Voice error correction method and system based on edge computing power

Abstract

The invention provides a voice error correction method and system based on edge computing power, and the method comprises the steps: obtaining voice data and environment data, carrying out the data alignment of the voice data and the environment data, and forming a voice data sample with an environment label; constructing a relational graph based on a graph neural network according to the voice data sample with the environment label; constructing a voice error correction model of fused graph information, inputting the relation graph into the voice error correction model for training to obtain a target model, deploying the target model to edge equipment, and realizing real-time reasoning by using edge computing power; when the edge device receives new voice data and environment data, a corresponding relation graph is constructed and input into a target model, a corrected text is generated and output, the target model judges the current environment state according to graph characteristics and dynamically adjusts an error correction strategy, and specifically, through deep cooperation of the voice and the environment data, the error correction accuracy is improved. And the accuracy and robustness of edge end voice error correction are effectively enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of speech error correction, and specifically relates to a speech error correction method and system based on edge computing power. Background Art

[0002] Amid the rapid growth of industrial automation and the Internet of Things (IoT), voice interaction technology has gained widespread adoption in industrial workshops due to its convenience and efficiency. Operators can now control equipment and record inspection information through voice commands, significantly improving work efficiency.

[0003] It should be noted that traditional speech recognition technology already has a high recognition accuracy, but in the application scenarios of industrial workshops, it often lacks effective consideration of environmental factors. That is, when in a noisy industrial workshop, the roar of machines, the noise of equipment operation, etc. will seriously interfere with the voice signal, resulting in a significant increase in the speech recognition error rate. Summary of the Invention

[0004] Based on this, an embodiment of the present invention provides a speech error correction method and system based on edge computing power, which aims to combine environmental data to accurately implement speech error correction in complex environments to meet the strict requirements for voice interaction reliability and accuracy in industrial scenarios.

[0005] A first aspect of an embodiment of the present invention provides a speech error correction method based on edge computing power, which is applied to an industrial workshop scenario where edge devices are installed. The method includes: Acquire voice data and environmental data, and align the voice data with the environmental data to form a voice data sample with an environmental label; Based on speech data samples with environmental labels, a relationship graph based on graph neural network is constructed; Constructing a speech error correction model that integrates graph information, inputting the relationship graph into the speech error correction model for training to obtain a target model, wherein the target model is deployed to the edge device to implement real-time inference using edge computing power; When the edge device receives new voice data and environmental data, it constructs a corresponding relationship graph and inputs it into the target model to generate a corrected text output. The target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features.

[0006] Furthermore, the step of acquiring voice data and environmental data, and aligning the voice data and the environmental data to form a voice data sample with an environmental label includes: Acquire speech data, use beamforming technology to capture the human voice in a directionally controlled manner, and use short-time Fourier transform to convert the time-domain speech signal into a frequency-domain Mel-spectrogram; Extracting acoustic features according to the frequency domain Mel-spectrogram; Acquire environmental data including at least temperature, humidity, ambient noise decibel value, device speed, and load, and normalize the environmental data; A timestamp alignment algorithm is used to match the acoustic features of the speech frame with the normalized environmental data to form a speech data sample with an environmental label.

[0007] Furthermore, in the step of constructing a relationship graph based on a graph neural network based on the speech data sample with the environment label, the relationship graph is composed of nodes and edges, the nodes include speech nodes and environment nodes, the acoustic features under the speech frame are used as node attributes, and each speech frame corresponds to a speech node; Each environmental parameter is treated as a separate node, and the collection timestamp of the corresponding environmental parameter is given as an attribute; Each speech node establishes undirected edges with all environmental nodes at the same moment, and the weights are dynamically adjusted based on the degree of impact of environmental factors on speech; Between environment nodes, edge connections are established for preset environment parameters, and edge weights are calculated using the Pearson correlation coefficient to characterize the correlation between parameters. In addition, the relationship graph is encoded into a graph structure and a graph embedding algorithm is used to encode the graph structure into a low-dimensional vector representation, preserving the topological relationship between speech and environmental factors.

[0008] Furthermore, the speech error correction model that integrates graph information includes a graph feature extraction layer and an attention fusion layer, wherein the graph feature extraction layer uses GNN to extract features from the relationship graph and outputs high-order semantic features of each node. The attention fusion layer uses a multi-head attention mechanism to fuse the acoustic features of the speech frame with the environmental graph features of the corresponding node to generate a speech representation that includes environmental perception. In addition, an adaptive weight function is designed to dynamically adjust the weight of the anti-noise module according to the characteristic values ​​of the environmental nodes.

[0009] Furthermore, in the step of inputting the relationship graph into the speech error correction model for training to obtain the target model, the cross entropy loss of speech error correction and the Laplace regularization loss of the graph structure are combined to ensure that the model maintains the semantic consistency of the graph structure while correcting speech errors. The loss function expression is: ; in, is the cross entropy loss of speech error correction, is the graph structure loss, is the balance coefficient.

[0010] Furthermore, the step of dynamically adjusting the error correction strategy by the target model based on the graph features to determine the current environment state includes: When the target model determines that the current environment state is a preset noise environment based on the graph features, the anti-noise module will be triggered. Specifically, a noise reduction algorithm based on deep learning is used to process the noisy speech features, and then the noise-reduced speech features are input into the target model to generate a corrected text output; When the target model determines that the current environmental state is the preset operating state of the device based on the graph features, for voice commands involving device operations, the device operating parameters in the environmental data are used to perform semantic verification on the commands; The risk of instruction execution is assessed through a pre-set rule base and machine learning models.

[0011] A second aspect of an embodiment of the present invention provides a speech error correction system based on edge computing power, which is used to implement the speech error correction method based on edge computing power provided in the first aspect of the embodiment of the present invention. The system includes: An acquisition module is used to acquire voice data and environmental data, and align the voice data and the environmental data to form a voice data sample with an environmental label; A construction module for constructing a graph neural network-based relationship graph based on speech data samples with environmental labels; A training module is configured to construct a speech error correction model integrating graph information, input the relationship graph into the speech error correction model for training, and obtain a target model, wherein the target model is deployed to the edge device to implement real-time inference using edge computing power; The input module is used to construct a corresponding relationship graph and input the target model when the edge device receives new voice data and environmental data, generating a corrected text output, wherein the target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features.

[0012] A third aspect of an embodiment of the present invention provides a computer-readable storage medium, including: The readable storage medium stores one or more programs, which, when executed by the processor, implement the speech error correction method based on edge computing power as described in the first aspect.

[0013] A fourth aspect of an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, it implements the speech error correction method based on edge computing power as described in the first aspect.

[0014] A speech error correction method and system based on edge computing power is provided in an embodiment of the present invention. The method and system obtain speech data and environmental data, and align the speech data and environmental data to form a speech data sample with an environmental label; construct a relationship graph based on a graph neural network based on the speech data sample with the environmental label; construct a speech error correction model that integrates graph information, and input the relationship graph into the speech error correction model for training to obtain a target model, wherein the target model is deployed to the edge device, and real-time reasoning is achieved by using edge computing power; when the edge device receives new speech data and environmental data, it constructs a corresponding relationship graph and inputs it into the target model to generate a corrected text output, wherein the target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features. Specifically, through the deep collaboration of speech and environmental data, the accuracy and robustness of edge speech error correction are effectively enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of an implementation method for speech error correction based on edge computing power provided in Example 1 of the present invention; Figure 2 This is a block diagram of a speech error correction system based on edge computing power provided in Example 2 of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0016] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0017] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0019] Example 1 The first embodiment of the present invention provides a voice error correction method based on edge computing power, which is applied to industrial workshop scenarios. Edge devices are installed in the industrial workshop, where edge devices refer to smart hardware devices located at the edge layer of the network, close to the data collection source or user terminal. They have certain local computing, storage and communication capabilities, and can process data directly at the "edge" end close to the data source without uploading all data to the cloud, thereby reducing latency, saving bandwidth, and maintaining functional availability in offline or weak network environments. For example, edge devices include smart sensors (such as microphone arrays, environmental sensors), smart cameras (with integrated voice collection functions), etc., please refer to Figure 1 , which is a flow chart for implementing a speech error correction method based on edge computing power, specifically including steps S01 to S04.

[0020] Step S01: Acquire voice data and environmental data, and align the voice data and the environmental data to form a voice data sample with an environmental label.

[0021] Specifically, we acquire voice data, use beamforming technology to capture human voice in a directionally accurate manner, and use short-time Fourier transform to convert the time-domain voice signal into a frequency-domain Mel-spectrogram. The formula is: ; in, is the time domain speech signal, is the Hamming window function, m is the frame index, k is the frequency index, and T is the frame length; According to the frequency domain Mel-frequency spectrogram, acoustic features are extracted. The acoustic features may be MFCC (Mel-frequency Cepstral Coefficients), PLP (Perceptual Linear Prediction Coefficients), etc. The Mel-frequency spectrogram is logarithmically transformed and then subjected to discrete cosine transform to extract low-frequency features. The formula is: ; Where S(k) is the Mel spectrum, K is the number of Mel filters, and n is the MFCC coefficient index; Acquire environmental data including at least temperature, humidity, ambient noise decibel value, device speed, and load, and normalize the environmental data. It is understandable that the device speed and load are device operating parameters. The normalization formula is: ; X is the original data, X min and X max are the minimum and maximum values ​​of the data respectively; A timestamp alignment algorithm is used to match the acoustic features of the speech frame with the normalized environmental data to form a speech data sample with an environmental label. Specifically, a time window Δt (e.g., 50ms) is defined, and the environmental data E(t) is aligned with the speech frame V(t) according to the timestamp t, which can be expressed as: .

[0022] Step S02: construct a relationship graph based on a graph neural network based on the speech data samples with environmental labels.

[0023] It should be noted that the relationship graph is composed of nodes and edges. The nodes include speech nodes and environment nodes. The acoustic features under the speech frame are used as node attributes. Each speech frame corresponds to a speech node. Taking MFCC features as an example, MFCC features are used as speech nodes v voice The attribute of dimension D voice , which can be expressed as: ; Each environmental parameter is treated as a node, and the acquisition timestamp of the corresponding environmental parameter is given as an attribute. For example, the normalized environmental parameter is used as the environmental node v env Attributes, such as the noise decibel value d Bnorm , temperature T norm etc., can be expressed as: ; Each speech node establishes an undirected edge with all environment nodes at the same moment, and the weight is dynamically adjusted according to the degree of influence of environmental factors on speech, such as the noise weight w noise It can be expressed as: ,It is understandable that the higher the noise decibel value is, the larger the corresponding edge weight is; Establish edge connections between environmental nodes for preset environmental parameters, such as temperature and humidity, and device speed and load current. Calculate edge weights using the Pearson correlation coefficient to characterize the correlation between parameters. In addition, the relationship graph is encoded by graph structure, and the graph embedding algorithm is used to encode the graph structure into a low-dimensional vector representation, retaining the topological relationship between speech and environmental factors. Specifically, the GraphSAGE aggregation operation is used to encode the graph structure, and the current node representation is updated by aggregating the features of neighboring nodes. The formula is: ; in, is the feature vector of node v in layer l, N(v) is the set of neighbor nodes, is the learnable weight matrix and σ is the activation function.

[0024] Step S03: construct a speech error correction model integrating graph information, input the relationship graph into the speech error correction model for training, and obtain a target model.

[0025] The target model is deployed to the edge device, and edge computing power is used to achieve real-time inference. Specifically, the speech error correction model that integrates graph information includes a graph feature extraction layer and an attention fusion layer. The graph feature extraction layer uses GNN (such as GCN, GAT) to extract features from the relationship graph and output high-order semantic features of each node. It should be noted that the propagation rule of the graph convolutional network (GCN) can be expressed as: ; in, is the adjacency matrix with self-loops added, for The diagonal node degree matrix of is the feature matrix of the l-th layer node, is the weight matrix.

[0026] The attention fusion layer uses a multi-head attention mechanism to fuse the acoustic features of the speech frame with the environment map features of the corresponding node to generate a speech representation that includes environment perception. The multi-head attention mechanism is used to calculate the weighted sum of the speech feature Q and the environment map features K and V, which is expressed as: ; ; ; Among them, Concat() is a concatenation operation. is the weight matrix, For the i The calculation results of the attention head are, 、 、 is the linear transformation matrix corresponding to Q, K, and V, is the dimension of K.

[0027] In addition, an adaptive weight function is designed to dynamically adjust the weight of the anti-noise module according to the characteristic value of the environment node. For example, when the decibel value exceeds the threshold, the weight of the anti-noise module is increased from 0.3 to 0.8 through the sigmoid function, thereby enhancing the ability to suppress noise in the speech signal. In this embodiment of the present invention, the weight of the anti-noise module is dynamically adjusted according to the decibel value of the noise, which is expressed as: ; Among them, α is the weight of the anti-noise module, β is the hyperparameter for adjusting the slope, and θ is the threshold.

[0028] Before model training, the dataset can be expanded through data augmentation techniques (such as adding synthetic noise and simulating abnormal device states). During model training, the cross-entropy loss of speech error correction is combined with the Laplace regularization loss of the graph structure to ensure that the model maintains the semantic consistency of the graph structure while correcting speech errors. The loss function expression is: ; ; ; in, is the cross entropy loss of speech error correction, is the graph structure loss (such as the smoothness loss of node embedding vectors), is the balance coefficient, N is the number of samples, C is the number of categories, is the predicted text, y is the real text, ( i,j ) is an edge, w ij is the edge weight, h i , h j Embedding vector for the node.

[0029] In the embodiment of the present invention, the AdamW optimizer is used, combined with the learning rate warm-up and cosine annealing strategy to prevent model overfitting and accelerate convergence. The AdamW optimizer is used to update θ, which is expressed as: ; Where η is the learning rate, , are the first-order and second-order moment estimates, and ϵ is the smoothing term.

[0030] The cosine annealing strategy can be expressed as: ; η min , η max is the minimum / maximum learning rate, T cur , T max is the current / total number of training rounds.

[0031] Step S04: When the edge device receives new voice data and environmental data, it constructs a corresponding relationship graph and inputs the target model to generate a corrected text output.

[0032] Among them, the target model dynamically adjusts the error correction strategy based on the current environmental status determined by the graph features. Specifically, when the target model determines that the current environmental status is a preset noise environment based on the graph features, the anti-noise module will be triggered. Specifically, a noise reduction algorithm based on deep learning (such as a convolutional neural network noise reduction model) is used to process the noisy speech features, and then the noise-reduced speech features are input into the target model to generate a corrected text output. More specifically, noise is divided into periodic noise (such as machine roars), random noise (such as workshop voices) and burst noise (such as abnormal noise caused by equipment failure). For different types, the corresponding noise reduction filter is called. For example, periodic noise can be based on an adaptive notch filter, and the formula is: ; r is the attenuation coefficient, ω0 is the noise frequency; Random noise can be reduced using a deep learning noise reduction model (such as DnCNN), which learns the mapping relationship between noise and clean speech through end-to-end training. When the target model determines that the current environmental state is the preset operating state of the device based on the graph features, for voice instructions involving device operations, the device operating parameters (such as speed, load) in the environmental data are used to perform semantic verification on the instructions. For example, the instructions recognized by the voice are combined with the current state of the device to determine whether the instructions are reasonable. If the current device speed has reached the upper limit, and the voice instruction is "increase the speed", the instruction is corrected or an alarm is issued. In an embodiment of the present invention, a safety domain model of the device operating state is constructed, and parameters such as speed and load are mapped to safety areas in a multidimensional space. For example, the device speed S and load L need to satisfy the inequality group: ; When a voice command involves device operation, the command parameters are compared with the safety domain. If they are out of range, error correction is triggered. The risk of instruction execution is assessed through a pre-set rule base and machine learning model. It should be noted that the risk quantification model is constructed to calculate the risk value based on the comprehensive instruction danger level, the current status of the equipment, and the urgency of the operation, which is expressed as: ; ω1, ω2, ω3 are weight coefficients; Then different treatment methods are adopted according to the risk value, for example: R≥0.8: Force the instruction to be modified and issue a warning (e.g., change "increase speed to limit" to "increase speed by 10%"); 0.5≤R<0.8: Provide multiple correction suggestions for operator confirmation; R<0.5: Only records potential risks and does not interfere with instruction execution.

[0033] In summary, an embodiment of the present invention proposes a speech error correction method based on edge computing power. The method obtains speech data and environmental data, and aligns the speech data and environmental data to form a speech data sample with an environmental label; constructs a relationship graph based on a graph neural network according to the speech data sample with the environmental label; constructs a speech error correction model that integrates graph information, inputs the relationship graph into the speech error correction model for training, and obtains a target model, wherein the target model is deployed to the edge device, and edge computing power is used to achieve real-time reasoning; when the edge device receives new speech data and environmental data, it constructs a corresponding relationship graph and inputs it into the target model to generate a corrected text output, wherein the target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features. Specifically, through the deep collaboration of speech and environmental data, the accuracy and robustness of edge speech error correction are effectively enhanced.

[0034] Example 2 The second embodiment of the present invention provides a speech error correction system 200 based on edge computing power, see Figure 2 , is a structural block diagram of a speech error correction method system based on edge computing power, the speech error correction system based on edge computing power 200 includes: An acquisition module 21 is configured to acquire speech data and environmental data, and align the speech data with the environmental data to form a speech data sample with an environmental label; A construction module 22 is configured to construct a graph neural network-based relationship graph based on the speech data samples with the environment label, wherein the relationship graph is composed of nodes and edges, wherein the nodes include speech nodes and environment nodes, and the acoustic features of the speech frame are used as node attributes, and each speech frame corresponds to a speech node; Each environmental parameter is treated as a separate node, and the collection timestamp of the corresponding environmental parameter is given as an attribute; Each speech node establishes undirected edges with all environmental nodes at the same moment, and the weights are dynamically adjusted based on the degree of impact of environmental factors on speech; Between environment nodes, edge connections are established for preset environment parameters, and edge weights are calculated using the Pearson correlation coefficient to characterize the correlation between parameters. In addition, the relationship graph is encoded into a graph structure using a graph embedding algorithm to encode the graph structure into a low-dimensional vector representation, preserving the topological relationship between speech and environmental factors; A training module 23 is configured to construct a speech error correction model integrating graph information, input the relationship graph into the speech error correction model for training, and obtain a target model, wherein the target model is deployed to the edge device, and edge computing power is used to implement real-time inference. The speech error correction model integrating graph information includes a graph feature extraction layer and an attention fusion layer, wherein the graph feature extraction layer uses GNN to extract features from the relationship graph and outputs high-order semantic features of each node. The attention fusion layer uses a multi-head attention mechanism to fuse the acoustic features of the speech frame with the environmental graph features of the corresponding node to generate a speech representation that includes environmental perception. In addition, an adaptive weight function is designed to dynamically adjust the weight of the anti-noise module according to the feature values ​​of the environmental nodes; Combining the cross entropy loss of speech error correction with the Laplace regularization loss of the graph structure ensures that the model maintains the semantic consistency of the graph structure while correcting speech errors. The loss function expression is: ; in, is the cross entropy loss of speech error correction, is the graph structure loss, is the balance coefficient; An input module 24 is configured to construct a corresponding relationship graph and input the graph into the target model when the edge device receives new voice data and environmental data, thereby generating a corrected text output. The target model dynamically adjusts the error correction strategy based on the graph features to determine the current environmental state. Specifically, when the target model determines that the current environmental state is a preset noise environment based on the graph features, the anti-noise module is triggered. Specifically, a deep learning-based noise reduction algorithm is used to process the noisy voice features, and then the noise-reduced voice features are input into the target model to generate a corrected text output. When the target model determines that the current environmental state is the preset operating state of the device based on the graph features, for voice commands involving device operations, the device operating parameters in the environmental data are used to perform semantic verification on the commands; The risk of instruction execution is assessed through a pre-set rule base and machine learning models.

[0035] Furthermore, in some other embodiments of the present invention, the acquisition module 21 includes: The transform unit is used to obtain voice data, use beamforming technology to capture the human voice in a direction, and use short-time Fourier transform to convert the time-domain voice signal into a frequency-domain Mel-spectrogram; An extraction unit, configured to extract acoustic features based on the frequency domain Mel-spectrogram; an acquisition unit, configured to acquire environmental data including at least temperature, humidity, ambient noise decibel value, device speed, and load, and normalize the environmental data; The matching unit is used to match the acoustic features of the speech frame with the normalized environmental data using a timestamp alignment algorithm to form a speech data sample with an environmental label.

[0036] Example 3 The third embodiment of the present invention provides an electronic device, see Figure 3 , is a structural block diagram of an electronic device, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, the speech error correction method based on edge computing power as described above is implemented.

[0037] In some embodiments, the processor 10 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip, used to run program codes or process data stored in the memory 20, such as executing access restriction programs.

[0038] The memory 20 includes at least one type of readable storage medium, including flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 20 may be an internal storage unit of the electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 may also be an external storage device of the electronic device, such as a plug-in hard disk equipped on the electronic device, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Furthermore, the memory 20 may include both an internal storage unit of the electronic device and an external storage device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or is about to be output.

[0039] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned edge computing-based speech error correction method.

[0040] Those skilled in the art will appreciate that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device.

[0041] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0042] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0043] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0044] The above embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A speech error correction method based on edge computing power, characterized in that: Applied to an industrial workshop scenario where edge devices are installed, the method includes: Acquire voice data and environmental data, and align the voice data with the environmental data to form a voice data sample with an environmental label; Based on speech data samples with environmental labels, a relationship graph based on graph neural network is constructed; Constructing a speech error correction model that integrates graph information, inputting the relationship graph into the speech error correction model for training to obtain a target model, wherein the target model is deployed to the edge device to implement real-time inference using edge computing power; When the edge device receives new voice data and environmental data, it constructs a corresponding relationship graph and inputs it into the target model to generate a corrected text output. The target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features.

2. The speech error correction method based on edge computing power according to claim 1 is characterized in that: The step of acquiring voice data and environmental data, and aligning the voice data and the environmental data to form a voice data sample with an environmental label includes: Acquire speech data, use beamforming technology to capture the human voice in a directionally controlled manner, and use short-time Fourier transform to convert the time-domain speech signal into a frequency-domain Mel-spectrogram; Extracting acoustic features according to the frequency domain Mel-spectrogram; Acquire environmental data including at least temperature, humidity, ambient noise decibel value, device speed, and load, and normalize the environmental data; A timestamp alignment algorithm is used to match the acoustic features of the speech frame with the normalized environmental data to form a speech data sample with an environmental label.

3. The speech error correction method based on edge computing power according to claim 2, characterized in that: In the step of constructing a relationship graph based on a graph neural network based on the speech data sample with the environment label, the relationship graph is composed of nodes and edges, the nodes include speech nodes and environment nodes, the acoustic features under the speech frame are used as node attributes, and each speech frame corresponds to a speech node; Each environmental parameter is treated as a separate node, and the collection timestamp of the corresponding environmental parameter is given as an attribute; Each speech node establishes undirected edges with all environmental nodes at the same moment, and the weights are dynamically adjusted based on the degree of impact of environmental factors on speech; Between environment nodes, edge connections are established for preset environment parameters, and edge weights are calculated using the Pearson correlation coefficient to characterize the correlation between parameters. In addition, the relationship graph is encoded into a graph structure and a graph embedding algorithm is used to encode the graph structure into a low-dimensional vector representation, preserving the topological relationship between speech and environmental factors.

4. The speech error correction method based on edge computing power according to claim 3 is characterized in that: The speech error correction model that integrates graph information includes a graph feature extraction layer and an attention fusion layer. The graph feature extraction layer uses GNN to extract features from the relationship graph and outputs high-order semantic features of each node. The attention fusion layer uses a multi-head attention mechanism to fuse the acoustic features of the speech frame with the environmental graph features of the corresponding node to generate a speech representation that includes environmental perception. In addition, an adaptive weight function is designed to dynamically adjust the weight of the anti-noise module according to the characteristic values ​​of the environmental nodes.

5. The speech error correction method based on edge computing power according to claim 4 is characterized in that: In the step of inputting the relationship graph into the speech error correction model for training to obtain the target model, the cross entropy loss of speech error correction and the Laplace regularization loss of the graph structure are combined to ensure that the model maintains the semantic consistency of the graph structure while correcting speech errors. The loss function expression is: ； in, is the cross entropy loss of speech error correction, is the graph structure loss, is the balance coefficient.

6. The speech error correction method based on edge computing power according to claim 5, characterized in that: The step of dynamically adjusting the error correction strategy by the target model based on the graph features to determine the current environment state includes: When the target model determines that the current environment state is a preset noise environment based on the graph features, the anti-noise module will be triggered. Specifically, a noise reduction algorithm based on deep learning is used to process the noisy speech features, and then the noise-reduced speech features are input into the target model to generate a corrected text output; When the target model determines that the current environmental state is the preset operating state of the device based on the graph features, for voice commands involving device operations, the device operating parameters in the environmental data are used to perform semantic verification on the commands; The risk of instruction execution is assessed through a pre-set rule base and machine learning models.

7. A speech error correction system based on edge computing power, characterized in that: For implementing the speech error correction method based on edge computing power according to any one of claims 1 to 6, the system comprises: An acquisition module is used to acquire voice data and environmental data, and align the voice data and the environmental data to form a voice data sample with an environmental label; A construction module for constructing a graph neural network-based relationship graph based on speech data samples with environmental labels; A training module is configured to construct a speech error correction model integrating graph information, input the relationship graph into the speech error correction model for training, and obtain a target model, wherein the target model is deployed to the edge device to implement real-time inference using edge computing power; The input module is used to construct a corresponding relationship graph and input the target model when the edge device receives new voice data and environmental data, generating a corrected text output, wherein the target model dynamically adjusts the error correction strategy based on the current environmental status according to the graph features.

8. A computer-readable storage medium, characterized in that include: The readable storage medium stores one or more programs, which, when executed by a processor, implement the edge computing-based speech error correction method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, it implements the edge computing-based speech error correction method described in any one of claims 1-6.

Images