Internet protocol intelligent all-in-one machine intelligent voice control method and system

Abstract

The present invention relates to the field of speech recognition technology, and provides an Internet protocol smart all-in-one machine intelligent voice control method and system, including: constructing an audio feature spectrum corresponding to a gain audio signal, performing time series analysis on the audio feature spectrum, and obtaining a sequence dependency; performing feature association processing on the audio feature spectrum to obtain an associated feature set, performing text mapping processing on the associated feature set to obtain a signal text sequence; analyzing the signal intent corresponding to the gain audio signal, calculating the intent coupling degree between the signal intents, and formulating a signal instruction corresponding to the gain audio signal; performing adaptation processing on the signal instruction and the control interface to obtain an adaptation control interface, setting a control sequence corresponding to the adaptation control interface, and using the adaptation control interface to execute the control processing of the smart all-in-one machine to obtain a control result. The present invention aims to improve the accuracy of the intelligent voice control of the Internet protocol smart all-in-one machine.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to an intelligent speech control method and system for an Internet protocol all-in-one smart machine. Background Art

[0002] In today's era of rapid development of intelligent technology, the field of intelligent voice control has made significant progress. From the perspective of existing technologies, existing voice control methods are mainly based on simple voice recognition principles, but the performance of existing voice recognition systems in complex environments is not stable enough. For example, in a noisy environment, background noise will interfere with the collection and recognition of voice signals, and the existing technology processes voice signals too simply, and only performs basic noise reduction processing, but cannot effectively distinguish and eliminate noises of different frequencies and intensities in a targeted manner, which leads to a decrease in the accuracy of semantic analysis of signals, and ultimately leads to poor voice control accuracy of smart all-in-one machines. Therefore, a method is needed to improve the accuracy of intelligent voice control of smart all-in-one machines. Summary of the invention

[0003] The present invention provides an Internet Protocol smart all-in-one machine intelligent voice control method and system, the main purpose of which is to improve the accuracy of the Internet Protocol smart all-in-one machine intelligent voice control.

[0004] To achieve the above-mentioned purpose, the present invention provides an Internet protocol intelligent all-in-one machine intelligent voice control method, comprising:

[0005] Using the smart all-in-one machine to collect ambient audio data of the application environment, performing signal purification processing on the ambient audio data to obtain a target audio signal;

[0006] Performing signal gain processing on the target audio signal to obtain a gained audio signal, inputting the gained audio signal into a pre-trained signal acoustic analysis model, using the feature perception layer in the signal acoustic analysis model to construct an audio feature spectrum corresponding to the gained audio signal, and using the recurrent neural network layer in the signal acoustic analysis model to perform time series analysis on the audio feature spectrum to obtain a sequence dependency;

[0007] In combination with the sequence dependency, the fully connected layer in the signal acoustic analysis model is used to perform feature association processing on the audio feature map to obtain an associated feature set, and the output layer in the signal acoustic analysis model is used to perform text mapping processing on the associated feature set to obtain a signal text sequence;

[0008] Based on the signal text sequence, analyzing the signal intention corresponding to the gain audio signal, calculating the intention coupling degree between the signal intentions, performing fuzzy optimization processing on the signal intention based on the intention coupling degree to obtain an optimized signal intention, and formulating a signal instruction corresponding to the gain audio signal based on the signal intention;

[0009] Query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, use the adapted control interface to execute control processing of the smart all-in-one machine to obtain a control result.

[0010] Optionally, the performing signal purification processing on the environmental audio data to obtain a target audio signal includes:

[0011] Cleaning the environmental audio data to obtain target audio data;

[0012] Performing analog-to-digital conversion on the target audio data to obtain an audio digital signal;

[0013] Calculating the signal frequency corresponding to the audio digital signal;

[0014] According to the signal frequency, the audio digital signal is subjected to signal purification processing to obtain a target audio signal.

[0015] Optionally, performing signal gain processing on the target audio signal to obtain a gained audio signal includes:

[0016] Performing discretization processing on the target audio signal to obtain a discrete audio signal;

[0017] Performing signal sampling processing on the discrete audio signal to obtain an audio signal value;

[0018] According to the audio signal value, the gain signal value corresponding to the target audio signal is calculated by the following formula:

[0019] ;

[0020] Where A represents the gain signal value corresponding to the target audio signal, represents the gain adjustment coefficient, represents the gain slope adjustment factor, represents the gain curvature coefficient, Indicates the signal value corresponding to the sampling time point a in the audio signal value, where a represents the sampling time point.

[0021] Based on the gain signal value, signal gain processing is performed on the target audio signal to obtain a gain audio signal.

[0022] Optionally, the constructing an audio feature spectrum corresponding to the gain audio signal by using a feature perception layer in the signal acoustic analysis model includes:

[0023] Extracting multi-dimensional audio features corresponding to the gain audio signal using the perception submodule in the feature perception layer;

[0024] Calculating the feature contribution value of the multi-dimensional audio feature using the feature evaluation function in the feature perception layer;

[0025] Based on the feature contribution value, using the selection function in the feature perception layer to perform feature selection processing on the multi-dimensional audio feature to obtain a target audio feature;

[0026] Using the feature fusion module in the feature perception layer to perform feature fusion on the target audio features to obtain fused audio features;

[0027] An audio feature map of the fused audio features is constructed using a visualization module in the feature perception layer.

[0028] Optionally, calculating the feature contribution value of the multi-dimensional audio feature by using the feature evaluation function in the feature perception layer includes:

[0029] Performing vectorization processing on the multi-dimensional audio features to obtain an audio feature vector;

[0030] Calculating the feature covariance between the multi-dimensional audio features according to the audio feature vector;

[0031] Calculating, based on the feature covariance, a feature stability coefficient corresponding to the multi-dimensional audio feature, and calculating a feature correlation coefficient between the multi-dimensional audio features;

[0032] In combination with the feature stability coefficient, the feature correlation coefficient and the audio feature vector, the feature contribution value of the multi-dimensional audio feature is calculated by the following feature evaluation function:

[0033] ;

[0034] Where D represents the feature contribution value of the multi-dimensional audio feature, Represents the audio feature vector corresponding to the bth feature in the multi-dimensional audio feature, b The serial number corresponding to the multi-dimensional audio feature, p represents the number of features corresponding to the multi-dimensional audio feature, Represents the feature correlation coefficient between the bth feature and the b+1th feature in the multi-dimensional audio feature. Represents the feature stability coefficient of the bth feature in the multi-dimensional audio feature.

[0035] Optionally, the using the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence includes:

[0036] Using the encoding layer in the output layer to encode the associated feature set to obtain an encoded feature set;

[0037] According to a preset feature-label mapping table, the probability function in the output layer is used to calculate the distribution probability corresponding to the encoding feature set;

[0038] Based on the distribution probability and the feature-label mapping table;

[0039] The decoding algorithm in the output layer is used to perform text mapping processing on the encoded feature set to obtain a signal text sequence.

[0040] Optionally, analyzing the signal intention corresponding to the gain audio signal based on the signal text sequence includes:

[0041] Identifying signal text characters in the signal text sequence, and calculating character weights corresponding to the signal text characters;

[0042] extracting key signal characters from the signal text characters based on the character weights;

[0043] Performing semantic analysis on the key signal characters to obtain signal semantics;

[0044] The character context corresponding to the signal text character is analyzed, and based on the character context and the signal semantics, the signal intention corresponding to the gain audio signal is analyzed.

[0045] Optionally, calculating the intention coupling degree between the signal intentions includes:

[0046] Analyze the intent attributes of the signal intent, and count the attribute frequency and joint frequency of each attribute in the intent attributes;

[0047] Constructing an attribute vector corresponding to the intent attribute, and calculating the attribute similarity between the intent attributes based on the attribute vector;

[0048] Combining the attribute frequency, the joint frequency and the attribute similarity, the intention coupling degree between the signal intentions is calculated by the following formula:

[0049] ;

[0050] Among them, H represents the intention coupling degree between signal intentions, It represents the joint frequency of the i-th and i+1-th attributes in the intention attribute. and They represent the attribute frequencies corresponding to the i-th and i+1-th attributes in the intent attributes, respectively. It represents the attribute similarity between the i-th and i+1-th attributes in the intent attribute, and i and i+1 both represent the serial numbers of the intent attribute.

[0051] Optionally, the setting of a control sequence corresponding to the adaptation control interface includes:

[0052] Querying the terminal control device corresponding to the adaptation control interface, and analyzing the device function characteristics corresponding to the terminal control device;

[0053] Based on the functional characteristics of the device, the device priority of the terminal control device is evaluated, and an operation flow chart of the terminal control device is obtained;

[0054] Based on the operation flow chart, analyzing the device operation logic between the terminal control devices;

[0055] Based on the device priority and the device operation logic, a control sequence corresponding to the adaptation control interface is set.

[0056] An Internet Protocol smart all-in-one machine intelligent voice control system, the system comprising:

[0057] A signal purification module is used to collect ambient audio data of the application environment using the smart all-in-one machine, perform signal purification processing on the ambient audio data, and obtain a target audio signal;

[0058] A timing analysis module, used to perform signal gain processing on the target audio signal to obtain a gained audio signal, input the gained audio signal into a pre-trained signal acoustic analysis model, use the feature perception layer in the signal acoustic analysis model to construct an audio feature spectrum corresponding to the gained audio signal, and use the recurrent neural network layer in the signal acoustic analysis model to perform timing analysis on the audio feature spectrum to obtain a sequence dependency;

[0059] A text sequence analysis module, used to combine the sequence dependency, use the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map to obtain an associated feature set, and use the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence;

[0060] A signal instruction formulation module, used for analyzing the signal intention corresponding to the gain audio signal based on the signal text sequence, calculating the intention coupling degree between the signal intentions, performing fuzzy optimization processing on the signal intention based on the intention coupling degree to obtain an optimized signal intention, and formulating a signal instruction corresponding to the gain audio signal based on the signal intention;

[0061] The control processing module is used to query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, the adapted control interface is used to execute control processing of the smart all-in-one machine to obtain a control result.

[0062] The present invention can remove invalid audio signals in the ambient audio data by performing signal purification processing on the ambient audio data, thereby improving the quality of the audio signal and providing a basis for the accuracy of subsequent signal processing results. The present invention can improve the signal strength of the target audio signal by performing signal gain processing on the target audio signal, thereby providing a basis for subsequent signal processing. The present invention combines the sequence dependency and utilizes the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map, so that the features with relationships in the audio feature map can be aggregated together, thereby facilitating subsequent text mapping processing. The present invention analyzes the signal intention corresponding to the gain audio signal based on the signal text sequence, so as to understand the communication purpose corresponding to the gain audio signal, thereby providing a basis for the formulation of subsequent signal instructions. The present invention maps the signal instruction to the control interface for adaptation processing, thereby facilitating the subsequent control processing of the smart all-in-one machine through the corresponding control interface, and sets the control sequence corresponding to the adapted control interface, so that the control interface can give priority to the important instructions of the smart all-in-one machine through the control sequence. Therefore, the embodiment of the present invention provides an Internet Protocol smart all-in-one machine intelligent voice control method and system, which can improve the accuracy of the Internet Protocol smart all-in-one machine intelligent voice control. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flowchart of an intelligent voice control method for an Internet protocol all-in-one smart machine provided by an embodiment of the present invention;

[0064] Figure 2 A functional module diagram of an Internet Protocol smart all-in-one machine intelligent voice control system provided by one embodiment of the present invention.

[0065] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0066] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0067] The embodiment of the present application provides an Internet Protocol smart all-in-one intelligent voice control method. In the embodiment of the present application, the execution subject of the Internet Protocol smart all-in-one intelligent voice control method includes but is not limited to at least one of the electronic devices such as the server, the terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the Internet Protocol smart all-in-one intelligent voice control method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0068] Reference Figure 1 FIG. 1 is a flow chart of an Internet protocol smart all-in-one machine intelligent voice control method provided by an embodiment of the present invention. In this embodiment, the Internet protocol smart all-in-one machine intelligent voice control method includes steps S1-S5.

[0069] S1. Use a smart all-in-one machine to collect ambient audio data of an application environment, perform analog-to-digital conversion on the ambient audio data to obtain an audio digital signal, calculate a signal frequency corresponding to the audio digital signal, and perform signal purification processing on the audio digital signal according to the signal frequency to obtain a target audio signal.

[0070] The present invention can remove invalid audio signals in the environmental audio data by performing signal purification processing on the environmental audio data, thereby improving the quality of the audio signal and ensuring the accuracy of the subsequent signal processing results. The smart all-in-one machine integrates multiple hardware functions into a device body, usually including a high-resolution display screen for clearly displaying images, videos and other information; it has a high-performance processor that can quickly process various complex data and calculations, and also integrates input devices such as cameras and microphones for image and sound acquisition. The environmental audio data is the collective data of all sounds in the environment where the smart all-in-one machine is located, and the target audio signal is the environmental audio data with invalid audio signals in the environmental audio data removed, such as noise and other interference components, retaining valuable audio signals.

[0071] As an embodiment of the present invention, the signal purification processing of the environmental audio data to obtain the target audio signal includes: cleaning the environmental audio data to obtain the target audio data, performing analog-to-digital conversion on the target audio data to obtain an audio digital signal, calculating the signal frequency corresponding to the audio digital signal, and performing signal purification processing on the audio digital signal according to the signal frequency to obtain the target audio signal.

[0072] Among them, the target audio data is the data obtained after removing invalid data in the environmental audio data, the audio digital signal is the digital expression of the target audio data, and the signal frequency is the number of changes of the audio digital signal in unit time.

[0073] Optionally, the cleaning process of the environmental audio data can be achieved through a box plot method; the analog-to-digital conversion of the target audio data can be achieved through an analog-to-digital converter, and the calculation of the signal frequency corresponding to the audio digital signal can be achieved through a fast Fourier transform algorithm; according to the signal frequency, the audio digital signal can be purified through a low-pass filter to obtain a target audio signal, such as removing signals with frequencies lower than 85 Hz and higher than 255 Hz.

[0074] S2. Perform signal gain processing on the target audio signal to obtain a gained audio signal, input the gained audio signal into a pre-trained signal acoustic analysis model, use the feature perception layer in the signal acoustic analysis model to construct an audio feature map corresponding to the gained audio signal, use the recurrent neural network layer in the signal acoustic analysis model to perform time series analysis on the audio feature map to obtain a sequence dependency.

[0075] The present invention can improve the signal strength of the target audio signal by performing signal gain processing on the target audio signal, thereby providing a basis for subsequent signal processing, wherein the gain audio signal is a signal obtained after the target audio signal is amplified.

[0076] As an embodiment of the present invention, performing signal gain processing on the target audio signal to obtain a gain audio signal includes: discretizing the target audio signal to obtain a discrete audio signal, performing signal sampling processing on the discrete audio signal to obtain an audio signal value, and calculating the gain signal value corresponding to the target audio signal according to the audio signal value by the following formula:

[0077] ;

[0078] Where A represents the gain signal value corresponding to the target audio signal, represents the gain adjustment coefficient, represents the gain slope adjustment factor, represents the gain curvature coefficient, Indicates the signal value corresponding to the sampling time point a in the audio signal value, where a represents the sampling time point.

[0079] Based on the gain signal value, signal gain processing is performed on the target audio signal to obtain a gain audio signal.

[0080] The discrete audio signal is a signal obtained by converting the target audio signal into a series of discrete point signals. The audio signal value is a numerical embodiment of the signal amplitude corresponding to the discrete audio signal at each discrete time point. The gain adjustment coefficient represents a key factor in the overall amplification or reduction of the signal. It is usually set according to the demand for signal strength. If you want the signal strength to be twice as strong as before, you can try to set α to 1 (assuming other conditions remain unchanged). You can also determine the appropriate α value through repeated experiments based on the human ear's hearing of the audio signal or the requirements of other signal processing systems for the input signal strength. The gain slope adjustment factor table is: It indicates how fast the gain changes in the initial stage. It can be set according to the characteristics of the signal and the processing requirements. If the signal needs to quickly reach a higher gain from the beginning, β is increased. This can be determined by prior knowledge of the signal, such as the type of signal (burst signal or slowly changing signal), the application scenario of the signal, etc. The gain curvature coefficient represents the degree of curvature of the curve of the gain changing with time, which is determined according to the signal processing goal and the nature of the signal. For example, if the signal gain is expected to stabilize quickly, γ is increased. The value of γ can be adjusted by testing and observing different types of signals and the requirements for the final output effect of the signal processing system.

[0081] Optionally, the discretization processing of the target audio signal can be implemented by a resample function; the signal sampling processing of the discrete audio signal can be implemented by a uniform sampling method, such as the Nyquist-Shannon sampling method; the target audio signal is directly replaced using the gain signal value to implement signal gain processing and obtain a gain audio signal.

[0082] The present invention utilizes the feature perception layer in the signal acoustic analysis model to extract features of the gain audio signal, so as to obtain the acoustic-related features of the gain audio signal, thereby providing a basis for the subsequent time-series analysis of the audio feature spectrum, wherein the signal acoustic analysis model is a model that gradually converts the audio signal into specific signal semantics, the feature perception layer is a neural layer for extracting the signal features of the gain audio signal, and the audio feature spectrum is a visual expression of the audio features of the gain audio signal in the form of a spectrum after feature extraction and organization, which shows the feature distribution of the gain audio signal in different dimensions (such as frequency, time, amplitude, etc.).

[0083] As an embodiment of the present invention, the method of constructing an audio feature map corresponding to the gain audio signal using the feature perception layer in the signal acoustic analysis model includes: extracting multidimensional audio features corresponding to the gain audio signal using the perception submodule in the feature perception layer, calculating feature contribution values ​​of the multidimensional audio features using the feature evaluation function in the feature perception layer, performing feature selection processing on the multidimensional audio features based on the feature contribution values ​​to obtain target audio features, performing feature fusion processing on the target audio features using the feature fusion module in the feature perception layer to obtain fused audio features, and constructing an audio feature map of the fused audio features using the visualization module in the feature perception layer.

[0084] Among them, the perception submodule is a functional unit in the feature perception layer for extracting the multidimensional audio features corresponding to the gain audio signal, the perception submodule is composed of a convolution kernel, the multidimensional audio features are representative features presented by various aspects of the gain audio signal in different dimensions (such as time domain, frequency domain, time-frequency domain, etc.), the feature evaluation function is a function used to calculate the importance of the multidimensional audio features, the feature contribution value represents the value of the importance of the multidimensional audio features, the selection function is a function used in the feature perception layer to select the multidimensional audio features, the selection function is compiled by a programming language, the feature fusion module is a functional module that fuses features in the feature perception layer, and is composed of a linear weighted sum algorithm, the fused audio features are the features obtained after the target audio features are fused, and the visualization module is a functional module used to visualize the fused audio features, and is composed of visualization tools, such as a line chart tool.

[0085] Further, as an optional embodiment of the present invention, the feature contribution value of the multidimensional audio feature is calculated using the feature evaluation function in the feature perception layer, including: vectorizing the multidimensional audio feature to obtain an audio feature vector, calculating the feature covariance between the multidimensional audio features according to the audio feature vector, calculating the feature stability coefficient corresponding to the multidimensional audio feature according to the feature covariance, and calculating the feature correlation coefficient between the multidimensional audio features, combining the feature stability coefficient, the feature correlation coefficient and the audio feature vector, and calculating the feature contribution value of the multidimensional audio feature through the following feature evaluation function:

[0086] ;

[0087] Where D represents the feature contribution value of the multi-dimensional audio feature, represents the audio feature vector corresponding to the bth feature in the multi-dimensional audio feature, b represents the sequence number corresponding to the multi-dimensional audio feature, and p represents the number of features corresponding to the multi-dimensional audio feature. Represents the feature correlation coefficient between the bth feature and the b+1th feature in the multi-dimensional audio feature. Represents the feature stability coefficient of the bth feature in the multi-dimensional audio feature.

[0088] Among them, the audio feature vector is the expression vector corresponding to the multidimensional audio feature, the feature covariance represents the change trend between the multidimensional audio features, the feature stability coefficient represents the stability degree corresponding to the multidimensional audio feature, and the feature correlation coefficient represents the correlation degree between the multidimensional audio features.

[0089] Furthermore, the vectorization processing of the multidimensional audio features can be achieved through a Mel-frequency cepstral coefficient (MFCC) algorithm; the calculation of the feature covariance between the multidimensional audio features can be achieved through a covariance formula; the feature stability coefficient corresponding to the multidimensional audio features can be obtained by 1-the feature covariance; the calculation of the feature correlation coefficient between the multidimensional audio features can be achieved through a Pearson correlation coefficient algorithm.

[0090] The present invention uses the recurrent neural network layer in the signal acoustic analysis model to perform time series analysis on the audio feature map, thereby exploring the regularity and intrinsic correlation pattern of the audio feature map in the time series, and providing a basis for subsequent feature association processing, wherein the recurrent neural network layer is a neural network structure for capturing the time series features and long-term dependencies in the audio feature map, and the sequence dependency is the association regularity of the audio feature map. Furthermore, the audio feature map can be time-series analyzed by the gate structure in the long short-term memory network in the recurrent neural network layer to obtain the sequence dependency, and the gate structure is composed of a forget gate, an input gate and an output gate.

[0091] S3. In combination with the sequence dependency, the fully connected layer in the signal acoustic analysis model is used to perform feature association processing on the audio feature map to obtain an associated feature set, and the output layer in the signal acoustic analysis model is used to perform text mapping processing on the associated feature set to obtain a signal text sequence.

[0092] The present invention combines the sequence dependency and utilizes the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map, so that the features with relationship in the audio feature map can be clustered together, thereby facilitating subsequent text mapping processing, wherein the associated feature set is the feature obtained after the associated features in the audio feature map are clustered together, and further, the features of the audio feature map with sequence dependency are associated together using the activation function in the fully connected layer to obtain the associated feature set, and the activation function includes the ReLU (Rectified Linear Unit) activation function.

[0093] The present invention uses the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set, so as to obtain the text expression information corresponding to the associated feature set, thereby facilitating the subsequent analysis and processing of the signal intention, wherein the signal text sequence is the expression and presentation of the audio features in text form obtained by converting the associated feature set based on certain mapping rules.

[0094] As an embodiment of the present invention, the method of using the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence includes: using the encoding layer in the output layer to encode the associated feature set to obtain an encoded feature set, according to a preset feature-label mapping table, using the probability function in the output layer to calculate the distribution probability corresponding to the encoded feature set, based on the distribution probability and the feature-label mapping table, using the decoding algorithm in the output layer to perform text mapping processing on the encoded feature set to obtain a signal text sequence.

[0095] Among them, the encoding layer is a neural layer that reduces the high-dimensional features in the associated feature set to low-dimensional features, and is composed of a linear encoding function. The encoding feature set is the low-dimensional features corresponding to the high-dimensional features in the associated feature set. The feature-label mapping table is a table of pre-constructed features and corresponding feature labels, which is constructed by counting a large number of features and feature labels. The probability function is a function used to calculate the probability of the encoding feature set appearing in the feature-label mapping table, such as a naive Bayes probability function. The decoding algorithm is an algorithm that finds the corresponding text for the features in the encoding feature set from the feature-label mapping table and outputs it, such as a Viterbi algorithm.

[0096] S4. Based on the signal text sequence, the signal intention corresponding to the gain audio signal is analyzed, the intention coupling degree between the signal intentions is calculated, and based on the intention coupling degree, the signal intention is fuzzy optimized to obtain the optimized signal intention, and based on the signal intention, the signal instruction corresponding to the gain audio signal is formulated.

[0097] The present invention analyzes the signal intention corresponding to the gain audio signal based on the signal text sequence, so as to understand the communication purpose corresponding to the gain audio signal and provide a basis for the formulation of subsequent signal instructions, wherein the signal intention indicates the signal communication purpose corresponding to the gain audio signal.

[0098] As an embodiment of the present invention, the analyzing the signal intention corresponding to the gain audio signal based on the signal text sequence includes: identifying the signal text characters in the signal text sequence, calculating the character weights corresponding to the signal text characters, extracting key signal characters from the signal text characters based on the character weights, performing semantic analysis on the key signal characters to obtain signal semantics, analyzing the character context corresponding to the signal text characters, and analyzing the signal intention corresponding to the gain audio signal based on the character context and the signal semantics.

[0099] Among them, the signal text character is the text information in the signal text sequence, the character weight represents the importance corresponding to the signal text character, the key signal character is a representative character in the signal text characters, the signal semantics is the meaning corresponding to the key signal character, and the character context is the information environment corresponding to the signal text character.

[0100] Furthermore, the recognition of the signal text characters in the signal text sequence can be achieved through OCR recognition technology; the calculation of the character weight corresponding to the signal text character can be achieved through the TF-IDF algorithm; when the character weight is greater than the preset weight, the key signal character is extracted from the signal text character, and the preset weight can be set to 0.8, or it can be set according to the actual application scenario; the semantic analysis of the key signal character can be achieved through semantic analysis; the character context corresponding to the signal text character can be obtained by combining the context analysis of the text; based on the character context and the signal semantics, the signal intention corresponding to the gain audio signal is analyzed. For example, if the signal semantics has "play" and the character context mentions "movie", it can be analyzed that the signal intention is to play the movie on the all-in-one machine.

[0101] By calculating the intention coupling degree between the signal intentions, the present invention can understand the similarity between the signal intentions, thereby facilitating the subsequent fuzzy optimization processing of the signal intentions, wherein the intention coupling degree represents the similarity between the signal intentions.

[0102] As an embodiment of the present invention, the calculation of the intention coupling degree between the signal intentions includes: analyzing the intention attributes of the signal intention, counting the attribute frequency and joint frequency of each attribute in the intention attributes, constructing an attribute vector corresponding to the intention attribute, and calculating the attribute similarity between the intention attributes based on the attribute vector. Combining the attribute frequency, the joint frequency and the attribute similarity, the intention coupling degree between the signal intentions is calculated by the following formula:

[0103] ;

[0104] Among them, H represents the intention coupling degree between signal intentions, It represents the joint frequency of the i-th and i+1-th attributes in the intention attribute. and They represent the attribute frequencies corresponding to the i-th and i+1-th attributes in the intent attributes, respectively. It represents the attribute similarity between the i-th and i+1-th attributes in the intent attribute, and i and i+1 both represent the serial numbers of the intent attribute.

[0105] Among them, the intention attribute is the intention feature corresponding to the signal intention, the attribute frequency is the frequency of each attribute in the intention attribute appearing alone, the joint frequency is the probability of different attributes in the intention attribute appearing at the same time, and the attribute similarity represents the degree of similarity between the intention attributes.

[0106] Optionally, the analysis of the intent attributes of the signal intent can be achieved through principal component analysis; the statistics of the attribute frequency and joint frequency of each attribute in the intent attributes can be achieved through direct counting method; the construction of the attribute vector corresponding to the intent attribute can be achieved through the bag-of-words model; the attribute similarity between the intent attributes can be achieved through the cosine similarity algorithm.

[0107] The present invention can improve the clarity of the signal intention by performing fuzzy optimization processing on the signal intention based on the intention coupling degree, formulate the signal instruction corresponding to the gain audio signal based on the signal intention, and then obtain the system operation instruction of the smart all-in-one machine, so as to facilitate the subsequent control processing of the smart all-in-one machine, wherein the optimized signal intention is a clear intention obtained after the intention with uncertainty in the signal intention is optimized, and the signal instruction is the specific operation execution content corresponding to the gain audio signal. Furthermore, the fuzzy optimization processing of the signal intention can be achieved through the fuzzy logic method. For example, in an intelligent temperature control system, the user gives the signal intention of "feeling a little hot indoors", and the system outputs the optimized signal intention of "lowering the temperature by 2-3 degrees Celsius" through fuzzy reasoning according to the fuzzy input of "a little hot" and the fuzzy rule of "temperature adjustment"; the formulation of the signal instruction corresponding to the gain audio signal can be achieved through the template generation method, and the signal intention is input into the preset instruction generation template to obtain the signal control instruction.

[0108] S5. Query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, use the adapted control interface to execute control processing of the smart all-in-one machine to obtain a control result.

[0109] The present invention maps the signal instruction to the control interface for adaptation processing, thereby facilitating the subsequent control processing of the smart all-in-one machine through the corresponding control interface, and sets a control sequence corresponding to the adapted control interface. The control sequence can be used to enable the control interface to give priority to important instructions of the smart all-in-one machine, wherein the control interface is an interface in the smart all-in-one machine for transmitting software instructions and hardware devices, the adapted control interface is an interface obtained by matching the signal instruction with the corresponding control interface, and the control sequence indicates the processing priority corresponding to the adapted control interface. Furthermore, the adaptation processing of the signal instruction and the control interface can be implemented through a driver.

[0110] As an embodiment of the present invention, the setting of the control sequence corresponding to the adaptation control interface includes: querying the terminal control device corresponding to the adaptation control interface, analyzing the device functional characteristics corresponding to the terminal control device, evaluating the device priority of the terminal control device based on the device functional characteristics, obtaining an operation flow chart of the terminal control device, analyzing the device operation logic between the terminal control devices based on the operation flow chart, and setting the control sequence corresponding to the adaptation control interface based on the device priority and the device operation logic.

[0111] Among them, the terminal control device is the execution device corresponding to the adaptation control interface, the device functional characteristics are the functional properties corresponding to the terminal control device, the device priority indicates the importance corresponding to the terminal control device, the operation flow chart is the overall operation chart of the terminal control device, and the device operation logic is the operation sequence between the terminal control devices.

[0112] Optionally, the query of the terminal control device corresponding to the adaptation control interface can be implemented through device management software; the device functional characteristics corresponding to the terminal control device can be analyzed and obtained through the device technical manual; the functional core degree of the device functional characteristics can be evaluated, and the device priority of the terminal control device can be evaluated according to the functional core degree; the operation flow chart of the terminal control device is obtained, and the device operation logic between the terminal control devices is determined based on the order in the operation flow chart; based on the device priority and the device operation logic, the control sequence corresponding to the adaptation control interface is set, and according to the device operation logic, the logical relationship of the adaptation control interface is determined, such as parallel, and the control sequence corresponding to the adaptation control interface is set according to the logical relationship and the device priority.

[0113] The present invention can remove invalid audio signals in the environmental audio data by performing signal purification processing on the environmental audio data, thereby improving the quality of the audio signal, and providing a basis for the accuracy of the subsequent signal processing results. The present invention can improve the signal strength of the target audio signal by performing signal gain processing on the target audio signal, thereby providing a basis for subsequent signal processing. The present invention combines the sequence dependency and uses the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map, so that the features with relationships in the audio feature map can be gathered together, thereby facilitating subsequent text mapping processing. The present invention analyzes the signal intention corresponding to the gain audio signal based on the signal text sequence, and can understand the communication purpose corresponding to the gain audio signal, thereby providing a basis for the formulation of subsequent signal instructions. The present invention maps the signal instruction to the control interface for adaptation processing, thereby facilitating the subsequent control processing of the intelligent all-in-one machine through the corresponding control interface, and sets the control sequence corresponding to the adapted control interface, so that the control interface can give priority to the important instructions of the intelligent all-in-one machine through the control sequence. Therefore, an intelligent voice control method for an Internet protocol intelligent all-in-one machine provided in an embodiment of the present invention can improve the accuracy of intelligent voice control of an Internet protocol intelligent all-in-one machine.

[0114] like Figure 2 , which is a functional module diagram of an Internet protocol smart all-in-one intelligent voice control system provided by one embodiment of the present invention.

[0115] The Internet protocol smart all-in-one machine intelligent voice control system 100 described in the present invention can be installed in an electronic device. According to the functions implemented, the Internet protocol smart all-in-one machine intelligent voice control system 100 can include a control processing module 101, a signal instruction formulation module 102, a text sequence analysis module 103, a timing analysis module 104 and a signal purification module 105. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0116] In this embodiment, the functions of each module / unit are as follows:

[0117] The signal purification module 101 is used to collect ambient audio data of the application environment using the smart all-in-one machine, perform signal purification processing on the ambient audio data, and obtain a target audio signal;

[0118] The timing analysis module 102 is used to perform signal gain processing on the target audio signal to obtain a gained audio signal, input the gained audio signal into a pre-trained signal acoustic analysis model, use the feature perception layer in the signal acoustic analysis model to construct an audio feature spectrum corresponding to the gained audio signal, and use the recurrent neural network layer in the signal acoustic analysis model to perform timing analysis on the audio feature spectrum to obtain a sequence dependency relationship;

[0119] The text sequence analysis module 103 is used to combine the sequence dependency, use the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map to obtain an associated feature set, and use the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence;

[0120] The signal instruction formulation module 104 is used to analyze the signal intention corresponding to the gain audio signal based on the signal text sequence, calculate the intention coupling degree between the signal intentions, perform fuzzy optimization processing on the signal intention based on the intention coupling degree to obtain the optimized signal intention, and formulate the signal instruction corresponding to the gain audio signal based on the signal intention;

[0121] The control processing module 105 is used to query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, the adapted control interface is used to execute control processing of the smart all-in-one machine to obtain a control result.

[0122] In detail, each module described in an Internet protocol smart all-in-one machine intelligent voice control system 100 described in the embodiment of the present application is used in the same manner as described above. Figure 1 The same technical means as the Internet Protocol smart all-in-one machine intelligent voice control method described in the text and can produce the same technical effects will not be repeated here.

[0123] In the several embodiments provided by the present invention, it should be understood that the provided methods and systems can be implemented in other ways. For example, the method embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An Internet Protocol smart all-in-one machine intelligent voice control method, characterized in that: The method comprises: Using the smart all-in-one machine to collect ambient audio data of the application environment, performing signal purification processing on the ambient audio data to obtain a target audio signal; Performing signal gain processing on the target audio signal to obtain a gained audio signal, inputting the gained audio signal into a pre-trained signal acoustic analysis model, using the feature perception layer in the signal acoustic analysis model to construct an audio feature spectrum corresponding to the gained audio signal, and using the recurrent neural network layer in the signal acoustic analysis model to perform time series analysis on the audio feature spectrum to obtain a sequence dependency; In combination with the sequence dependency, the fully connected layer in the signal acoustic analysis model is used to perform feature association processing on the audio feature map to obtain an associated feature set, and the output layer in the signal acoustic analysis model is used to perform text mapping processing on the associated feature set to obtain a signal text sequence; Based on the signal text sequence, analyzing the signal intention corresponding to the gain audio signal, calculating the intention coupling degree between the signal intentions, performing fuzzy optimization processing on the signal intention based on the intention coupling degree to obtain an optimized signal intention, and formulating a signal instruction corresponding to the gain audio signal based on the signal intention; Query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, use the adapted control interface to execute control processing of the smart all-in-one machine to obtain a control result.

2. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The performing signal purification processing on the environmental audio data to obtain a target audio signal includes: Cleaning the environmental audio data to obtain target audio data; Performing analog-to-digital conversion on the target audio data to obtain an audio digital signal; Calculating the signal frequency corresponding to the audio digital signal; According to the signal frequency, the audio digital signal is subjected to signal purification processing to obtain a target audio signal.

3. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The performing signal gain processing on the target audio signal to obtain a gained audio signal includes: Performing discretization processing on the target audio signal to obtain a discrete audio signal; Performing signal sampling processing on the discrete audio signal to obtain an audio signal value; According to the audio signal value, the gain signal value corresponding to the target audio signal is calculated by the following formula: ； Where A represents the gain signal value corresponding to the target audio signal, represents the gain adjustment coefficient, represents the gain slope adjustment factor, represents the gain curvature coefficient, Indicates the signal value corresponding to the sampling time point a in the audio signal value, where a represents the sampling time point. Based on the gain signal value, signal gain processing is performed on the target audio signal to obtain a gain audio signal.

4. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The step of constructing an audio feature spectrum corresponding to the gain audio signal by using the feature perception layer in the signal acoustic analysis model includes: Extracting multi-dimensional audio features corresponding to the gain audio signal using the perception submodule in the feature perception layer; Calculating the feature contribution value of the multi-dimensional audio feature using the feature evaluation function in the feature perception layer; Based on the feature contribution value, using the selection function in the feature perception layer to perform feature selection processing on the multi-dimensional audio feature to obtain a target audio feature; Using the feature fusion module in the feature perception layer to perform feature fusion on the target audio features to obtain fused audio features; An audio feature map of the fused audio features is constructed using a visualization module in the feature perception layer.

5. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 4, characterized in that: The calculating the feature contribution value of the multi-dimensional audio feature by using the feature evaluation function in the feature perception layer includes: Performing vectorization processing on the multi-dimensional audio features to obtain an audio feature vector; Calculating the feature covariance between the multi-dimensional audio features according to the audio feature vector; Calculating, based on the feature covariance, a feature stability coefficient corresponding to the multi-dimensional audio feature, and calculating a feature correlation coefficient between the multi-dimensional audio features; In combination with the feature stability coefficient, the feature correlation coefficient and the audio feature vector, the feature contribution value of the multi-dimensional audio feature is calculated by the following feature evaluation function: ； Where D represents the feature contribution value of the multi-dimensional audio feature, Represents the audio feature vector corresponding to the bth feature in the multi-dimensional audio feature, b The serial number corresponding to the multi-dimensional audio feature, p represents the number of features corresponding to the multi-dimensional audio feature, Represents the feature correlation coefficient between the bth feature and the b+1th feature in the multi-dimensional audio feature. Represents the feature stability coefficient of the bth feature in the multi-dimensional audio feature.

6. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The step of using the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence includes: Using the encoding layer in the output layer to encode the associated feature set to obtain an encoded feature set; According to a preset feature-label mapping table, the probability function in the output layer is used to calculate the distribution probability corresponding to the encoding feature set; Based on the distribution probability and the feature-label mapping table; The decoding algorithm in the output layer is used to perform text mapping processing on the encoded feature set to obtain a signal text sequence.

7. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The analyzing the signal intention corresponding to the gain audio signal based on the signal text sequence includes: Identifying signal text characters in the signal text sequence, and calculating character weights corresponding to the signal text characters; extracting key signal characters from the signal text characters based on the character weights; Performing semantic analysis on the key signal characters to obtain signal semantics; The character context corresponding to the signal text character is analyzed, and based on the character context and the signal semantics, the signal intention corresponding to the gain audio signal is analyzed.

8. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The calculating the intention coupling degree between the signal intentions includes: Analyze the intent attributes of the signal intent, and count the attribute frequency and joint frequency of each attribute in the intent attributes; Constructing an attribute vector corresponding to the intent attribute, and calculating the attribute similarity between the intent attributes based on the attribute vector; Combining the attribute frequency, the joint frequency and the attribute similarity, the intention coupling degree between the signal intentions is calculated by the following formula: ； Among them, H represents the intention coupling degree between signal intentions, It represents the joint frequency of the i-th and i+1-th attributes in the intention attribute. and They represent the attribute frequencies corresponding to the i-th and i+1-th attributes in the intent attributes, respectively. It represents the attribute similarity between the i-th and i+1-th attributes in the intent attribute, and i and i+1 both represent the serial numbers of the intent attribute.

9. The method for controlling an Internet protocol all-in-one intelligent machine by intelligent voice according to claim 1, characterized in that: The step of setting the control sequence corresponding to the adaptation control interface includes: Querying the terminal control device corresponding to the adaptation control interface, and analyzing the device function characteristics corresponding to the terminal control device; Based on the functional characteristics of the device, the device priority of the terminal control device is evaluated, and an operation flow chart of the terminal control device is obtained; Based on the operation flow chart, analyzing the device operation logic between the terminal control devices; Based on the device priority and the device operation logic, a control sequence corresponding to the adaptation control interface is set.

10. An Internet Protocol smart all-in-one intelligent voice control system, characterized in that: The system comprises: A signal purification module is used to collect ambient audio data of the application environment using the smart all-in-one machine, perform signal purification processing on the ambient audio data, and obtain a target audio signal; A timing analysis module, used to perform signal gain processing on the target audio signal to obtain a gained audio signal, input the gained audio signal into a pre-trained signal acoustic analysis model, use the feature perception layer in the signal acoustic analysis model to construct an audio feature spectrum corresponding to the gained audio signal, and use the recurrent neural network layer in the signal acoustic analysis model to perform timing analysis on the audio feature spectrum to obtain a sequence dependency; A text sequence analysis module, used to combine the sequence dependency, use the fully connected layer in the signal acoustic analysis model to perform feature association processing on the audio feature map to obtain an associated feature set, and use the output layer in the signal acoustic analysis model to perform text mapping processing on the associated feature set to obtain a signal text sequence; A signal instruction formulation module, used for analyzing the signal intention corresponding to the gain audio signal based on the signal text sequence, calculating the intention coupling degree between the signal intentions, performing fuzzy optimization processing on the signal intention based on the intention coupling degree to obtain an optimized signal intention, and formulating a signal instruction corresponding to the gain audio signal based on the signal intention; The control processing module is used to query the control interface corresponding to the smart all-in-one machine, adapt the signal instruction to the control interface to obtain an adapted control interface, and set a control sequence corresponding to the adapted control interface. Based on the control sequence, the adapted control interface is used to execute control processing of the smart all-in-one machine to obtain a control result.

Images