ​​as the final speech features.

[0112] S23, data integration: merge the pre-processed smart device data and voice feature data into item sets according to timestamps, and build an IoT perception database by integrating them into the same data framework.

[0113] It is understandable that by integrating smart device data and voice feature data based on timestamps, accurate comparison and association analysis can be performed, and by building an IoT perception database, preparations can be made for mining association rules.

[0114] S3, user intention feedback generation: The user intention is predicted by using the IoT instruction library through the collected IoT perception data. After receiving the wake-up instruction, the predicted user intention is actively fed back to the user in the form of voice, for example: "Do you want to turn on the bedroom light?" The construction of the IoT instruction library includes the following steps:

[0115] S31, frequent item set mining: using Apriori algorithm or FP-growth algorithm to mine frequent item sets in IoT sensing database by setting the minimum support threshold;

[0116] S32. Association rule generation: Based on the results of frequent item set mining, all possible association rules are generated, and the final association rules are screened by calculating support and confidence.

[0117] Association rules are implications of the form X⇒Y, where X={x1, x2, …, x m}⊂I,Y={y1,y2,…,y n}⊂I, X and Y are different item sets and are not empty. X⇒Y means that in a transaction, if the antecedent X appears, then the consequent Y will appear with a certain probability. Assuming there is an association rule X⇒Y, then the support Support(X⇒Y)=P(X,Y) represents the possibility of a certain item set X∪Y appearing in the transaction database; the confidence Confidence(X⇒Y)= P(Y|X)= P(X,Y) / P(X) represents the probability of the consequent occurring under the condition that the antecedent occurs, which is the possibility of a conditional probability occurring.

[0118] S33. Evaluate rules: Test the effectiveness of rules in actual applications, check their accuracy and response time, and adjust algorithm parameters such as support and confidence thresholds through actual usage feedback to improve speech recognition accuracy and response efficiency.

[0119] S34. Establish an IoT instruction library: Build an IoT instruction library through empirical rules and association rules, where the empirical rules include automated tasks created based on set trigger conditions and actuator operation behaviors, so as to improve the IoT instruction library and enhance user experience and satisfaction.

[0120] It should be noted that in the subsequent operation of the system, in the case of incorrect user intent feedback, the user command is added to the IoT perception data through voice recognition, and the association rules are generated to improve the IoT command library. It is also used to correct the empirical rules. For the executed empirical rules, the empirical rules are removed after the user issues a pause execution command. In the above way, the IoT command library is continuously improved during the operation of the system, making the system more responsive and the user intent feedback more accurate, while reducing the amount of calculation for voice recognition in the system.

[0121] S4, execution instruction judgment: if the generated user intention feedback is correct, then a device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library;

[0122] It is understandable that if the user's intention feedback is correct, complex voice recognition will not be performed, which improves the efficiency of the system response and reduces the amount of data calculation for the system.

[0123] If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online;

[0124] The device execution action after online voice recognition and the corresponding IoT perception data are used to correct the user control rules with incorrect user intention feedback, and the modified user control rules are matched with the voice features and stored in the voice rule library.

[0125] It is understandable that the IoT instruction library mainly makes rule decisions based on the collected real-time information, while the voice rule library is established by collecting correct user rules for online voice recognition during operation. In the subsequent application process, as the voice rule library is enriched, user voice recognition can be completed through the voice rule library, reducing the frequency of online voice recognition and improving the responsiveness of the system.

[0126] Among them, the online speech recognition includes the following steps:

[0127] S41, obtaining a feature vector through the extracted speech features, inputting the speech feature vector into the constructed deep convolutional neural network recognition model, and obtaining relatively rough speech acoustic features through the convolution layer;

[0128] Construct a deep convolutional neural network recognition model, input the extracted features into the model for classification training, and the output is the final recognition result to achieve speech recognition. A deep neural network contains four layers, namely convolution layer, pooling layer, fully connected layer and softmax layer.

[0129] S42, after the output, enters the second convolution layer, extracts fine speech features after mapping by nonlinear activation function, transmits the convolutional speech features to the pooling layer, uses the maximization method to reduce the convolution parameter error, and further obtains more accurate speech features;

[0130] S43, the convolution layer and the pooling layer output the speech feature map and then send it to the fully connected layer, and obtain the corresponding posterior probability after the convolution operation; the CTC of the softmax layer is used to classify and optimize the speech features extracted by the above layers, and the decoding operation is used to recognize the speech features and then output them.

[0131] In this embodiment, the IoT command library is used to feedback user intentions, while the voice rule library is used to quickly recognize voice when the feedback of user intentions is incorrect. That is, the IoT command library is for wake-up commands, and the voice rule library is for control commands. By setting the voice rule library, the system's personalization level for automatically generating user rules is improved, making the voice control system more suitable for practical applications. At the same time, the IoT command library and the voice rule library are used to guide users to rapidly and progressively realize the control of smart devices, providing a more intuitive and user-friendly interaction method, allowing users to quickly and accurately control devices through natural language, thereby improving operational efficiency.

[0132] The present invention simultaneously collects real-time information of smart home devices and user voice command information, first generates user intention feedback through the Internet of Things command library based on the real-time information, enhances context perception to grasp user needs and actively controls the device, then judges the user intention feedback and performs voice recognition in a progressive manner to obtain the user's control intention, quickly and accurately controls the device, and improves operating efficiency; at the same time, during the operation of the system, the accuracy of user intention feedback generation is improved by continuously updating the Internet of Things command library, and a voice rule library is established to quickly identify user control commands, reduce the frequency of online voice recognition, and improve the responsiveness of the system.

[0133] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An intelligent voice recognition control system based on the Internet of Things, characterized by: It includes IoT sensing data acquisition module, IoT sensing data processing module, user intention feedback generation module, execution control module and voice online recognition module; The IoT sensing data processing module is used to process the collected IoT sensing data, complete sensor data standardization and voice data feature extraction, and the execution steps include: Smart device data processing: By filtering out outliers and noise in the data, converting the data into a unified format and range, and filling in missing data; Voice data processing: preprocessing and feature extraction of collected voice command data; Data integration: The pre-processed smart device data and voice feature data are merged into item sets according to timestamps, and the IoT perception database is constructed by integrating them into the same data framework; The user intention feedback generation module is used to predict the user intention by using the IoT instruction library through the collected IoT perception data, and after receiving the wake-up instruction, actively feed back the predicted user intention to the user in the form of voice; The execution control module is used to judge the user intention feedback and execute the operation. If the generated user intention feedback is correct, the device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library; If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online; The device execution action after online speech recognition and the corresponding IoT perception data are used to correct the user control rules with incorrect user intention feedback, and the corrected user control rules are matched with the recognized speech features and stored in the speech rule library; The construction of the IoT instruction library includes the following steps: Frequent item set mining: Use association rule mining algorithm to mine frequent item sets in the IoT perception database by setting the minimum support threshold; Association rule generation: Based on the results of frequent item set mining, all possible association rules are generated, and the final association rules are screened by calculating support and confidence; Evaluate rules: Test the effectiveness of rules in actual applications, check their accuracy and response time, and adjust algorithm parameters through actual usage feedback; Establish an IoT instruction library: Build an IoT instruction library through empirical rules and association rules, where the empirical rules include automated tasks created based on set trigger conditions and actuator operation behaviors.

2. According to claim 1, an intelligent voice recognition control system based on the Internet of Things is characterized by: The IoT sensing data collection module is used for the smart devices and microphone devices of the smart home control system to collect real-time information of the home environment and the user's voice command information; And formulate the IoT perception data collection format based on time series, including time, voice command, device number, device name, device data, device status and device category; Among them, real-time information includes various types of information such as switch quantity, analog quantity, internal quantity and parameter; voice commands include wake-up commands and control commands.

3. The intelligent voice recognition control system based on the Internet of Things according to claim 1 is characterized in that: The online voice recognition module is used to online recognize the collected user voice control instructions, and specifically includes the following steps: The feature vector is obtained by extracting the speech features, and the speech feature vector is input into the constructed deep convolutional neural network recognition model to obtain the rough speech acoustic features through the convolution layer; After the output, it enters the second convolution layer, and after being mapped by a nonlinear activation function, fine speech features are extracted. The convolutional speech features are sent to the pooling layer, and the maximization method is used to reduce the convolution parameter error. The convolutional layer and the pooling layer output the speech feature map and then send it to the fully connected layer, and the corresponding posterior probability is obtained after the convolution operation. The CTC of the softmax layer is used to classify and optimize the speech features extracted by the above layers, and the decoding operation is used to recognize the speech features and then output them.

4. The intelligent voice recognition control system based on the Internet of Things according to claim 3 is characterized in that: The construction of the deep convolutional neural network recognition model includes: The extracted features are input into the model for classification training, and the output result is the final recognition result; the deep neural network is set up with four layers, namely convolution layer, pooling layer, fully connected layer and softmax layer.

5. An intelligent speech recognition control method based on the Internet of Things, applied to an intelligent speech recognition control system based on the Internet of Things as claimed in any one of claims 1 to 4, characterized in that: The following steps are involved: IoT sensing data collection: collect real-time information about the home environment and user voice command information through smart devices and microphone devices of the smart home control system; Formulate a time-series-based IoT sensing data collection format, including time, voice command, device number, device name, device data, device status, and device category; IoT perception data processing: By processing the collected IoT perception data, sensor data standardization and voice data feature extraction are completed; User intention feedback generation: The user intention is predicted by using the collected IoT perception data and the IoT command library. After receiving the wake-up command, the predicted user intention is actively fed back to the user in the form of voice. Execution instruction judgment: If the generated user intention feedback is correct, a device execution instruction is generated, and the instruction execution record and the corresponding IoT perception data are used to generate user control rules and add them to the IoT instruction library; If the user's intention feedback is wrong, the control instruction in the voice command is matched with the voice rule library to find the corresponding user rule. If the match fails, the voice is recognized online; The device execution action after online voice recognition and the corresponding IoT perception data are used to correct the user control rules with incorrect user intention feedback, and the modified user control rules are matched with the voice features and stored in the voice rule library.

6. The intelligent voice recognition control method based on the Internet of Things according to claim 5 is characterized in that: The voice data feature extraction comprises the following steps: Perform preprocessing operations on the collected voice command data, including signal filtering and endpoint detection; The wavelet transform method is used to perform time-frequency domain conversion on the recorded speech signal to obtain the wavelet coefficients of each signal. The wavelet coefficient spectrum is obtained by calculating the wavelet coefficients, and the wavelet coefficients are arranged from low to high and then spliced. The speech feature sequence is established based on the splicing results. Calculate the recorded speech feature sequence, and the calculation result is the power spectrum calculation result; Use filters to perform filtering analysis on speech signals; Compensate for the background noise generated by speech and calculate the long-time frame power of speech; After adjusting the power, the time-frequency and power of the multi-channel speech are normalized. The process of normalizing the speech power is as follows: the average power of the recorded speech sample is estimated, and the estimated result is subtracted from the current power; Performing power function nonlinear processing on the power normalized value, and transforming the speech nonlinear processing signal sequence by discrete cosine transform to obtain speech feature parameters; The first-order and second-order differences of the acquired speech feature parameters are then concatenated with the original values ​​as the final speech features.

Images