Speech recognition method and device, computer device, and storage medium

Abstract

Embodiments of the present application provide a speech recognition method and device, computer equipment and a storage medium, belonging to the technical field of financial technology. The method comprises: obtaining target speech data; inputting the target speech data into a preset target speech recognition model; wherein the target speech recognition model comprises a target time feature extraction network, a target space feature extraction network and a target speech recognition network; performing feature extraction on the target speech data through the target time feature extraction network to obtain target speech time sequence features; performing feature extraction on the target speech data through the target space feature extraction network to obtain target speech space features; and performing speech recognition on the target speech time sequence features and the target speech space features through the target speech recognition network. The embodiments of the present application can improve the accuracy of speech recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of financial technology, and particularly relates to a voice recognition method and device, computer equipment and a storage medium. BACKGROUND

[0002] With the combination of computer technology and the financial field, more and more intelligent platforms provide financial services to users to improve the experience of users when handling financial business. For example, in the insurance business service, in order to save the configuration of insurance business personnel, the conventional insurance consultation question is usually answered by voice by the financial intelligent robot. However, in order to improve the accuracy of voice reply and improve the experience of users in the question and answer process, it is necessary to recognize the voice of the user, and determine the meaning of the question of the user according to the voice recognition result. In the related art, the voice recognition is mainly achieved by recognizing the voice features of different time sequences, and completing the voice recognition according to the voice features, but the voice recognition is only achieved from the voice features in the time dimension, and the real meaning of the voice cannot be accurately recognized, which affects the accuracy of voice recognition. Therefore, how to improve the accuracy of voice recognition has become a technical problem to be solved. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a voice recognition method and device, computer equipment and a storage medium, which aims to improve the accuracy of voice recognition.

[0004] To achieve the above-mentioned purpose, a first aspect of the embodiments of the present application provides a voice recognition method, which comprises the following steps:

[0005] obtaining target voice data;

[0006] inputting the target voice data into a preset target voice recognition model; wherein the target voice recognition model comprises a target time feature extraction network, a target space feature extraction network and a target voice recognition network;

[0007] extracting features of the target voice data through the target time feature extraction network to obtain target voice time sequence features;

[0008] extracting features of the target voice data through the target space feature extraction network to obtain target voice space features;

[0009] recognizing the target voice time sequence features and the target voice space features through the target voice recognition network.

[0010] In some embodiments, the target spatial feature extraction network is a compressed excitation network, the compressed excitation network comprising: a linear layer and a convolutional layer; the feature extraction of the target speech data by the target spatial feature extraction network to obtain target speech spatial features comprises:

[0011] the feature extraction of the target speech data by the linear layer to obtain at least two initial speech spatial features; wherein each initial speech spatial feature represents speech features on a spatial channel;

[0012] the splicing processing of the at least two initial speech spatial features by the linear layer to obtain candidate speech spatial features;

[0013] the weighted calculation of the candidate speech spatial features by the convolutional layer to obtain the target speech spatial features.

[0014] In some embodiments, the splicing processing of the at least two initial speech spatial features by the linear layer to obtain candidate speech spatial features comprises:

[0015] the dimension expansion of each initial speech spatial feature by the linear layer to obtain preliminary speech spatial features;

[0016] the splicing processing of the at least two preliminary speech spatial features by the linear layer to obtain the candidate speech spatial features.

[0017] In some embodiments, the weighted calculation of the candidate speech spatial features by the convolutional layer to obtain the target speech spatial features comprises:

[0018] the pooling processing of the candidate speech spatial features by the convolutional layer to obtain speech pooling features; the normalization processing of the speech pooling features by the convolutional layer to obtain a preset weight of each spatial channel;

[0019] the weighted summation of the candidate speech spatial features and the preset weight by the convolutional layer to obtain the target speech spatial features.

[0020] In some embodiments, the time feature extraction network is a self-attention network, the self-attention network comprising: a feature extraction layer and a self-attention layer; the feature extraction of the target speech data by the target time feature extraction network to obtain target speech time features comprises:

[0021] the feature extraction of the target speech data by the feature extraction layer to obtain preliminary speech time sequence features;

[0022] performing self-attention calculation on the preliminary speech time sequence feature through the self-attention layer to obtain self-attention weights;

[0023] performing weighted calculation on the preliminary speech time sequence feature according to the self-attention weights to obtain the target speech time sequence feature.

[0024] In some embodiments, the performing speech recognition on the target speech time sequence feature and the target speech spatial feature through the target speech recognition network comprises:

[0025] performing feature fusion on the target speech time sequence feature and the target speech spatial feature through the target speech recognition network to obtain target speech fusion features;

[0026] performing speech recognition on the target speech fusion features through the target speech recognition network.

[0027] In some embodiments, before the inputting the target speech data into the preset target speech recognition model, the method further comprises:

[0028] constructing the target speech recognition model, specifically comprising:

[0029] obtaining training speech data;

[0030] inputting the training speech data into a preset original speech recognition model; wherein the original speech recognition model comprises an original time feature extraction network, an original spatial feature extraction network and an original speech recognition network:

[0031] performing feature extraction on the training speech data through the original time feature extraction network to obtain training speech time sequence features;

[0032] performing feature extraction on the training speech data through the original spatial feature extraction network to obtain training speech spatial features;

[0033] performing speech recognition on the training speech time sequence features and the training speech spatial features through the original speech recognition network to obtain a speech recognition result;

[0034] performing parameter adjustment on the original speech recognition model according to a preset speech verification result and the speech recognition result to obtain the target speech recognition model.

[0035] To achieve the above object, a second aspect of the embodiment of the present application proposes a speech recognition device, which comprises:

[0036] a data obtaining module configured to obtain target speech data;

[0037] The voice data processing module is configured to input the target voice data into a preset target voice recognition model; the target voice recognition model comprises a target time feature extraction network, a target space feature extraction network, and a target voice recognition network.

[0038] The time sequence feature extraction module is configured to perform feature extraction on the target voice data by using the target time feature extraction network to obtain target voice time sequence features.

[0039] The space feature extraction module is configured to perform feature extraction on the target voice data by using the target space feature extraction network to obtain target voice space features.

[0040] The voice recognition module is configured to perform voice recognition on the target voice time sequence features and the target voice space features by using the target voice recognition network.

[0041] To achieve the above object, a third aspect of the embodiment of the present application provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method of the first aspect when executing the computer program.

[0042] To achieve the above object, a fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method of the first aspect.

[0043] The voice recognition method and device, computer device and storage medium provided by the present application input target voice data into a target voice recognition model comprising a target time feature extraction network, a target space feature extraction network and a target voice recognition network, extract target voice time sequence features from the target voice data by using the target time feature extraction network, extract target voice space features from the target voice data by using the target space feature extraction network, and then perform voice recognition on the target voice time sequence features and the target voice space features by using the target voice recognition network. Therefore, the voice recognition is combined with the voice time sequence features and the voice space features, the different types of voice features in the voice recognition are enhanced, the features in the voice data are better learned, and the accuracy of the voice recognition is improved. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a flowchart of the voice recognition method provided by the embodiment of the present application;

[0045] Figure 2 is a flowchart of the voice recognition method provided by another embodiment of the present application;

[0046] Figure 3is a structural schematic diagram of a target speech recognition model in a speech recognition method provided by an embodiment of the present application.

[0047] Figure 4 is Figure 1 is a flowchart of step S103 in the method.

[0048] Figure 5 is a structural schematic diagram of a compression excitation network in a speech recognition method provided by an embodiment of the present application.

[0049] Figure 6 is Figure 1 is a flowchart of step S104 in the method.

[0050] Figure 7 is Figure 6 is a flowchart of step S602 in the method.

[0051] Figure 8 is a process schematic diagram of linear layer processing target speech data in a speech recognition method provided by an embodiment of the present application.

[0052] Figure 9 is Figure 6 is a flowchart of step S603 in the method.

[0053] Figure 10 is Figure 1 is a flowchart of step S105 in the method.

[0054] Figure 11 is a whole flowchart of a speech recognition method provided by an embodiment of the present application.

[0055] Figure 12 is a structural schematic diagram of a speech recognition device provided by an embodiment of the present application.

[0056] Figure 13 is a hardware structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0058] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a manner different from the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification and claims and the above drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0060] First, let's analyze some of the terms used in this application:

[0061] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0062] Speech recognition (ASR), also known as automatic speech recognition, is the process by which a computer or device converts spoken speech signals into corresponding text or instructions. Speech recognition systems analyze the frequency, amplitude, and temporal characteristics of speech signals to extract basic speech units (such as phonemes and words), and then combine these units into understandable text or commands through matching and modeling.

[0063] Self-attention mechanism, also known as internal attention mechanism, is a special form of attention mechanism used to process sequential data (such as text). Each element in the sequence can interact with other elements in the sequence, thereby capturing global dependencies. Self-attention is widely used in the Transformer model and is one of its core components.

[0064] Squeeze-and-Excitation Network (SENet): is a technology to improve the architecture of Convolutional Neural Network (CNN), aiming to enhance the ability of the network to dynamically adjust the importance of features. The core idea of SENet is to improve the feature representation ability by explicitly modeling the dependence between channels. SENet introduces a "Squeeze-and-Excitation" module, referred to as SE module, which can be inserted into existing convolutional neural networks and improve the performance of the model by reweighting channel features.

[0065] Feedforward Neural Network (FNN): is the most basic and simplest artificial neural network structure. The information in the feedforward network only propagates in one direction, i.e. from the input layer to the output layer through one or more hidden layers, without loop or feedback connection.

[0066] In order to adapt to the intelligent development, different types of financial service platforms are set up in the financial field, and in order to improve the user experience when using the financial service platform, the voice recognition function is increased, and the user can perform financial business consultation and handling on the financial service platform through voice. Therefore, the accuracy of voice recognition affects the user experience on the financial service platform, and improving the accuracy of voice recognition has become a key concern in the financial field.

[0067] In related technologies, voice recognition is mainly achieved through attention mechanism, which weights and summarizes the input feature sequence through attention mechanism, realizes the learning of prosodic information and the weight distribution of voice features at different time sequences. Since the time sequence information in the voice signal is very important, in the voice recognition process, the voice features at different time sequences are mixed and then voice recognition is performed. However, only voice features in the time sequence dimension are used for voice recognition, which lacks voice information in space, affecting the accuracy of voice recognition.

[0068] Based on this, the embodiments of the present application provide a voice recognition method and device, computer equipment and storage medium, which extracts target voice time sequence features from target voice data through a target time feature extraction network, extracts target voice space features from target voice data through a target space feature extraction network, and then combines target voice time sequence features and target voice space features to perform voice recognition. In the voice recognition process, the voice features between time and space are integrated, the perception and expression ability of voice recognition is enhanced, and the accuracy of voice recognition is improved.

[0069] The voice recognition method and device, computer equipment and storage medium provided by the embodiments of the present application are specifically described by the following embodiments. First, the voice recognition method in the embodiments of the present application is described.

[0070] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. The artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results.

[0071] The artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The artificial intelligence software technology mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0072] The voice recognition method provided by the embodiments of the present application relates to the field of artificial intelligence. The voice recognition method provided by the embodiments of the present application can be applied in a terminal, can be applied in a server end, and can also be software running in a terminal or a server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc.; the server end can be configured as an independent physical server, can be configured as a server cluster or a distributed system composed of multiple physical servers, can also be configured as a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and basic cloud computing services such as big data and artificial intelligence platforms; and the software can be an application for implementing the voice recognition method, but is not limited to the above forms.

[0073] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0074] It should be noted that in various specific embodiments of the present application, when relevant processing needs to be performed on data related to the identity or characteristics of the user, such as user information, user behavior data, user history data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of such data will comply with relevant laws, regulations, and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for enabling the embodiments of the present application to function normally will be obtained.

[0075] Figure 1 is an optional flowchart of a speech recognition method provided by the embodiments of the present application, Figure 1 The method in can include but is not limited to steps S101-S105.

[0076] Step S101, obtaining target speech data;

[0077] Step S102, inputting the target speech data into a preset target speech recognition model; wherein the target speech recognition model includes a target time feature extraction network, a target space feature extraction network, and a target speech recognition network;

[0078] Step S103, performing feature extraction on the target speech data through the target time feature extraction network to obtain target speech time sequence features;

[0079] Step S104, performing feature extraction on the target speech data through the target space feature extraction network to obtain target speech space features;

[0080] Step S105, performing speech recognition on the target speech time sequence features and the target speech space features through the target speech recognition network.

[0081] The steps S101-S105 shown in the embodiments of the present application are performed by constructing a target speech recognition module including a target time feature extraction network, a target space feature extraction network, and a target speech recognition network, extracting target speech time sequence features from the target speech data through the target time feature extraction network, extracting target speech space features from the target speech data through the target space feature extraction network, and then combining the target speech time sequence features and the target speech space features to perform speech recognition, which can enhance the perception and expression ability in the speech recognition process to improve the accuracy of speech recognition.

[0082] In step S101 of some embodiments, the voice recognition method applies an insurance service platform in the financial field, and target voice data is collected through an audio recording device in the insurance service platform, or uploaded to the insurance service platform through a user terminal. The embodiment does not make specific limitations on the acquisition method of the target voice data.

[0083] Before step S101 of some embodiments, the voice recognition method further includes constructing a target voice recognition model, which is used to recognize the voice content of the target voice data. In some embodiments, the target voice recognition model includes a target time feature extraction network, a target space feature extraction network, and a target voice recognition network. The target time feature extraction network is a self-attention network composed of multiple self-attention layers, each of which contains a self-attention mechanism to capture the context information of the target voice data in time sequence, thereby improving the accuracy of voice recognition. The target space feature extraction network is a compressed excitation network, abbreviated as SENet, which includes a feature perception module, and the feature perception module is an SE module that can enhance the perception ability of the convolution layer to the importance of the feature and output voice features representing spatial relationships. The target voice recognition network is a feedforward network that performs voice recognition based on the input voice features and outputs voice recognition results.

[0084] In some embodiments, the target voice recognition model is obtained by training an original voice recognition model, and the original voice recognition model includes an original time feature extraction network, an original space feature extraction network, and an original voice recognition network. The original time feature extraction network is trained to obtain the target time feature extraction network, and the structure of the original time feature extraction network is the same as that of the target time feature extraction network. The original space feature extraction network is trained to obtain the target space feature extraction network, and the structure of the original space feature extraction network is the same as that of the target space feature extraction network. The original voice recognition network is trained to obtain the target voice recognition network, and the structure of the original voice recognition network is the same as that of the target voice recognition network.

[0085] Please refer to Figure 2 In some embodiments, constructing the target voice recognition model can include but is not limited to steps S201 to S206:

[0086] Step S201, obtaining training voice data;

[0087] Step S202, inputting the training voice data into a preset original voice recognition model; wherein the original voice recognition model includes an original time feature extraction network, an original space feature extraction network, and an original voice recognition network:

[0088] Step S203, feature extraction is performed on the training speech data by the original time feature extraction network to obtain training speech time sequence features;

[0089] Step S204, feature extraction is performed on the training speech data by the original space feature extraction network to obtain training speech space features;

[0090] Step S205, speech recognition is performed on the training speech time sequence features and the training speech space features by the original speech recognition network to obtain a speech recognition result.

[0091] Step S206, parameter adjustment is performed on the original speech recognition model according to the preset speech verification result and the speech recognition result to obtain a target speech recognition model.

[0092] In step S201 of some embodiments, the training speech data is acquired in the same way as the target speech data, that is, by the insurance service platform. However, the training speech data is not acquired in real time by the insurance service platform, but is acquired before the speech recognition function of the insurance service platform is started, and is also acquired periodically during the use of the speech recognition function of the insurance service platform. It should be noted that the training speech data is acquired before the insurance service platform starts the speech recognition function, and is used to build the target speech recognition model. After the insurance service platform starts the speech recognition function, the target speech data collected according to the preset period is stored in the pre-set database, and the target speech data is taken out from the database as training speech data according to the preset period, so the training speech data after the speech recognition function is started is used to update the target speech recognition model, and the recognition performance of the target speech recognition model is constantly improved.

[0093] In step S202 of some embodiments, the original speech recognition model composed of the original time feature extraction network, the original space feature extraction network and the original speech recognition network is built, so that the original speech recognition model can acquire speech features in time and space, and perform speech recognition after the speech features in time and space are fused, so that the speech recognition can understand the mutual relationship of different parts of the speech, and can better understand the complete speech data to achieve accurate speech recognition.

[0094] It should be noted that the original spatial feature extraction network is a compression excitation network, the compression excitation network is a network specially used for image classification, which can optimize the attention allocation of the spatial channel in speech recognition by automatically learning the inter-channel correlation, and can extract key features from a large amount of speech data, and output speech features representing more accurate spatial relationships. The original time feature extraction network adopts a self-attention mechanism to obtain speech features from the time sequence of the speech data, and the self-attention mechanism can pay attention to the correlation of the speech data at different time sequences, and output speech features representing more accurate time sequence relationships.

[0095] In step S203 of some embodiments, the original time feature extraction network is a self-attention network, and the self-attention network includes a plurality of self-attention layers. The self-attention mechanism of the self-attention layer can capture the speech features of the training speech data in the time sequence, and output training speech time sequence features accurately representing the time sequence relationship. It should be noted that the training speech time sequence features are represented by a sequence, and the values in the sequence represent the speech information at different time points. The context information of the training speech data on the time axis can be determined by the training speech time sequence features.

[0096] In step S204 of some embodiments, the original spatial feature extraction network is a compression excitation network, and the compression excitation network enhances the perception ability of the convolutional layer to the speech features by introducing an SE module. Therefore, the original spatial feature extraction network extracts speech features representing more accurate spatial relationships from the training speech data. It should be noted that the training speech spatial features can represent the spatial relationship of the speech, so as to judge the relationship between different parts of the speech in the space, which is beneficial to understanding the information in the speech data.

[0097] In step S205 of some embodiments, the training speech time sequence features and the training speech spatial features are simultaneously input into the original speech recognition network, the original speech recognition network collects the speech features in the time sequence and the space for speech recognition, which can more accurately identify the information in the speech features, so as to output accurate speech recognition results. It should be noted that the speech recognition result can be the phoneme, speech content or speech command of the training speech data, and the speech recognition result is not limited here.

[0098] In step S206 of some embodiments, the speech verification result is the recognition result of the training speech data pre-configured, which is used to judge the accuracy of the original speech recognition model in speech recognition. Specifically, the recognition loss data is obtained by loss calculation on the language verification result and the language recognition result, and the recognition loss data represents the recognition error of the original speech recognition model. The original speech recognition model is adjusted according to the recognition loss data, and the original speech recognition model after parameter adjustment is used as the target speech recognition model after the recognition loss data reaches the minimum value.

[0099] For example, if the original speech recognition model is used to identify the content of the training speech data, the speech verification result is set in advance, and the speech verification result is a pre-set insurance question such as “What insurance can a person aged 50 with coronary heart disease purchase?”. The corresponding speech data is manually recorded as training speech data according to the insurance question, and then the training speech data is input into the original speech model to obtain a speech recognition result. If the speech recognition result is “What insurance can a person aged 50 with coronary heart disease purchase?”, the recognition loss data between the speech verification result and the language recognition result is calculated, and the recognition loss data represents the error of speech content recognition. The original speech recognition model is adjusted through the recognition loss data until the speech recognition result output by the original speech recognition model is “What insurance can a person aged 50 with coronary heart disease purchase?”, so as to complete the training of the original speech recognition model, and the trained original speech recognition model is used as the target speech recognition model.

[0100] As disclosed above, after the construction of the target speech recognition model is completed, when the speech recognition function of the insurance service platform is started, if the target speech data is received, the target speech data is directly input into the target speech recognition model, and the target speech data is recognized by the target speech recognition model.

[0101] In step S102 of some embodiments, the structure of the target speech recognition model is as shown in Figure 3 As can be seen from Figure 3 It can be seen that the target speech recognition model includes a target time feature extraction network, a target space feature extraction network, and a target speech recognition network. The target time feature extraction network is a self-attention network, the target space feature extraction network is a compressed excitation network, and the target speech recognition network is a feedforward network. The target speech data is input into the self-attention network and the compressed excitation network, respectively. The features output by the self-attention network and the compressed excitation network are fused and input into the feedforward network. The feedforward network combines the speech features in space and time to perform speech recognition, thereby improving the accuracy of speech recognition.

[0102] In some embodiments, the self-attention network includes a feature extraction layer and at least two self-attention layers. Each self-attention layer includes a self-attention mechanism component and a normalization layer. The self-attention mechanism component is used to identify the importance of different time points in the speech features, and the normalization layer is used to normalize the importance of different time points.

[0103] Please refer to Figure 4 In some embodiments, step S103 can include but is not limited to steps S401 to S403:

[0104] In step S401, the feature extraction layer is used to perform feature extraction on the target voice data to obtain preliminary voice time sequence features.

[0105] In step S402, the self-attention layer is used to perform self-attention calculation on the preliminary voice time sequence features to obtain self-attention weights.

[0106] In step S403, the preliminary voice time sequence features are calculated by weighting according to the self-attention weights to obtain target voice time sequence features.

[0107] In step S401 of some embodiments, the feature extraction layer is also referred to as a linear transformation layer, and the target voice data is mapped into a space corresponding to the query, key and value by the linear transformation layer to obtain the preliminary voice time sequence features. It should be noted that the preliminary voice time sequence features are a sequence containing query vectors, key vectors and value vectors, and the importance of the voice features at different time points can be calculated by the query vectors, key vectors and value vectors.

[0108] In step S402 of some embodiments, the self-attention layer performs self-attention calculation on the preliminary voice time sequence features, that is, the similarity between the query vectors and the key vectors is calculated as the self-attention weights. It should be noted that the self-attention weights represent the importance of the voice features at different time points, and the voice features at different time points are focused on by the self-attention weights.

[0109] In step S403 of some embodiments, the self-attention weights and the preliminary voice time sequence features are calculated by weighting, that is, the feature importance in the preliminary voice time sequence features is adjusted to obtain the target voice time sequence features, and the target voice time sequence features can accurately represent the voice information at different time points.

[0110] In steps S401 to S403 shown in the present embodiment, the feature extraction layer is used to first perform feature extraction on the target voice data to obtain preliminary voice time sequence features, then the self-attention layer is used to perform self-attention calculation on the preliminary voice time sequence features to calculate the self-attention weights representing the importance of the voice features at different time points, and finally the preliminary voice time sequence features are calculated by weighting according to the self-attention weights to output the target voice time sequence features representing the time relationship more accurately.

[0111] Referring to Figure 5 In some embodiments, the compression excitation network comprises a linear layer and a convolutional layer. It should be noted that at least two linear layers are provided, and the outputs of the at least two linear layers are connected to the convolutional layer. The SE module is added in the convolutional layer, and the SE module can enhance the perception ability of the convolutional layer and improve the performance of the compression excitation network.

[0112] Referring to Figure 6In some embodiments, step S104 can include but is not limited to steps S601-S603:

[0113] In step S601, feature extraction is performed on the target speech data by a linear layer to obtain at least two initial speech space features; each initial speech space feature represents speech features on a space channel.

[0114] In step S602, at least two initial speech space features are spliced by a linear layer to obtain candidate speech space features.

[0115] In step S603, the candidate speech space features are weighted and calculated by a convolution layer to obtain target speech space features.

[0116] In step S601 of some embodiments, the linear layer extracts features from the target speech data, i.e., maps the target speech data to each space channel by the linear layer to obtain initial speech space features. The at least two initial speech space features are features mapped by different space channels, representing speech features in different space channels. It should be noted that at least two linear layers are provided, each linear layer maps the target speech data to a space, and different linear layers are configured in different spaces.

[0117] In step S602 of some embodiments, at least two initial speech space features are spliced by a linear layer. It should be noted that splicing is to combine initial speech space features mapped by different space channels, so the candidate speech space features represent speech features of different space relationships.

[0118] In step S603 of some embodiments, the convolution layer adds an SE module, which can calculate the importance of each space channel in the convolution layer, and then weight and calculate the candidate speech space features of different space channels according to the importance of the channel to obtain speech features representing more accurate space relationships.

[0119] In steps S601-S603 shown in this embodiment, a compression excitation network containing a linear layer and a convolution layer is set to extract at least two initial speech space features from the target speech data by the linear layer, splice the at least two initial speech space features into candidate speech space features by the linear layer, and finally weight and calculate the candidate speech space features by the convolution layer to obtain target speech space features representing more accurate space relationships.

[0120] Please refer to Figure 7 In some embodiments, step S602 can include but is not limited to steps S701-S702:

[0121] Step S701: Dimensional augmentation is performed on each initial speech space feature through a linear layer to obtain preliminary speech space features;

[0122] Step S702: At least two preliminary speech space features are concatenated through a linear layer to obtain candidate speech space features.

[0123] In step S701 of some embodiments, the dimension of the initial speech spatial features is the same as the dimension of the target speech data. However, the candidate speech spatial features are speech features that contain different spatial features, so the dimension of the candidate speech spatial features is larger than that of the initial speech spatial features. Therefore, the initial speech spatial features cannot be directly concatenated into candidate speech spatial features. Therefore, a linear layer is used to expand the dimension of each initial speech spatial feature. In this embodiment, the initial speech spatial features are expanded by one channel dimension to obtain preliminary speech spatial features.

[0124] For example, if the target speech data is defined as X, and the shape of the target speech data is (B, T, D), and the shape of the initial speech space feature is also (B, T, D), then by expanding the initial speech space feature by one channel dimension, a preliminary speech space feature is obtained, and the shape of the preliminary speech space feature is (B, 1, T, D).

[0125] In step S702 of some embodiments, at least two preliminary speech space features are concatenated into candidate speech space features, that is, speech space features from different spatial channels are concatenated into candidate speech space features containing multiple spatial channels.

[0126] For example, such as Figure 8 As shown, the shape of the preliminary speech space features is (B, I, T, D). In this embodiment, C linear layers are set, meaning there are C preliminary speech space features. By concatenating the C preliminary speech space features, the target speech space features with the shape (B, C, T, D) are obtained.

[0127] Specifically, for the target speech data X, it is represented by different spatial channels as shown in equations (1) to (3):

[0128] X′ i =X*W1 i +b1 i (1)

[0129] X′ i =X′ i .unsqueeze (2)

[0130] X″=concat(X′1,…,X′ i (3)

[0131] In the formula, X′ iW1 i b1 i b1 unsqueeze represents dimension expansion on the initial speech spatial feature, X" represents the candidate speech spatial feature, and concat represents splicing of the preliminary speech spatial features.

[0132] In steps S701 to S702 shown in the embodiment, the preliminary speech spatial features are obtained by dimension expansion on the initial speech spatial features, and the at least two preliminary speech spatial features are spliced into the candidate speech spatial features, thereby constructing the candidate speech spatial features under different spatial channels.

[0133] Referring to Figure 9 In some embodiments, step S603 includes but is not limited to steps S901 to S903.

[0134] In step S901, the convolution layer is used to perform pooling processing on the candidate speech spatial features to obtain speech pooling features.

[0135] In step S902, the convolution layer is used to perform normalization processing on the speech pooling features to obtain preset weights of each spatial channel.

[0136] In step S903, the convolution layer is used to perform weighted summation on the candidate speech spatial features and the preset weights to obtain target speech spatial features.

[0137] It should be noted that the convolution layer of the embodiment adds an SE module, and the SE module includes two stages, namely compression and excitation. The compression stage performs pooling processing on the speech features in each spatial channel, and the excitation stage is used to generate weights of each spatial channel.

[0138] In step S901 of some embodiments, the convolution layer performs pooling processing on the candidate speech spatial features, and the pooling processing is the compression stage of the SE module, that is, the speech features of each spatial channel are determined through the candidate speech spatial features, and the speech features of each spatial channel are compressed into a global description vector. It should be noted that the speech pooling features are global description vectors, and the global information of each spatial channel is determined through the global description vectors.

[0139] Specifically, the convolution layer is used to perform pooling processing on the candidate speech spatial features to obtain speech pooling features as shown in formula (4):

[0140]

[0141] wherein T and D are height and width of the candidate speech spatial feature respectively, z is the speech pooling feature, and X"(i,j) is a value of the input candidate speech spatial feature at position (i,j).

[0142] In step S902 of some embodiments, the normalization processing is also the excitation stage of the SE module, and the speech pooling feature is a global description vector of each spatial channel. By normalizing the global description vector, a preset weight of each spatial channel is generated, and the preset weight represents the importance of each spatial channel.

[0143] Specifically, the preset weight obtained by normalizing the speech pooling feature through the convolution layer is shown in formula (5):

[0144] S = F ex (z) = softmax(z) (5)

[0145] wherein softmax is a normalization function.

[0146] In step S903 of some embodiments, at least two selected speech spatial features are obtained by weighting and summing the candidate speech spatial features and the preset weight, i.e., weighting different spatial channels of the candidate speech spatial features according to the preset weight, and then summing the at least two selected speech spatial features to obtain the target speech spatial feature.

[0147] For example, the preset weight S is obtained through the above formula (4) and (5), and the preset weight is assigned to the candidate speech spatial feature to obtain at least two selected speech spatial features as shown in formula (6):

[0148] X" c =S x X" (6)

[0149] After obtaining the selected speech spatial feature, the target speech spatial feature is obtained by summing the at least two selected speech spatial features as shown in formula (7):

[0150]

[0151] wherein X" c is the selected speech spatial feature corresponding to the cth spatial channel.

[0152] In steps S901 to S903 shown in the embodiment, the convolution layer is used to pool and normalize the candidate speech spatial feature, so that the compression and activation stages of the SE module are realized, the preset weight representing different spatial channels is obtained, and the target speech spatial feature representing more accurate spatial relationship is obtained by weighting and summing the candidate speech spatial feature through the preset weight.

[0153] Please refer to Figure 10In some embodiments, step S105 can include, but is not limited to, steps S1001-S1002:

[0154] In step S1001, the target speech timing feature and the target speech spatial feature are fused by a target speech recognition network to obtain a target speech fusion feature.

[0155] In step S1002, the target speech fusion feature is recognized by the target speech recognition network.

[0156] It should be noted that the target speech recognition network is a feedforward network, and the target speech recognition network is trained by the original speech recognition network. During the training process of the original speech recognition network, the training speech timing feature and the training speech spatial feature are learned. Therefore, the trained target speech recognition network has learned rich speech features, and can accurately recognize the target speech timing feature and the target speech spatial feature.

[0157] In step S1001 of some embodiments, the target speech timing feature and the target speech spatial feature are fused into a target speech fusion feature, so that the target speech recognition network can recognize the target speech fusion feature.

[0158] In step S1002 of some embodiments, the target speech recognition network is a feedforward network, and the feedforward network can extract effective information from the speech feature. Therefore, the feedforward network of the present embodiment can extract more effective speech information from the target speech fusion feature which fuses the target speech spatial feature and the target speech timing feature, and realize accurate speech recognition.

[0159] In steps S1001-S1002 shown in the present embodiment, the target speech timing feature and the target speech spatial feature are first fused into a target speech fusion feature, and then the target speech fusion feature is recognized, so that the speech recognition operation is more simple and accurate.

[0160] In summary, as shown in Figure 11 , the overall flowchart of the speech recognition method of the present embodiment is shown. Figure 11 The overall flowchart of the speech recognition method of the present embodiment is shown.

[0161] In step S1101, a target speech recognition model is obtained by training a preset original speech recognition model according to training speech data and a preset speech verification result.

[0162] It should be noted that the embodiment is applied to the insurance service field, and voice data is entered as training voice data in advance according to preconfigured voice question and answer text through setting of an insurance service platform, and a voice question and answer sentence in the voice question and answer text corresponding to each training voice data is taken as a voice verification result. The original voice recognition model includes an original time feature extraction network, an original space feature extraction network and an original voice recognition network, training voice time sequence features are extracted from the training voice data through the original time feature extraction network, training voice space features are extracted from the training voice data through the original space feature extraction network, and the original voice recognition network is used to perform voice recognition on the training voice time sequence features and the training voice space features to obtain a voice recognition result. It should be noted that the voice recognition result and the voice verification result are used for loss calculation to obtain recognition loss data, and the recognition loss data represents recognition error of the original voice recognition model, and the original voice recognition model is adjusted in parameter through the recognition loss data. Specifically, in the original voice recognition model adjustment process, the original time feature extraction network, the original space feature extraction network and the original voice recognition network are adjusted in parameter based on the recognition loss data. It should be noted that the original voice recognition network is first adjusted in parameter, the original voice recognition network learns voice time sequence features and voice space features, and then the original voice recognition network learns richer voice features to obtain a target voice recognition network with more accurate voice recognition.

[0163] In step S1102, target voice data is obtained, and the target voice data is input into a target voice recognition model; wherein the target voice recognition model includes a target time feature extraction network, a target space feature extraction network and a target voice recognition network.

[0164] As disclosed above, the target voice recognition model is obtained after the training of the original voice recognition model is completed, and the target voice recognition model is applied to a starting process of a voice recognition function of the insurance service platform. When the insurance service platform receives a voice recognition starting instruction, voice data is collected in real time as target voice data through an audio recording device.

[0165] In some embodiments, the target voice recognition model includes a target time feature extraction network, a target space feature extraction network and a target voice recognition network, the target time feature extraction network is a self-attention network, the self-attention network can automatically learn voice features at different time points and output accurate voice time sequence features, the target space feature extraction network is a compression excitation network, the compression excitation network can enhance attention on key voice features in voice data and output accurate voice space features, and the target voice recognition network is a feedforward network, the feedforward network can extract effective voice information based on voice features and realize accurate voice recognition.

[0166] Step S1103, the target time feature extraction network is used to extract features of the target voice data, and target voice time sequence features are obtained.

[0167] It should be noted that the target time feature extraction network is a self-attention network, and the self-attention network includes a feature extraction layer and a self-attention layer. The feature extraction layer is also called a linear transformation layer. The target voice data is mapped to a space corresponding to a query, a key, and a value through the linear transformation layer to obtain preliminary voice time sequence features. Then, the self-attention layer is used to perform self-attention calculation on the preliminary voice time sequence features to obtain self-attention weights representing the importance of each feature in the preliminary voice time sequence features. Finally, the preliminary voice time sequence features are weighted and calculated based on the self-attention weights to obtain target voice time sequence features representing more accurate time sequence relationships.

[0168] Step S1104, the target space feature extraction network is used to extract features of the target voice data, and target voice space features are obtained.

[0169] It should be noted that the target space feature extraction network is an excitation compression network, and the excitation compression network includes a linear layer and a convolution layer. The linear layer is used to map the target voice data to different space channels to obtain at least two initial voice space features. Then, the linear layer is used to expand one dimension of each initial voice space feature to obtain preliminary voice space features. Therefore, the at least two preliminary voice space features after the dimension expansion are spliced to obtain candidate voice space features, and the candidate voice space features are voice features after the dimension change. Further, the convolution layer is used to perform pooling processing on the candidate voice space features to obtain voice pooling features representing global description vectors, and the global description vectors represent global information of each space channel. Therefore, the global description vectors are normalized to obtain preset weights representing the importance of each space channel. Finally, the preset weights and the candidate voice space features are weighted and summed to obtain target voice space features representing more accurate spatial relationships.

[0170] Step S1105, the target voice recognition network is used to perform voice recognition on the target voice time sequence features and the target voice space features.

[0171] As disclosed above, after the target voice time sequence features and the target voice space features are extracted, the voice recognition is performed in combination with the target voice time sequence features and the target voice space features. In this way, the voice recognition process can better understand the voice features of the voice data, and the accuracy of voice recognition is improved.

[0172] It should be noted that after the voice recognition is completed, feedback data of the voice recognition performed by the target voice recognition network is received, and the target voice recognition network is updated based on the feedback data to improve the accuracy of the voice recognition performed by the target voice recognition network.

[0173] Please refer toFigure 12 The embodiment of the present application further provides a speech recognition device, which can implement the speech recognition method, and the device comprises:

[0174] The data acquisition module 1201 is configured to acquire target speech data.

[0175] The speech data processing module 1202 is configured to input the target speech data into a preset target speech recognition model, wherein the target speech recognition model comprises a target time feature extraction network, a target space feature extraction network and a target speech recognition network.

[0176] The time sequence feature extraction module 1203 is configured to perform feature extraction on the target speech data through the target time feature extraction network to obtain target speech time sequence features.

[0177] The space feature extraction module 1204 is configured to perform feature extraction on the target speech data through the target space feature extraction network to obtain target speech space features.

[0178] The speech recognition module 1205 is configured to perform speech recognition on the target speech time sequence features and the target speech space features through the target speech recognition network.

[0179] The specific implementation of the speech recognition device is basically the same as the specific embodiment of the speech recognition method, and thus will not be described here.

[0180] The embodiment of the present application further provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the speech recognition method when executing the computer program. The computer device can be any intelligent terminal, such as a tablet computer or a vehicle-mounted computer.

[0181] Please refer to Figure 13 , Figure 13 The hardware structure of the computer device of another embodiment is illustrated, and the computer device comprises:

[0182] The processor 1301 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit) or one or more integrated circuits, and is configured to execute a related program to implement the technical solutions provided by the embodiments of the present application.

[0183] The memory 1302 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), etc. The memory 1302 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the related program codes are stored in the memory 1302 and are called and executed by the processor 1301 to implement the voice recognition method of the embodiments of the present application.

[0184] The input / output interface 1303 is configured to realize information input and output.

[0185] The communication interface 1304 is configured to realize the communication interaction between the device and other devices. The communication can be realized by a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0186] The bus 1305 is configured to transmit information between various components (for example, the processor 1301, the memory 1302, the input / output interface 1303, and the communication interface 1304) of the device.

[0187] The processor 1301, the memory 1302, the input / output interface 1303, and the communication interface 1304 are connected to each other through the bus 1305 to realize the communication connection between the device.

[0188] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the voice recognition method.

[0189] The memory is a non-transitory computer readable storage medium, which can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, for example, at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor. These remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0190] The voice recognition method and device, the computer device, and the storage medium provided by the embodiments of the present application can learn more voice features of voice data by extracting target voice time sequence features and target voice space features from target voice data respectively and jointly performing voice recognition on the target voice time sequence features and the target voice space features. Based on more voice features, more accurate voice recognition can be performed.

[0191] The embodiments described in the present application are for more clearly illustrating the technical solutions of the present application, and do not constitute a limitation on the technical solutions provided by the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the present application are also applicable to similar technical problems.

[0192] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the present application, and can include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0193] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, that is, can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment.

[0194] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the function modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0195] The terms "first", "second", "third", "fourth" and the like (if any) in the specification of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0196] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the front and rear associated objects. "At least one of the following" or the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c can be single or multiple.

[0197] In several embodiments provided in the application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the above units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed mutual ones can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0198] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0199] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0200] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0201] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A voice recognition method, characterized by, The method comprises: acquiring target voice data; inputting the target voice data into a preset target voice recognition model; wherein the target voice recognition model comprises a target time feature extraction network, a target space feature extraction network, and a target voice recognition network; extracting features of the target voice data through the target time feature extraction network to obtain target voice time sequence features; extracting features of the target voice data through the target space feature extraction network to obtain target voice space features; performing voice recognition on the target voice time sequence features and the target voice space features through the target voice recognition network; the target space feature extraction network is a compressed excitation network, and the compressed excitation network comprises a linear layer and a convolution layer; the target voice space features are obtained by extracting features of the target voice data through the target space feature extraction network, comprising: extracting features of the target voice data through the linear layer to obtain at least two initial voice space features; wherein each initial voice space feature represents voice features on a space channel; splicing at least two initial voice space features through the linear layer to obtain candidate voice space features; performing weighted calculation on the candidate voice space features through the convolution layer to obtain the target voice space features.

2. The method of claim 1, wherein, the splicing of at least two initial voice space features through the linear layer to obtain candidate voice space features comprises: performing dimension expansion on each initial voice space feature through the linear layer to obtain preliminary voice space features; splicing at least two preliminary voice space features through the linear layer to obtain the candidate voice space features.

3. The method of claim 1, wherein, the weighted calculation of the candidate voice space features through the convolution layer to obtain the target voice space features comprises: performing pooling processing on the candidate voice space features through the convolution layer to obtain voice pooling features; performing normalization processing on the voice pooling features through the convolution layer to obtain a preset weight of each space channel; performing weighted summation on the candidate voice space features and the preset weight through the convolution layer to obtain the target voice space features.

4. The method according to any one of claims 1 to 3, characterized in that, the time feature extraction network is a self-attention network, and the self-attention network comprises a feature extraction layer and a self-attention layer; the target voice time sequence features are obtained by extracting features of the target voice data through the target time feature extraction network, comprising: extracting features of the target voice data through the feature extraction layer to obtain preliminary voice time sequence features; performing self-attention calculation on the preliminary voice time sequence features through the self-attention layer to obtain self-attention weights; performing weighted calculation on the preliminary voice time sequence features according to the self-attention weights to obtain the target voice time sequence features.

5. The method according to any one of claims 1 to 3, characterized in that, the voice recognition on the target voice time sequence features and the target voice space features through the target voice recognition network comprises: The target speech fusion feature is obtained by performing feature fusion on the target speech time sequence feature and the target speech space feature through the target speech recognition network. The target speech recognition network is used to perform speech recognition on the target speech fusion feature.

6. The method according to any one of claims 1 to 3, characterized in that, Before the target speech data is input into the preset target speech recognition model, the method further comprises: The target speech recognition model is constructed, specifically comprising: obtaining training speech data; The training speech data is input into a preset original speech recognition model; wherein the original speech recognition model comprises an original time feature extraction network, an original space feature extraction network and an original speech recognition network: The training speech data is feature-extracted through the original time feature extraction network to obtain training speech time sequence features; The training speech data is feature-extracted through the original space feature extraction network to obtain training speech space features; The original speech recognition network is used to perform speech recognition on the training speech time sequence features and the training speech space features to obtain a speech recognition result; The original speech recognition model is parameter-adjusted according to a preset speech verification result and the speech recognition result to obtain the target speech recognition model.

7. A speech recognition apparatus characterized by comprising: The device comprises: a data acquisition module configured to acquire target speech data; a speech data processing module configured to input the target speech data into a preset target speech recognition model; wherein the target speech recognition model comprises a target time feature extraction network, a target space feature extraction network and a target speech recognition network; a time sequence feature extraction module configured to extract features of the target speech data through the target time feature extraction network to obtain target speech time sequence features; a space feature extraction module configured to extract features of the target speech data through the target space feature extraction network to obtain target speech space features; a speech recognition module configured to perform speech recognition on the target speech time sequence features and the target speech space features through the target speech recognition network; The target space feature extraction network is a compression excitation network, which comprises a linear layer and a convolution layer; the target speech space features are obtained by extracting features of the target speech data through the target space feature extraction network, which comprises: extracting features of the target speech data through the linear layer to obtain at least two initial speech space features; wherein each initial speech space feature represents speech features on a space channel; splicing at least two initial speech space features through the linear layer to obtain candidate speech space features; performing weighted calculation on the candidate speech space features through the convolution layer to obtain the target speech space features.

8. A computer device, comprising: The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the speech recognition method in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the speech recognition method in any one of claims 1 to 6.

Images