Equipment information processing system and method based on multi-source data

Abstract

The invention discloses an equipment information processing system and method based on multi-source data, and relates to the technical field of information analysis processing, and the system comprises an equipment information collection module, a data pre-division module, an equipment information analysis module and a target equipment performance estimation module. Signal feature data, historical identification data and use information received by equipment are acquired through the equipment information acquisition module, and the historical identification data of the equipment are pre-divided into different reference data through the data pre-division module. A device information analysis module establishes a first estimation model and a second estimation model for estimating the sensitivity of a device recognition voice instruction according to reference data generated after pre-division processing operation, and a target device performance estimation module selects an optimal estimation model to estimate the sensitivity of a target recognition voice instruction. And pre-estimation is carried out after the use environment is matched, so that the error of the pre-estimation result is effectively reduced, and the accuracy of the equipment function pre-estimation result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information analysis and processing, and in particular to a device information processing system and method based on multi-source data. Background Art

[0002] As the Internet of Things technology gradually improves, some devices will be equipped with voice control functions to facilitate the control of devices. However, after the device has been used for a certain period of time, the sensitivity of the device's voice recognition function may gradually weaken over time, and the probability of problems such as longer device response time and device malfunction due to voice recognition errors will gradually increase. Therefore, it is particularly important to analyze and estimate the device's voice recognition function in a timely manner, which will help prompt timely improvement of the device's voice recognition function and ensure the device's voice controllability and accurate control;

[0003] However, the voice recognition environments of devices capable of voice control will be different and uncontrollable. Due to the different usage environments of the devices, the voice signals recognized by the devices will have different biases. The voice recognition environments of some devices are more inclined to recognize voice signals with less noise, while the voice recognition environments of some devices are more inclined to recognize voice signals with more noise. In the prior art, the estimated results of the sensitivity of the voice recognition function of the device are generally obtained by referring to the analysis of all data, and the reference data is not pre-divided. Some reference data may not be suitable for the usage environment of the target device. The estimated results may have large errors, which is not conducive to improving the accuracy of the device function estimation results. Summary of the invention

[0004] The object of the present invention is to provide a device information processing system and method based on multi-source data to solve the problems raised in the prior art.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: a device information processing system based on multi-source data, the system comprising a device information acquisition module, a data pre-division module, a device information analysis module and a target device performance estimation module, the device information acquisition module collects signal feature data, historical identification data and usage information received by the device;

[0006] Pre-dividing the historical identification data of the device into different reference data through the data pre-dividing module;

[0007] Establishing, by the device information analysis module, a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing the voice command according to the reference data generated after the pre-division processing operation;

[0008] The target device data is acquired through the target device performance estimation module, and the best estimation model is selected to estimate the sensitivity of the target recognition voice command.

[0009] Preferably, the device information acquisition module includes a signal feature acquisition unit, an identification data acquisition unit and a usage information acquisition unit;

[0010] The signal feature acquisition unit is used to collect feature data of voice signals to be recognized that were previously received by different devices of the same type, wherein the feature data refers to the signal-to-noise ratio of the voice signal, and the device refers to a device that can be controlled by voice. The same type means that the type and model of the device are the same;

[0011] The recognition data collection unit is used to collect historical recognition data of the device, wherein the historical recognition data includes the number of recognition errors of the voice command and the recognition response time of the successful recognition of the voice command, wherein the recognition response time refers to the interval between the time when the voice command is issued and the time when the device receives the voice command and completes the corresponding operation according to the voice command;

[0012] The usage information collection unit is used to collect the usage time information of the device.

[0013] Preferably, the data pre-division module includes an adaptation coefficient analysis unit and a data division processing unit;

[0014] The signal-to-noise ratio separation value of the speech signal is set by the adaptation coefficient analysis unit, and the speech signal with a high signal-to-noise ratio and the speech signal with a low signal-to-noise ratio are separated by the signal-to-noise ratio separation value, and the number of times different devices have previously recognized the speech signal with a high signal-to-noise ratio and the number of times they have recognized the speech signal with a low signal-to-noise ratio are counted, and the adaptation coefficients of the recognition data of different devices as reference data are analyzed according to the statistical results, and the number of times a device has previously recognized the speech signal with a high signal-to-noise ratio and the number of times it has recognized the speech signal with a low signal-to-noise ratio are compared. If the number of times the device recognizes the speech signal with a high signal-to-noise ratio is not less than the number of times it recognizes the speech signal with a low signal-to-noise ratio, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the first adaptation coefficient; if the number of times the device recognizes the speech signal with a high signal-to-noise ratio is less than the number of times it recognizes the speech signal with a low signal-to-noise ratio, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the second adaptation coefficient;

[0015] The first adaptation coefficient threshold and the second adaptation coefficient threshold are set by the data division processing unit. If the first adaptation coefficient exceeds the first adaptation coefficient threshold, the identification data of the corresponding device is divided into the first reference data. If the second adaptation coefficient exceeds the second adaptation coefficient threshold, the identification data of the corresponding device is divided into the second reference data, and the identification data of the remaining undivided devices are not used as reference data.

[0016] Preferably, the device information analysis module includes a partition data analysis unit and an estimation model establishment unit;

[0017] The divided data analysis unit is used for the device whose identification data is divided into the first reference data or the second reference data: dividing the used time of a random device into several equal parts, analyzing the identification data of the random device, and calculating the sensitivity coefficient of the random device in recognizing the voice command at different stages of the used time;

[0018] The estimation model establishment unit is used for devices whose identification data are classified into the first reference data: the usage time and sensitivity coefficient of the device are combined into several groups of target training data, and several first estimation models are established after separate fitting training is performed on the several groups of target training data; for devices whose identification data are classified into the second reference data: the usage time and sensitivity coefficient of the device are combined into several groups of target training data, and several second estimation models are established after separate fitting training is performed on the several groups of target training data.

[0019] Preferably, the target device performance estimation module includes a target information acquisition unit, a reference model selection unit and a sensitivity coefficient estimation unit;

[0020] The target information acquisition unit is used to acquire target device information for which sensitivity coefficient estimation is currently required, including usage time information of the current device and the number of times the current device has previously recognized a voice signal with a high signal-to-noise ratio and the number of times the current device has previously recognized a voice signal with a low signal-to-noise ratio, wherein the current device is of the same type as the device whose recognition data is classified into the reference data;

[0021] The reference model selection unit is used to analyze the current device information: if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is not less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, select the best estimation model for estimating the sensitivity coefficient of the current device recognizing a voice command from a plurality of first estimation models; if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, select the best estimation model for estimating the sensitivity coefficient of the current device recognizing a voice command from a plurality of second estimation models;

[0022] The sensitivity coefficient estimation unit substitutes the usage time of the current device into the optimal estimation model to estimate the sensitivity coefficient of the current device for recognizing voice commands.

[0023] A device information processing method based on multi-source data comprises the following steps:

[0024] S1: collects signal feature data, historical identification data and usage information received by the device;

[0025] S2: pre-dividing the historical identification data of the device into different reference data;

[0026] S3: establishing a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing voice commands according to the reference data generated after the pre-division processing operation;

[0027] S4: Obtain target device data and select the best estimation model to estimate the sensitivity of the target recognition voice command.

[0028] Preferably, S1 includes: collecting a data set of signal-to-noise ratios of speech signals to be recognized that were previously received by different devices of the same type as {SNR1, SNR2, ...SNR k}, k represents the number of items in the signal-to-noise ratio data set, collects historical recognition data of the device, the historical recognition data includes the number of recognition errors of voice commands and the recognition response time of successfully recognized voice commands, and collects the usage time information of the device.

[0029] Preferably, S2 includes: setting the signal-to-noise ratio separation value L of the speech signal, L=[∑ k i＝1 (SNR i )] / k, i represents the i-th item in the signal-to-noise ratio data set. If the signal-to-noise ratio of the speech signal exceeds L, the corresponding speech signal is a speech signal with a high signal-to-noise ratio; otherwise, the corresponding speech signal is a speech signal with a low signal-to-noise ratio. It is counted that the number of times that a total of n devices have previously identified a speech signal with a high signal-to-noise ratio is not less than the number of times that they have identified a speech signal with a low signal-to-noise ratio. The number of times that a random device among the n devices has previously identified a speech signal with a high signal-to-noise ratio and the number of times that a random device has previously identified a speech signal with a low signal-to-noise ratio are a and b respectively. The first adaptation coefficient of the recognition data of the random device as the reference data is W j , W j =(ab) / (a+b), the first adaptation coefficient set of the identification data of n devices as reference data is W={W1, W2, ...W j , ...W n}, set the first adaptation coefficient threshold to W ’ , W ’ is the average value of the first adaptation coefficients in the set W, and the first adaptation coefficients in the set W are compared with W ’ Compare: If the first adaptation coefficient exceeds W ’ , dividing the identification data of the corresponding device into the first reference data;

[0030] It is counted that the number of times that a total of m devices have previously recognized voice signals with a high signal-to-noise ratio is less than the number of times that they have recognized voice signals with a low signal-to-noise ratio. The number of times that a random device among the m devices has previously recognized voice signals with a high signal-to-noise ratio and the number of times that it has recognized voice signals with a low signal-to-noise ratio are c and d respectively. The second adaptation coefficient of the corresponding device's recognition data as reference data is w e , w e =(dc) / (c+d), and the second adaptation coefficient set of the m devices' identification data as reference data is w={w1, w2, ...w e , ...w m}, set the second adaptation coefficient threshold to w ’ , w ’ is the average value of the second adaptation coefficients in set w, and the second adaptation coefficients in set w are compared with w one by one. ’ Compare: If the second adaptation coefficient exceeds w ’ , dividing the identification data of the corresponding device into the second reference data.

[0031] Preferably, S3 includes: counting that there are a total of f devices whose identification data are classified into the first reference data, retrieving the usage time of a random device among the f devices as t, dividing t into r equal parts, and obtaining a usage time set {T1, T2, ...T r}, where T1 = t / r, T2 = 2t / r, T r = t, retrieve that a total of h voice commands were recognized before the usage time of the corresponding device ends at T1, where the number of recognition errors is A, and the recognition response time of successfully recognizing the voice command is G. The sensitivity coefficient of the corresponding device recognizing the voice command before the usage time ends at T1 is U1, U1 = (hA) / h + 1 / G, and the set of sensitivity coefficients of the corresponding device recognizing the voice command at different stages of the usage time is U = {U1, U2, ...U r}, and a random set of target training data for f devices is obtained as {(T1, U1), (T2, U2), ... (T r , U r )}, after performing straight-line fitting training on a random set of target training data, a random first estimation model is established: y=θ1*x+θ2, θ1 and θ2 represent fitting coefficients, and after performing the same analysis on the data of f devices, f sets of target training data for f devices are obtained, and after performing separate fitting training on each set of data in the f sets of target training data, f first estimation models are established;

[0032] It is counted that there are c devices whose identification data are classified into the second reference data. After analyzing the c device data in the same way, c groups of target training data for the c devices are obtained. After performing separate fitting training on each group of data in the c groups of target training data, c second estimation models are established.

[0033] Preferably, the step S4 includes: obtaining that the usage time of the target device for which the sensitivity coefficient estimation is currently required is T ’ , the target device that needs to estimate the sensitivity coefficient has previously identified a speech signal with a high signal-to-noise ratio, which is E, and a speech signal with a low signal-to-noise ratio, which is F. By comparison, E>F is obtained. The best estimation model is selected from the f first estimation models. The difference in the number of historical signal recognitions of the current device is |EF|. The set of historical signal recognition difference values ​​of the f devices is {D1, D2, ...D f}, the matching degree set of the historical signal identification information of the current device and the historical signal identification information of f devices is {1 / ||EF|-D1|, 1 / ||EF|-D2|, ... 1 / ||EF|-D f |}, select the device with the highest matching degree with the historical signal recognition information of the current device as the Jth device, take the first estimation model corresponding to the Jth device as the best estimation model, and obtain the first estimation model corresponding to the Jth device as: y=v1*x+v2, where v1 and v2 represent the fitting coefficients of the first estimation model corresponding to the Jth device, let x=T ’ , the estimated sensitivity coefficient of the current device in recognizing voice commands is v1*(T ’ )+v2.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention collects multi-source historical data, including signal feature data, historical recognition data and usage information received by the device, analyzes the multi-source historical data, analyzes the changes in the sensitivity of voice signal recognition of the same type of devices, establishes an estimation model as a reference model for estimating the sensitivity of voice signal recognition of the current target device, and estimates the sensitivity coefficient of voice command recognition of the target device, which is helpful to timely discover abnormal situations of the voice recognition function of the device and timely improve the voice recognition function of the device, and ensure the voice controllability and accurate control of the device;

[0036] Before analyzing multi-source historical data, the data is pre-divided and processed according to the usage environment of different devices. Different estimation models are established for devices that recognize voice signals with less noise in the usage environment and devices that recognize voice signals with more noise in the usage environment. The usage environment data of the current device is matched with the reference data, and the data of the current device is substituted into the estimation model corresponding to the device with matching usage environment to estimate the sensitivity coefficient. The sensitivity coefficient of the current device for voice recognition in the corresponding usage environment is estimated, which effectively reduces the error of the estimation result and improves the accuracy of the device function estimation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural schematic diagram of a device information processing system based on multi-source data of the present invention;

[0038] Figure 2 The present invention is a flowchart of a method for processing device information based on multi-source data. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Embodiment 1:

[0041] like Figure 1 As shown, this embodiment provides a device information processing system based on multi-source data, and the system includes: a device information acquisition module, a data pre-division module, a device information analysis module and a target device performance estimation module, wherein the device information acquisition module acquires signal feature data, historical recognition data and usage information received by the device; the data pre-division module pre-divides the historical recognition data of the device into different reference data; the device information analysis module establishes a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing voice commands based on the reference data generated after the pre-division processing operation; the target device performance estimation module acquires the target device data, and selects the best estimation model to estimate the sensitivity of the target recognition voice commands.

[0042] The device information collection module includes a signal feature collection unit, an identification data collection unit and a usage information collection unit; the signal feature collection unit is used to collect feature data of voice signals that need to be recognized that have been received in the past by different devices of the same type. The feature data refers to the signal-to-noise ratio of the voice signal. The device refers to a device that can be controlled by voice. The same type means that the type and model of the device are the same; the identification data collection unit is used to collect historical recognition data of the device. The historical recognition data includes the number of recognition errors of the voice command and the recognition response time of the successful recognition of the voice command. The recognition response time refers to the interval between the time when the voice command is issued and the time when the device receives the voice command and completes the corresponding operation according to the voice command; the usage information collection unit is used to collect the usage time information of the device.

[0043] The data pre-division module includes an adaptation coefficient analysis unit and a data division processing unit; the signal-to-noise ratio separation value of the voice signal is set by the adaptation coefficient analysis unit, and the voice signal with a high signal-to-noise ratio and the voice signal with a low signal-to-noise ratio are separated by the signal-to-noise ratio separation value, and the number of times different devices have previously recognized voice signals with a high signal-to-noise ratio and the number of times they have recognized voice signals with a low signal-to-noise ratio are counted, and the recognition data of different devices are analyzed as the adaptation coefficient of reference data based on the statistical results, and the number of times a device has previously recognized voice signals with a high signal-to-noise ratio and the number of times it has recognized voice signals with a low signal-to-noise ratio are compared. If the number of times the device has recognized voice signals with a high signal-to-noise ratio is not less than the number of times it has recognized voice signals with a low signal-to-noise ratio, the adaptation coefficient of the device is used as the reference data. times, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the first adaptation coefficient; if the number of times the device recognizes a voice signal with a high signal-to-noise ratio is less than the number of times it recognizes a voice signal with a low signal-to-noise ratio, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the second adaptation coefficient; the first adaptation coefficient threshold and the second adaptation coefficient threshold are set by the data division processing unit, if the first adaptation coefficient exceeds the first adaptation coefficient threshold, the recognition data of the corresponding device is divided into the first reference data, if the second adaptation coefficient exceeds the second adaptation coefficient threshold, the recognition data of the corresponding device is divided into the second reference data, and the recognition data of the remaining devices that have not been divided are not used as reference data.

[0044] The device information analysis module includes a divided data analysis unit and an estimation model establishment unit; the divided data analysis unit is used for devices whose identification data is divided into the first reference data or the second reference data: the used time of a random device is evenly divided into several equal parts, the identification data of a random device is analyzed, and the sensitivity coefficient of the random device in recognizing voice commands at different stages of the used time is calculated; the estimation model establishment unit is used for devices whose identification data is divided into the first reference data: the used time and the sensitivity coefficient of the device are combined into several groups of target training data, and several first estimation models are established after separate fitting training is performed on the several groups of target training data; for devices whose identification data is divided into the second reference data: the used time and the sensitivity coefficient of the device are combined into several groups of target training data, and several second estimation models are established after separate fitting training is performed on the several groups of target training data.

[0045] The target device performance estimation module includes a target information acquisition unit, a reference model selection unit and a sensitivity coefficient estimation unit; the target device information for which sensitivity coefficient estimation is currently required is acquired through the target information acquisition unit, including the usage time information of the current device and the number of times the current device has previously recognized a voice signal with a high signal-to-noise ratio and the number of times the current device has recognized a voice signal with a low signal-to-noise ratio, and the type of the current device is the same as the type of the device whose recognition data is classified into the reference data; the reference model selection unit is used to analyze the current device information: if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is not less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, the best estimation model for estimating the sensitivity coefficient of the current device for recognizing voice commands is selected from a plurality of first estimation models; if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, the best estimation model for estimating the sensitivity coefficient of the current device for recognizing voice commands is selected from a plurality of second estimation models; the usage time of the current device is substituted into the best estimation model through the sensitivity coefficient estimation unit to estimate the sensitivity coefficient of the current device for recognizing voice commands.

[0046] Embodiment 2:

[0047] like Figure 2 As shown, this embodiment provides a device information processing method based on multi-source data, which is implemented based on the device information processing system in the embodiment, and specifically includes the following steps:

[0048] S1: Collect the signal feature data, historical recognition data and usage information received by the device: The signal-to-noise ratio data set of the speech signals to be recognized that were previously received by different devices of the same type is {SNR1, SNR2, ...SNR k}, k represents the number of items in the signal-to-noise ratio data set, collects historical recognition data of the device, the historical recognition data includes the number of recognition errors of voice commands and the recognition response time of successfully recognized voice commands, and collects the usage time information of the device;

[0049] S2: Pre-divide the historical recognition data of the device into different reference data: Set the signal-to-noise ratio separation value L of the speech signal, L = [∑ k i＝1 (SNR i )] / k, i represents the i-th item in the signal-to-noise ratio data set. If the signal-to-noise ratio of the speech signal exceeds L, the corresponding speech signal is a speech signal with a high signal-to-noise ratio; otherwise, the corresponding speech signal is a speech signal with a low signal-to-noise ratio. It is counted that the number of times that a total of n devices have previously identified a speech signal with a high signal-to-noise ratio is not less than the number of times that they have identified a speech signal with a low signal-to-noise ratio. The number of times that a random device among the n devices has previously identified a speech signal with a high signal-to-noise ratio and the number of times that a random device has previously identified a speech signal with a low signal-to-noise ratio are a and b respectively. The first adaptation coefficient of the recognition data of the random device as the reference data is W j , W j =(ab) / (a+b), the first adaptation coefficient set of the identification data of n devices as reference data is W={W1, W2, ...W j , ...W n}, set the first adaptation coefficient threshold to W ’ , W ’ is the average value of the first adaptation coefficients in the set W, and the first adaptation coefficients in the set W are compared with W ’ Compare: If the first adaptation coefficient exceeds W ’ , dividing the identification data of the corresponding device into the first reference data;

[0050] It is counted that the number of times that a total of m devices have previously recognized voice signals with a high signal-to-noise ratio is less than the number of times that they have recognized voice signals with a low signal-to-noise ratio. The number of times that a random device among the m devices has previously recognized voice signals with a high signal-to-noise ratio and the number of times that it has recognized voice signals with a low signal-to-noise ratio are c and d respectively. The second adaptation coefficient of the corresponding device's recognition data as reference data is w e , w e =(dc) / (c+d), and the second adaptation coefficient set of the m devices' identification data as reference data is w={w1, w2, ...w e , ...w m}, set the second adaptation coefficient threshold to w ’ , w ’ is the average value of the second adaptation coefficients in set w, and the second adaptation coefficients in set w are compared with w one by one. ’Compare: If the second adaptation coefficient exceeds w ’ , dividing the identification data of the corresponding device into the second reference data;

[0051] S3: Establishing a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing voice commands according to the reference data generated after the pre-division processing operation: It is counted that there are a total of f devices whose recognition data are divided into the first reference data, and the usage time of a random device among the f devices is t. T is evenly divided into r equal parts, and the usage time set is {T1, T2, ...T r}, where T1 = t / r, T2 = 2t / r, T r =t, and retrieve that a total of h voice commands have been recognized before the usage time of the corresponding device ends at T1. For example, if the usage time of a random device is 300 days, 300 days are divided into 10 equal parts, and the usage time set obtained is {30, 60, 90, 120, 150, 180, 210, 240, 270, 300}, and T1 is 30 days, then the corresponding device retrieved has recognized a total of h voice commands before 30 days including the 30th day, where the number of recognition errors is A, and the recognition response time for successfully recognizing the voice command is G. The sensitivity coefficient of the corresponding device recognizing the voice command before the usage time ends at T1 is U1, U1=(hA) / h+1 / G, and the sensitivity coefficient set of the corresponding device recognizing the voice command at different stages of the usage time is U={U1, U2, ...U r}, and a random set of target training data for f devices is obtained as {(T1, U1), (T2, U2), ... (T r , U r )}, after performing straight-line fitting training on a random set of target training data, a random first estimation model is established: y=θ1*x+θ2, θ1 and θ2 represent fitting coefficients, and after performing the same analysis on the data of f devices, f sets of target training data for f devices are obtained, and after performing separate fitting training on each set of data in the f sets of target training data, f first estimation models are established;

[0052] It is counted that there are c devices whose identification data are classified into the second reference data, and the c device data are analyzed in the same way to obtain c groups of target training data for the c devices. After performing separate fitting training on each group of data in the c groups of target training data, c second estimation models are established. The second estimation model is established in the same way as the first estimation model.

[0053] S4: Obtain target device data and select the best estimation model to estimate the sensitivity of the target recognition voice command: The usage time of the target device for which the sensitivity coefficient estimation is currently required is obtained as T’ , the target device that needs to estimate the sensitivity coefficient has previously identified a speech signal with a high signal-to-noise ratio, which is E, and a speech signal with a low signal-to-noise ratio, which is F. By comparison, E>F is obtained. The best estimation model is selected from the f first estimation models. The difference in the number of historical signal recognitions of the current device is |EF|. The set of historical signal recognition difference values ​​of the f devices is {D1, D2, ...D f}, the matching degree set of the historical signal identification information of the current device and the historical signal identification information of f devices is {1 / ||EF|-D1|, 1 / ||EF|-D2|, ... 1 / ||EF|-D f |}, select the device with the highest matching degree with the historical signal recognition information of the current device as the Jth device, take the first estimation model corresponding to the Jth device as the best estimation model, and obtain the first estimation model corresponding to the Jth device as: y=v1*x+v2, where v1 and v2 represent the fitting coefficients of the first estimation model corresponding to the Jth device, let x=T ’ , the estimated sensitivity coefficient of the current device in recognizing voice commands is v1*(T ’ )+v2;

[0054] For example, it is found that the number of times the current device recognizes a speech signal with a high signal-to-noise ratio is greater than the number of times it recognizes a speech signal with a low signal-to-noise ratio. The best estimation model is selected from f first estimation models as the first estimation model corresponding to the Jth device. The first estimation model corresponding to the Jth device is obtained as: y = -0.003x + 0.99. It is found that the usage time of the target device for which the sensitivity coefficient estimation is currently required is 100 days. Let x = T ’ =100, the estimated sensitivity coefficient of the current device in recognizing voice commands is v1*(T ’ )+v2=0.69.

[0055] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A device information processing system based on multi-source data, characterized in that: The system includes a device information acquisition module, a data pre-division module, a device information analysis module and a target device performance estimation module; Collecting signal feature data, historical identification data and usage information received by the device through the device information collection module; Pre-dividing the historical identification data of the device into different reference data through the data pre-dividing module; Establishing, by the device information analysis module, a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing the voice command according to the reference data generated after the pre-division processing operation; The target device data is acquired through the target device performance estimation module, and the best estimation model is selected to estimate the sensitivity of the target recognition voice command.

2. The device information processing system based on multi-source data according to claim 1, characterized in that: The device information acquisition module includes a signal feature acquisition unit, an identification data acquisition unit and a usage information acquisition unit; The signal feature acquisition unit is used to collect feature data of voice signals to be recognized that were previously received by different devices of the same type, wherein the feature data refers to the signal-to-noise ratio of the voice signal, and the device refers to a device that can be controlled by voice. The same type means that the type and model of the device are the same; The recognition data collection unit is used to collect historical recognition data of the device, wherein the historical recognition data includes the number of recognition errors of the voice command and the recognition response time of the successful recognition of the voice command, wherein the recognition response time refers to the interval between the time when the voice command is issued and the time when the device receives the voice command and completes the corresponding operation according to the voice command; The usage information collection unit is used to collect the usage time information of the device.

3. The device information processing system based on multi-source data according to claim 2, characterized in that: The data pre-division module includes an adaptation coefficient analysis unit and a data division processing unit; The signal-to-noise ratio separation value of the speech signal is set by the adaptation coefficient analysis unit, and the speech signal with a high signal-to-noise ratio and the speech signal with a low signal-to-noise ratio are separated by the signal-to-noise ratio separation value, and the number of times different devices have previously recognized the speech signal with a high signal-to-noise ratio and the number of times they have recognized the speech signal with a low signal-to-noise ratio are counted, and the adaptation coefficients of the recognition data of different devices as reference data are analyzed according to the statistical results, and the number of times a device has previously recognized the speech signal with a high signal-to-noise ratio and the number of times it has recognized the speech signal with a low signal-to-noise ratio are compared. If the number of times the device recognizes the speech signal with a high signal-to-noise ratio is not less than the number of times it recognizes the speech signal with a low signal-to-noise ratio, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the first adaptation coefficient; if the number of times the device recognizes the speech signal with a high signal-to-noise ratio is less than the number of times it recognizes the speech signal with a low signal-to-noise ratio, the adaptation coefficient of the recognition data of the corresponding device as the reference data is used as the second adaptation coefficient; The first adaptation coefficient threshold and the second adaptation coefficient threshold are set by the data division processing unit. If the first adaptation coefficient exceeds the first adaptation coefficient threshold, the identification data of the corresponding device is divided into the first reference data. If the second adaptation coefficient exceeds the second adaptation coefficient threshold, the identification data of the corresponding device is divided into the second reference data.

4. The device information processing system based on multi-source data according to claim 3, characterized in that: The device information analysis module includes a partition data analysis unit and an estimation model establishment unit; The divided data analysis unit is used for the device whose identification data is divided into the first reference data or the second reference data: dividing the used time of a random device into several equal parts, analyzing the identification data of the random device, and calculating the sensitivity coefficient of the random device in recognizing the voice command at different stages of the used time; The estimation model establishment unit is used for devices whose identification data are classified into the first reference data: the usage time and sensitivity coefficient of the device are combined into several groups of target training data, and several first estimation models are established after separate fitting training is performed on the several groups of target training data; for devices whose identification data are classified into the second reference data: the usage time and sensitivity coefficient of the device are combined into several groups of target training data, and several second estimation models are established after separate fitting training is performed on the several groups of target training data.

5. The device information processing system based on multi-source data according to claim 4, characterized in that: The target device performance estimation module includes a target information acquisition unit, a reference model selection unit and a sensitivity coefficient estimation unit; The target information acquisition unit is used to acquire target device information for which sensitivity coefficient estimation is currently required, including usage time information of the current device and the number of times the current device has previously recognized a voice signal with a high signal-to-noise ratio and the number of times the current device has previously recognized a voice signal with a low signal-to-noise ratio; The reference model selection unit is used to analyze the current device information: if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is not less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, select the best estimation model for estimating the sensitivity coefficient of the current device recognizing a voice command from a plurality of first estimation models; if the number of times the current device recognizes a voice signal with a high signal-to-noise ratio is less than the number of times the current device recognizes a voice signal with a low signal-to-noise ratio, select the best estimation model for estimating the sensitivity coefficient of the current device recognizing a voice command from a plurality of second estimation models; The sensitivity coefficient estimation unit substitutes the usage time of the current device into the optimal estimation model to estimate the sensitivity coefficient of the current device for recognizing voice commands.

6. A device information processing method based on multi-source data, characterized in that: include: S1: collects signal feature data, historical identification data and usage information received by the device; S2: pre-dividing the historical identification data of the device into different reference data; S3: establishing a first estimation model and a second estimation model for estimating the sensitivity of the device in recognizing voice commands according to the reference data generated after the pre-division processing operation; S4: Obtain target device data and select the best estimation model to estimate the sensitivity of the target recognition voice command.

7. The device information processing method based on multi-source data according to claim 6, characterized in that: S1 includes: collecting a data set of signal-to-noise ratios of speech signals to be recognized that were previously received by different devices of the same type, {SNR1, SNR2, ...SNR k }, k represents the number of items in the signal-to-noise ratio data set, the historical recognition data of the acquisition device, the historical recognition data including the number of recognition errors of the voice command and the recognition response time of the successful recognition of the voice command, and the usage time information of the acquisition device.

8. The device information processing method based on multi-source data according to claim 7, characterized in that: The S2 includes: setting the signal-to-noise ratio separation value L of the speech signal, L = [∑ k i＝1 (SNR i )] / k, i represents the i-th item in the signal-to-noise ratio data set. If the signal-to-noise ratio of the speech signal exceeds L, the corresponding speech signal is a speech signal with a high signal-to-noise ratio; otherwise, the corresponding speech signal is a speech signal with a low signal-to-noise ratio. It is counted that the number of times that a total of n devices have previously identified a speech signal with a high signal-to-noise ratio is not less than the number of times that they have identified a speech signal with a low signal-to-noise ratio. The number of times that a random device among the n devices has previously identified a speech signal with a high signal-to-noise ratio and the number of times that a random device has previously identified a speech signal with a low signal-to-noise ratio are a and b respectively. The first adaptation coefficient of the recognition data of the random device as the reference data is W j , W j =(ab) / (a+b), the first adaptation coefficient set of the identification data of n devices as reference data is W={W1, W2, ...W j , ...W n }, set the first adaptation coefficient threshold to W ’ , the first adaptation coefficients in the set W are compared with W ’ Compare: If the first adaptation coefficient exceeds W ’ , dividing the identification data of the corresponding device into the first reference data; It is counted that the number of times that a total of m devices have previously recognized voice signals with a high signal-to-noise ratio is less than the number of times that they have recognized voice signals with a low signal-to-noise ratio. The number of times that a random device among the m devices has previously recognized voice signals with a high signal-to-noise ratio and the number of times that it has recognized voice signals with a low signal-to-noise ratio are c and d respectively. The second adaptation coefficient of the corresponding device's recognition data as reference data is w e , w e =(dc) / (c+d), and the second adaptation coefficient set of the m devices' identification data as reference data is w={w1, w2, ...w e , ...w m }, set the second adaptation coefficient threshold to w ’ , the second adaptation coefficients in set w are compared with w ’ Compare: If the second adaptation coefficient exceeds w ’ , dividing the identification data of the corresponding device into the second reference data.

9. The device information processing method based on multi-source data according to claim 8, characterized in that: S3 includes: counting that there are a total of f devices whose identification data are classified into the first reference data, retrieving the usage time of a random device among the f devices as t, dividing t into r equal parts, and obtaining a usage time set {T1, T2, ...T r }, retrieve that a total of h voice commands were recognized before the corresponding device's usage time ends at T1, where the number of recognition errors is A, and the recognition response time of successfully recognizing a voice command is G. The sensitivity coefficient of the corresponding device recognizing voice commands before the usage time ends at T1 is obtained as U1, U1=(hA) / h+1 / G, and the set of sensitivity coefficients for the corresponding device to recognize voice commands at different stages of the usage time is obtained as U={U1, U2, ...U r }, and a random set of target training data for f devices is obtained as {(T1, U1), (T2, U2), ... (T r , U r )}, after performing straight-line fitting training on a random set of target training data, a random first estimation model is established: y=θ1*x+θ2, θ1 and θ2 represent fitting coefficients, and after performing the same analysis on the data of f devices, f sets of target training data for f devices are obtained, and after performing separate fitting training on each set of data in the f sets of target training data, f first estimation models are established; It is counted that there are c devices whose identification data are classified into the second reference data. After analyzing the c device data in the same way, c groups of target training data for the c devices are obtained. After performing separate fitting training on each group of data in the c groups of target training data, c second estimation models are established.

10. The device information processing method based on multi-source data according to claim 9, characterized in that: S4 includes: obtaining the usage time of the target device for which the sensitivity coefficient estimation is currently required as T ’ , the target device that needs to estimate the sensitivity coefficient has previously identified a speech signal with a high signal-to-noise ratio, which is E, and a speech signal with a low signal-to-noise ratio, which is F. By comparison, E>F is obtained. The best estimation model is selected from the f first estimation models. The difference in the number of historical signal recognitions of the current device is |EF|. The set of historical signal recognition difference values ​​of the f devices is {D1, D2, ...D f }, the matching degree set of the historical signal identification information of the current device and the historical signal identification information of f devices is {1 / ||EF|-D1|, 1 / ||EF|-D2|, ... 1 / ||EF|-D f |}, select the device with the highest matching degree with the historical signal recognition information of the current device as the Jth device, take the first estimation model corresponding to the Jth device as the best estimation model, and obtain the first estimation model corresponding to the Jth device as: y=v1*x+v2, where v1 and v2 represent the fitting coefficients of the first estimation model corresponding to the Jth device, let x=T ’ , the estimated sensitivity coefficient of the current device in recognizing voice commands is v1*(T ’ )+v2.

Images