Hospital guidance order classification processing method and device, equipment and storable medium

Abstract

The invention provides a hospital guide order classification processing method and device, computer equipment and a computer readable storage medium, and belongs to the field of artificial intelligence and the field of digital medical treatment. The hospital guide order classification processing method comprises the steps of obtaining target voice and target text of a target hospital guide order; through a target audio coding layer in a feature extraction module, target voice is decoupled to obtain a decoupled audio feature of the target hospital guide order; through a target text coding layer in a feature extraction module, decoupling a target text to obtain a decoupled text feature of the target hospital guide order; and performing classification processing based on the decoupling text features through a trained classification module to obtain a classification recommendation result of the target hospital guide order. The diagnosis guide order data is analyzed through machine learning in combination with the audio data and the text data, the generality information of the text and the voice is captured through feature decoupling, the accuracy of classification recommendation of the diagnosis guide order is improved, and then the diagnosis guide order processing efficiency is improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and digital medicine, and specifically to a method, device, computer equipment and computer-readable storage medium for classifying and processing medical guide orders. Background Art

[0002] With the development of the times, the application of artificial intelligence technology is becoming more and more extensive. Artificial intelligence technology is used to classify complex data so that the classification results can be used for further data processing. For example, in the medical field, medical guide orders can be classified to recommend them to the corresponding department for processing.

[0003] However, the inventors of the present application discovered during the actual research and development process that the relevant technology, in order to improve the classification accuracy of the classification task, mainly extracts text features separately for text information and auditory features separately for voice information, and then directly concatenates the text features and auditory features for classification. However, the direct concatenation of text features and auditory features is difficult to capture information that is more conducive to classification, which leads to a decrease in classification accuracy. Summary of the invention

[0004] The present application provides a method, device, computer equipment and computer-readable storage medium for classifying and processing medical guide orders, which belong to the fields of artificial intelligence and digital medicine, and can improve the accuracy of medical guide order classification results.

[0005] In a first aspect, the present application provides a method for classifying and processing medical guide orders, comprising:

[0006] Obtaining a target voice of a target medical guidance order and a target text of the target medical guidance order;

[0007] The target speech is decoupled through the target audio coding layer in the feature extraction module to obtain the decoupled audio features of the target medical guidance order, wherein the target audio coding layer is obtained by extracting the audio coding layer of the trained extraction module, and the trained extraction module is obtained by training the preset extraction module based on the feature loss between text and speech of the sample medical guidance order;

[0008] The target text is decoupled through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order, wherein the target text encoding layer is obtained by extracting the text encoding layer of the trained extraction module;

[0009] Through the trained classification module, classification processing is performed based on the decoupled audio features and the decoupled text features to obtain a classification recommendation result for the target medical guidance order.

[0010] In a second aspect, the present application provides a device for classifying and processing medical guidance orders, comprising:

[0011] An acquisition unit, used to acquire a target voice of a target medical guidance order and a target text of the target medical guidance order;

[0012] A decoupling unit is used to decouple the target speech through a target audio coding layer in a feature extraction module to obtain a decoupled audio feature of the target medical guidance order, wherein the target audio coding layer is obtained by extracting the audio coding layer of a trained extraction module, and the trained extraction module is obtained by training a preset extraction module based on a feature loss between text and speech of a sample medical guidance order;

[0013] The decoupling unit is further used to perform decoupling processing on the target text through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order, wherein the target text encoding layer is obtained based on the extraction processing of the text encoding layer of the trained extraction module;

[0014] A classification unit is used to perform classification processing based on the decoupled audio features and the decoupled text features through a trained classification module to obtain a classification recommendation result of the target medical guidance order.

[0015] In a third aspect, the present application also provides a computer device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method for classifying and processing medical guide orders when executing the computer program.

[0016] In a fourth aspect, the present application also provides a computer-readable storage medium on which a computer program is stored, and the computer program is loaded by a processor to execute the method for classifying and processing medical guide orders.

[0017] In the present application, a preset extraction module is trained using the feature loss between the text and speech of a sample medical guidance order to obtain a trained extraction module, so that there is more potential shared information between the features extracted by the trained extraction module from the sample text of the sample medical guidance order and the features extracted from the sample speech of the sample medical guidance order, and then a feature extraction module is obtained by extracting the trained extraction module. As a result, the feature extraction module decouples the target speech of the target medical guidance order and the target text of the target medical guidance order, and can more specifically extract potential common information in the speech data and text data, and these common information contain potential medical information. Therefore, the decoupled audio features and decoupled text features obtained by decoupling can more specifically capture the medical information in the target medical guidance order, effectively improving the accuracy of the classification recommendation of the medical guidance order, thereby improving the efficiency of the medical guidance order processing, and providing support for the subsequent development of medical auxiliary technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a schematic diagram of the architecture of the medical guide order classification processing system provided in an embodiment of the present application;

[0020] Figure 2 This is a flow chart of a method for classifying and processing medical guide orders provided in an embodiment of the present application;

[0021] Figure 3 This is a schematic diagram of an implementation process of the method for classifying and processing medical guide orders provided in an embodiment of the present application;

[0022] Figure 4 It is a schematic diagram of a training process of a preset extraction module provided in an embodiment of the present application;

[0023] Figure 5 It is a flow chart of a preset extraction module through text-speech feature loss training of a sample medical guidance order provided in an embodiment of the present application;

[0024] Figure 6 It is a schematic diagram of the structure of an embodiment of a device for classifying and processing medical guidance orders provided in an embodiment of the present application;

[0025] Figure 7 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The embodiments of the present application provide a method, apparatus, computer device and computer-readable storage medium for classifying and processing medical guide orders, which comprehensively analyze the voice data and text data of a target medical guide order, decouple the target voice of the target medical guide order and the target text of the target medical guide order, capture the common information of the target text and the target voice, and use the decoupled audio features and decoupled text features obtained by decoupling to perform classification recommendations, thereby effectively improving the accuracy of the classification recommendations of medical guide orders and thereby improving the efficiency of medical guide order processing.

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0029] In the description of the embodiments of the present application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0030] In order to enable any person skilled in the art to implement and use the present application, the following description is provided. In the following description, details are listed for the purpose of explanation. It should be understood that those of ordinary skill in the art can recognize that the present application can also be implemented without using these specific details. In other examples, the known process will not be elaborated in detail to avoid unnecessary details that make the description of the present application embodiment obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the widest range of principles and features disclosed in accordance with the embodiments of the present application.

[0031] The embodiment of the present application provides a method, apparatus, computer device and computer-readable storage medium for classifying and processing medical guidance orders. The computer device may be a smart phone, tablet computer, laptop computer, desktop computer, wearable electronic device, portable calculator and other smart devices.

[0032] Some embodiments of the present application are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the embodiments and features in the embodiments provided by the present application can be combined with each other.

[0033] With the development of science and technology, artificial intelligence technology is increasingly widely used in social life. Voice contains a variety of information such as semantics, logic, emotions, and physiology. Through the analysis of voice data by artificial intelligence technology, the key features in the data can be more accurately extracted, which improves the experience of human-computer interaction and promotes the development of fields such as shopping recommendations, intelligent question and answer, and digital medicine.

[0034] In a medical guidance application scenario, terminal device A uploads a guidance order data to server B, and the server distributes it to the terminal device C of the corresponding department. In order to better classify the guidance order in the above scenario, this application provides a guidance order classification processing method, which can extract the voice data and text data of the guidance order for intelligent classification, saving manpower and material resources.

[0035] Based on this, see Figure 1 , Figure 1 This is a schematic diagram of the architecture of the medical guide order classification processing system provided in the embodiment of the present application. The medical guide order classification processing method provided in the embodiment of the present application can be applied to Figure 1 The medical consultation order classification processing system shown in FIG. Figure 1 As shown, the medical guide order classification processing system includes a terminal device and a server. The server involved in this application can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud storage, and big data and artificial intelligence platforms. The terminal device can be a computer device such as a smart phone, a tablet computer, a laptop computer, a smart bracelet, etc., but is not limited to this. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here. There is no limit on the number of servers and terminal devices.

[0036] Combination Figure 1 The architecture shown in the figure, a process of the medical guidance order classification processing system provided by the embodiment of the present application includes steps S1 to S6, wherein:

[0037] S1. Terminal device A uploads a target medical guidance order to server B, and server B extracts the target voice of the target medical guidance order.

[0038] S2. Server B extracts signal features of the target speech to obtain acoustic features to be classified of the target speech.

[0039] S3. Server B performs speech recognition processing on the acoustic features of the target speech to be classified to obtain the target text of the target medical guidance order.

[0040] S4. Server B performs vectorization processing on the target text to obtain the text features to be classified of the target text.

[0041] S5. Server B performs feature decoupling on the acoustic features to be classified and the text features to be classified to obtain decoupled audio features and decoupled text features.

[0042] S6. Based on the decoupled audio features and the decoupled text features, the classification recommendation results of the target medical guidance order are determined. Further, server B distributes the target medical guidance order to the terminal device C of the corresponding department according to the classification recommendation results.

[0043] The execution subject of the medical guide order classification processing method provided in the embodiment of the present application may be the medical guide order classification processing device provided in the embodiment of the present application, or the computer device provided in the embodiment of the present application, or Figure 1 The server B of the medical guidance order classification processing system shown in the figure, wherein the medical guidance order classification processing device can be implemented in the form of hardware or software.

[0044] The following is an introduction to the classification and processing method of the medical guidance order provided by this application. Figure 2 , Figure 2 101 is a flowchart of a method for classifying and processing medical guide orders provided in an embodiment of the present application. The method for classifying and processing medical guide orders includes steps 101 to 104, wherein:

[0045] 101. Obtain a target voice and a target text of a target medical guidance order.

[0046] Among them, the target medical guidance orders refer to the medical guidance orders to be classified and recommended.

[0047] There are many forms of target speech, including:

[0048] (1) In some embodiments, the target medical guidance order is stored in a preset database. In this case, step 101 may specifically include: directly reading the voice data of the target medical guidance order in the preset database.

[0049] (2) In some embodiments, the target voice is the original voice input by the user in real time. In this case, step 101 may specifically include: obtaining the original voice input by the user in real time as the target voice for the target medical guidance order.

[0050] (3) In some embodiments, the target speech is the speech obtained after the original speech input by the user is processed by noise reduction, tuning, compression, etc. In this case, step 101 may specifically include: obtaining the original speech input by the user; performing noise reduction and compression on the original speech input by the user to obtain the processed speech, and using the processed speech as the target speech.

[0051] The target text may be expressed in a variety of forms, including, for example:

[0052] (1) In some embodiments, the target text is text data stored in a preset database of target medical guidance orders. In this case, step 101 may specifically include: directly reading the text data of the target medical guidance order in the preset database.

[0053] (2) In some embodiments, the target text is the target text obtained by converting the target speech into the target speech through speech recognition technology (Automatic Speech Recognition, ASR), wherein speech recognition is an artificial intelligence technology that converts human voice into text. The current mainstream speech recognition models include Hidden Markov Model (HMM), Transformer-based models, Deep Neural Networks (DNN), etc. The specific speech recognition method and model can be selected according to the actual situation. At this time, step 101 can specifically include: obtaining the target speech of the target medical guidance order, and converting the target speech into the target text through the speech recognition model.

[0054] Since speech not only contains text information, but also contains information such as timbre and rhythm, for example, the speech data and text data from the same medical guide order contain the same or similar key information, and the semantics, timbre, rhythm and other information of the same user can express the user's medical preferences. This application describes and analyzes the same information from different perspectives, which can make the information analysis more comprehensive and rich, thereby improving the accuracy of the classification recommendation results of the medical guide order.

[0055] 102. Decouple the target speech through the target audio coding layer in the feature extraction module to obtain the decoupled audio features of the target medical guidance order.

[0056] The target audio coding layer is obtained by extracting the audio coding layer of the trained extraction module, and the trained extraction module is obtained by training the preset extraction module using the text-speech feature loss of the sample guidance order. The decoupled audio feature is a feature vector extracted from the target speech containing content potentially shared with the target text; the sample guidance order is a set of guidance order data used to train the preset extraction module; the text-speech feature loss is the feature loss between the text-related features of the sample text extracted from the sample guidance order and the spectrum-related features of the sample speech; the text-related features of the sample text are feature vectors containing content information potentially shared with the sample speech; the spectrum-related features of the sample speech are feature vectors containing content information potentially shared with the sample text.

[0057] In some embodiments, the target audio coding layer can be trained by referring to steps 201 to 205 below. For details, please refer to the relevant description below. To simplify the description, it will not be repeated here.

[0058] Exemplarily, firstly, the target speech is vectorized to obtain the acoustic features to be classified of the target speech; then, the acoustic features to be classified of the target speech are input into the target audio coding layer in the feature extraction module for feature decoupling. The acoustic features to be classified may be Mel Frequency Cepstral Coefficients (MFCC) features, Filter Bank (FBank) features, Linear Predictive Coding (LPC) features, and the like.

[0059] Exemplarily, the process of "vectorizing the target speech and obtaining the acoustic features to be classified of the target speech" can be as follows: pre-emphasize, frame and window the speech signal of the target speech to make the speech signal suitable for Fourier transform; apply the fast Fourier algorithm to each short-time window to obtain the corresponding spectrum; calculate each spectrum through a Mel filter, and output the Mel spectrum as the acoustic features to be classified of the target speech. In the embodiment of the present application, by converting the original audio signal into a Mel spectrum for subsequent processing, the detailed information in the audio data is retained, the complexity of the data is reduced, and it is beneficial to improve the accuracy of the classification recommendation of subsequent medical guidance orders, thereby improving the feasibility of the solution.

[0060] There are many specific implementation methods for step 102, which include, for example:

[0061] (1) In some embodiments, the target audio coding layer in the feature extraction module includes a spectrum association encoder and a spectrum private encoder. The spectrum association encoder in the target audio coding layer is used to extract the decoupled audio features of the target medical guidance order; and the spectrum private encoder in the target audio coding layer is used to extract the target spectrum private features containing the audio domain attribute information of the target speech for subsequent auxiliary classification.

[0062] (2) In some embodiments, the target audio coding layer in the feature extraction module includes a spectrum correlation encoder, and the spectrum correlation encoder in the target audio coding layer is used to decouple the target speech to obtain the decoupled audio features of the target medical guidance order.

[0063] In the embodiment of the present application, since the feature extraction module is obtained by extracting the trained extraction module, and the trained extraction module is obtained by training the preset extraction module using the text-speech feature loss of the sample medical guidance order, the decoupled audio features obtained by the feature extraction module contain information that is potentially associated with the decoupled text features, thereby reducing the impact of differences in different types of data, thereby more accurately extracting key feature information of the target medical guidance order, which is beneficial to improving the classification accuracy of the target medical guidance order and improving the efficiency of medical assistance work.

[0064] 103. Decoupling the target text is performed through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order.

[0065] The target text encoding layer is obtained by extracting the text encoding layer of the trained extraction module. The decoupled text feature is a feature vector extracted from the target text that contains potential shared content with the target speech.

[0066] In some embodiments, the target text encoding layer can be trained by referring to steps 201 to 205 below. For details, please refer to the relevant instructions below. For the sake of simplicity, they will not be repeated here.

[0067] Exemplarily, first, the target text is vectorized to obtain the text features to be classified of the target text; then, the text features to be classified of the target text are input into the target text encoding layer in the feature extraction module for feature decoupling.

[0068] Exemplarily, the process of "vectorizing the target text to obtain the text features to be classified of the target text" can be as follows: the target text is converted into text features to be classified using the Doc2vec algorithm. Doc2vec can represent texts of different lengths as feature vectors of specific lengths. More specifically, the Distributed Memory Model of Paragraph Vectors (PV-DM) model is used to slide samples from the context of the paragraphs of the target text, using sentence vectors and word vectors as input, and outputting fixed-length paragraph vectors as the text features to be classified of the target text. In the embodiment of the present application, by vectorizing the text for subsequent processing, the detailed information in the text data is retained, which is conducive to improving the accuracy of the classification recommendation of subsequent medical guidance orders, thereby improving the feasibility of the solution.

[0069] There are many specific implementations of step 103, which include, for example:

[0070] (1) In some embodiments, the target text encoding layer in the feature extraction module includes a text association encoder and a text private encoder. The decoupled text features of the target medical guidance order are extracted by the text association encoder in the target text encoding layer; and the target text private features containing the text domain attribute information of the target text are extracted by the text private encoder in the target text encoding layer for use in subsequent auxiliary classification.

[0071] (2) In some embodiments, the target text encoding layer in the feature extraction module includes a text association encoder, and the target text is decoupled by the text association encoder in the target text encoding layer to obtain the decoupled text features of the target medical guidance order.

[0072] In the embodiment of the present application, the feature extraction module obtained by training the preset extraction module using the text-speech feature loss of the sample medical guidance order can more specifically extract the shared information of the text domain and the spectral domain in the target medical guidance order. Therefore, the decoupled text features obtained by decoupling the target text through the feature extraction module contain content potentially shared with the audio domain. The present application analyzes the medical guidance order data from the perspectives of the text domain and the audio domain, and effectively extracts the key features in the target medical guidance order by using the potential shared information between different types of data, reducing the impact of the differences between different types of data, which is conducive to improving the accuracy of subsequent classification processing.

[0073] 104. Through the trained classification module, classification processing is performed based on the decoupled audio features and the decoupled text features to obtain the classification recommendation results of the target medical guidance order.

[0074] Exemplarily, the classification module may include using a convolutional neural network model to perform convolution pooling operations on an input feature vector and output a classification result.

[0075] The training process of the classification module may include: performing feature extraction on the sample medical guidance order based on the trained extraction module, performing feature fusion on the text-related features and spectrum-related features of the sample medical guidance order to obtain sample fusion features, and using the sample fusion feature set as the training data set to train a preset classification module to obtain a trained classification module.

[0076] Exemplarily, the classification result can be represented as a numerical value distributed in a discrete interval, for example, 1 or 2. If the score is 1, the classification recommendation result is department 1. If the score is 2, the classification recommendation result is department 2, and so on. There is no limitation here.

[0077] There are many ways to implement step 104, and by way of example, they may include:

[0078] (1) In some embodiments, please refer to Figure 3 , Figure 3 It is a schematic diagram of an implementation process of the method for classifying and processing medical guide orders provided in an embodiment of the present application.

[0079] The decoupled audio features and decoupled text features are vectorized and concatenated, and the autoencoder is used to perform feature fusion to obtain the fused features of the target medical guidance order.

[0080] Through the trained classification module, classification processing is performed based on the fusion features to obtain the classification recommendation results of the target medical guidance orders.

[0081] (2) In some embodiments, auxiliary text and auxiliary audio are generated based on decoupled audio features and decoupled text features through an auxiliary information generation layer in a feature extraction module, wherein the auxiliary information generation layer in the feature extraction module is obtained by extracting the generation layer of the trained extraction module.

[0082] Among them, the auxiliary information generation layer of the feature extraction module includes a text generator and an audio generator. The method of generating auxiliary text and generating auxiliary audio is similar to the method of generating text and audio by the generation layer in the extraction module preset in the later training. For details, please refer to the relevant instructions later, which will not be repeated here.

[0083] Through the target text encoding layer in the feature extraction module, feature decoupling processing is performed based on the auxiliary text to obtain the auxiliary text features.

[0084] Through the target audio coding layer in the feature extraction module, feature decoupling processing is performed based on the auxiliary audio to obtain the auxiliary audio features.

[0085] At this time, step 104 may include steps 104A-104B:

[0086] 104A. Fusing the decoupled audio features and the decoupled text features to obtain fused features of the target medical guidance order.

[0087] 104B. Through the trained classification module, classification processing is performed based on the fusion features, auxiliary text features, and auxiliary audio features to obtain the classification recommendation results of the target medical guidance order.

[0088] Exemplarily, step 104B may include: respectively inputting fusion features, auxiliary text features, and auxiliary audio features through a trained classification module, performing weighted summation on the output scores, and obtaining a classification recommendation result for the target medical guidance order.

[0089] Compared with the method of extracting features for text information and voice information separately, the decoupled audio features and decoupled text features extracted in the present application contain potential shared information of the target text and target voice, and perform feature fusion based on the decoupled audio features and decoupled text features, thereby achieving the complementarity of text information and audio information, and capturing the characteristics of the target medical guidance orders in a more targeted manner, thereby improving the accuracy of classification recommendations for medical guidance orders.

[0090] In an embodiment of the present application, by obtaining the target voice and target text of the target medical guidance order, a comprehensive analysis of the target medical guidance order is performed from the two perspectives of text domain and audio domain; through the feature extraction module, decoupling processing is performed based on the target voice and target text to obtain decoupled audio features and decoupled text features, which effectively reduces the impact of different data types of text and audio on feature fusion, and more efficiently extracts potential correlation features between text information and audio information; through the trained classification module, feature fusion is performed based on the decoupled audio features and decoupled text features and then classification processing is performed to obtain the classification recommendation results of the target medical guidance order, which helps to assist medical staff more efficiently, provide better quality medical guidance services, and improve work efficiency.

[0091] For some examples, see Figure 4 , Figure 4 201 is a schematic diagram of a training process of a preset extraction module provided in an embodiment of the present application. The training process of the audio coding layer of the feature extraction module and the text coding layer of the feature extraction module includes steps 201 to 205, which are as follows:

[0092] 201. Through the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the sample text of the sample medical guidance order to obtain a text feature group.

[0093] The sample medical guidance order is a set of medical guidance order data used for training, and the sample text is text data obtained based on the sample voice through voice recognition technology.

[0094] Exemplarily, the text encoding layer in the preset extraction module includes a text-private encoder and a text-associated encoder. The text-associated encoder is used to extract text-associated features that express similar information to the audio domain. The text-private encoder is used to extract text-private features that are private to the text domain attributes. The text feature group includes text-associated features and text-private features.

[0095] The text association encoder and the text private encoder can adopt a convolutional neural network with the same structure, including a one-dimensional convolution layer, a pooling layer, and a fully connected layer. Therefore, step 201 specifically includes: vectorizing the sample text to obtain the text features to be classified of the sample text; inputting the text features to be classified of the sample text into the text association encoder to obtain text association features; inputting the text features to be classified of the sample text into the text private encoder to obtain text private features.

[0096] It should be noted that the specific method of obtaining the to-be-classified text features of the sample text can be referred to in the above step 103, and will not be described again here for simplicity.

[0097] 202. Perform feature decoupling processing based on the sample voice of the sample medical guidance order through the audio coding layer in the preset extraction module to obtain a spectrum feature group.

[0098] Exemplarily, the audio coding layer in the preset extraction module includes a spectrum-private encoder and a spectrum-association encoder. The spectrum-association encoder is used to extract spectrum-association features that express similar information to the text domain. The spectrum-private encoder is used to extract spectrum-private features that are private to the audio domain attributes. The spectrum feature group includes spectrum-association features and spectrum-private features.

[0099] The spectrum association encoder and the spectrum private encoder can adopt a convolutional neural network with the same structure, including a two-dimensional convolution layer, a pooling layer, and a fully connected layer. Therefore, step 202 specifically includes: extracting signal features of the sample speech to obtain acoustic features to be classified of the sample speech; inputting the acoustic features to be classified of the sample speech into the spectrum association encoder to obtain spectrum association features; inputting the acoustic features to be classified of the sample speech into the spectrum private encoder to obtain spectrum private features.

[0100] It should be noted that the specific method of obtaining the acoustic features to be classified of the sample speech can be referred to in the above step 102, and will not be repeated here for the sake of simplicity.

[0101] 203. Based on the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss, a preset extraction module is trained to obtain a trained extraction module.

[0102] In some embodiments, see Figure 5 , Figure 5 This is a flow chart of a preset extraction module provided by the embodiment of the present application through the text-speech feature loss training of the sample medical guidance order. Step 203 may specifically include the following steps 203A to 203E:

[0103] 203A. Obtain the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss.

[0104] 203B. Perform feature cross processing based on the text feature group and the spectrum feature group through the generation layer in the preset extraction module to obtain comparison text and comparison audio.

[0105] In some embodiments, step 203B may specifically include the following steps A1 to A4:

[0106] A1. Through the generation layer in the preset extraction module, the intermediate text and the intermediate audio of the sample medical guidance order are obtained based on the text feature group and the spectrum feature group.

[0107] Among them, the intermediate text contains the text domain attribute information of the sample medical guidance order, and the content information extracted from the sample voice and shared with the text domain; the intermediate audio contains the audio domain attribute information of the sample medical guidance order, and the content information extracted from the sample text and shared with the audio domain.

[0108] In some embodiments, step A1 may specifically include the following steps B1-B2:

[0109] B1. Generate the intermediate text of the sample medical guidance order based on the text private features of the text feature group and the spectrum association features of the spectrum feature group through the text generator in the generation layer in the preset extraction module.

[0110] B2. Generate the intermediate audio of the sample medical guidance order based on the spectrum private features of the spectrum feature group and the text association features of the text feature group through the audio generator in the generation layer in the preset extraction module.

[0111] Exemplarily, a trained long short-term memory network model (LSTM) is used as a text generator; a trained long short-term memory network model is used as a spectrum generator. It is understandable that the text generator can also be a statistically based model, a neural network language model, etc., and the spectrum generator can also be a recurrent neural network (RNN), a Transformer, etc., which are not limited here.

[0112] In an embodiment of the present application, intermediate audio and intermediate text are generated through steps B1 to B2 for subsequent training, thereby improving the accuracy of the extraction module after training, and then enabling the feature extraction module to capture the potentially shared features of the text data and voice data in the target medical guide order, thereby improving the accuracy of the feature extraction module in decoupling key features, making the classification results of the target medical guide order more accurate, and improving the feasibility of the solution.

[0113] A2. Through the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the intermediate text to obtain the intermediate text feature group.

[0114] Among them, the intermediate text feature group includes intermediate text associated features and intermediate text private features. The intermediate text associated features are feature vectors extracted from the intermediate text and contain content information potentially associated with the audio domain. The intermediate text private features are feature vectors extracted from the intermediate text and contain attribute information of the text domain.

[0115] Specifically, step A2 may include: inputting the intermediate text of the sample medical guidance order into a text association encoder to obtain intermediate text association features; inputting the intermediate text of the sample medical guidance order into a text private encoder to obtain intermediate text private features.

[0116] A3. Perform feature decoupling processing based on the intermediate audio through the audio coding layer in the preset extraction module to obtain an intermediate spectrum feature group.

[0117] Among them, the intermediate spectrum feature group includes intermediate spectrum associated features and intermediate spectrum private features. The intermediate spectrum associated features are feature vectors extracted from the intermediate audio and contain content information potentially associated with the text domain. The intermediate spectrum private features are feature vectors extracted from the intermediate audio and contain attribute information of the audio domain.

[0118] Specifically, step A3 may include: inputting the intermediate audio of the sample medical guidance order into a spectrum association encoder to obtain intermediate spectrum association features; inputting the intermediate audio of the sample medical guidance order into a spectrum private encoder to obtain intermediate spectrum private features.

[0119] A4. Through the generation layer in the preset extraction module, generation processing is performed based on the intermediate text feature group and the intermediate spectrum feature group to obtain the comparison text and the comparison audio.

[0120] Among them, the comparison text contains the attribute information of the text domain in the sample medical guidance order, and the content information associated with the text domain and the audio domain in the sample medical guidance order; the comparison audio contains the attribute information of the audio domain in the sample medical guidance order, and the content information associated with the audio domain and the text domain in the sample medical guidance order.

[0121] In the embodiment of the present application, text information and voice information are used to complement each other, and comparative text and comparative audio are generated through steps A1 to A4 for subsequent training of a preset extraction module, thereby improving the accuracy of the trained extraction module, improving the accuracy of feature decoupling of the feature extraction module, improving the accuracy of classification recommendations for target medical guidance orders, and improving the operability of the solution.

[0122] 203C. Based on the text similarity between the sample text and the comparison text of the sample medical guidance order, obtain the text cycle consistency loss of the preset extraction module.

[0123] 203D. Based on the audio similarity between the sample voice of the sample medical guidance order and the comparison audio, obtain the audio cycle consistency loss of the preset extraction module.

[0124] 203E. Based on the text-speech feature loss, the text cycle consistency loss, and the audio cycle consistency loss, the preset extraction module is trained to obtain a trained extraction module.

[0125] Since the voice data and text data of the same sample medical guidance order contain domain attribute information and content information of potential associations between different types of data, the features obtained by separating the text domain attribute information from the feature information of the sample text and the features obtained by separating the audio domain attribute information from the feature information of the sample voice express similar content information. In steps 203A to 203E, the text encoding layer of the preset extraction module is used to output text-related features and text-private features by decoupling the sample text, and the audio encoding layer of the preset extraction module is used to output spectrum-related features and spectrum-private features by decoupling the sample voice.

[0126] The more similar the text-related features and the spectrum-related features are, the more content information that is potentially associated with the audio domain is contained in the text-related features extracted from the sample text, and the more content information that is potentially associated with the text domain is contained in the spectrum-related features extracted from the sample speech. Therefore, during the training process, the feature loss between text and speech is minimized to make the extracted text-related features and spectrum-related features as similar as possible, so that the accuracy of the text encoding layer in the trained extraction module and the audio encoding layer in the trained extraction module is higher.

[0127] The more similar the comparison text generated by feature cross-combination is to the sample text, the higher the accuracy of extracting the attribute information of the text domain of the sample text as the text-private feature, and the higher the accuracy of extracting the content information potentially associated with the audio domain as the text-associated feature. Therefore, during the training process, the text cycle consistency loss is minimized to make the comparison text as similar as possible to the sample text, so that the text encoding layer in the trained extraction module can perform feature decoupling more accurately.

[0128] The more similar the comparison audio generated by feature cross-combination is to the sample speech, the higher the accuracy of extracting the attribute information of the audio domain of the sample speech as the spectrum private feature, and the higher the accuracy of extracting the content information potentially associated with the text domain as the spectrum association feature. Therefore, during the training process, the audio cycle consistency loss is minimized to make the comparison audio and the sample speech as similar as possible, so that the audio coding layer in the trained extraction module can perform feature decoupling more accurately.

[0129] Therefore, the preset extraction module can be trained based on the following loss function:

[0130]

[0131] L1=||T,T′||

[0132] L2=||M,M′||

[0133] Among them, L1 is the text cycle consistency loss, T is the sample text of the sample medical guidance order, and T′ is the comparison text obtained by inputting the preset extraction module of the sample medical guidance order; L2 is the audio cycle consistency loss, M is the sample voice of the sample medical guidance order, and M′ is the comparison audio obtained by inputting the preset extraction module of the sample medical guidance order; is the text-to-speech feature loss, The text association encoder in the text encoding layer in the preset extraction module, Indicates the text association features obtained by inputting the sample text of the sample medical guidance order into the text association encoder. The spectrum correlation encoder in the audio coding layer in the preset extraction module, It indicates that the sample speech of the sample medical guidance order is input into the spectrum association encoder, and the sample association spectrum features are obtained.

[0134] In some embodiments, step 203 may further include:

[0135] By using the text association encoder of the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the contrasting text to obtain the contrasting text features.

[0136] By using the spectrum correlation encoder of the audio encoding layer in the preset extraction module, feature decoupling processing is performed based on the contrast audio to obtain the contrast audio feature.

[0137] The loss between contrasting text features and contrasting audio features is obtained as the decoupled contrast loss.

[0138] Through the prediction layer in the preset extraction module, classification processing is performed based on contrasting text features and contrasting audio features to obtain the cross entropy loss of the preset extraction module, wherein the prediction layer is a preset convolutional neural network.

[0139] At this time, step 203E also includes:

[0140] Based on the text-speech feature loss, text cycle consistency loss, audio cycle consistency loss, decoupled contrast loss, and cross entropy loss, the preset extraction module is trained to obtain the trained extraction module.

[0141] At this point, the preset extraction module can be trained based on the following loss function:

[0142]

[0143] in, To decouple the contrast loss, Indicates the contrast text features obtained by inputting the contrast text into the text association encoder. It represents the contrast audio feature obtained by inputting the contrast audio into the spectrum correlation encoder. ce is the cross entropy loss, N is the number of sample guidance orders, J is the number of guidance classification categories, and y il represents a symbolic function, where if the true category of the i-th sample medical guidance order belongs to category l, it takes 1, otherwise it takes 0, p il It represents the probability that the prediction layer predicts that the i-th sample medical consultation order belongs to category l.

[0144] The more similar the contrasting text features and the contrasting audio features are, the more content information that is potentially related between the text domain and the audio domain is contained in the contrasting text features and the contrasting audio features, which means that the generation layer in the trained extraction module can retain more key features of the sample guidance orders when generating text data and audio data. Therefore, during the training process, the decoupling comparison loss is made as small as possible to make the contrasting text features and the contrasting audio features as similar as possible, thereby making the text encoding layer in the trained extraction module and the audio encoding layer in the trained extraction module more accurate.

[0145] During the training process, the cross entropy loss is made as small as possible so that the prediction layer of the trained extraction module can predict the true category of the sample guidance order with higher accuracy, thereby making the text encoding layer in the trained extraction module and the audio encoding layer in the trained extraction module more accurate in extracting key features.

[0146] In an embodiment of the present application, the preset extraction module is trained through step 203 to obtain a trained extraction module. The feature extraction module obtained based on the trained extraction module can more specifically capture the features containing potential related content in the text domain and audio domain of the target medical guidance order, thereby improving the accuracy of the classification recommendation results of the target medical guidance order.

[0147] 204. Performing extraction processing based on the trained audio coding layer of the extraction module to obtain a target audio coding layer of the feature extraction module;

[0148] There are many specific implementations of step 204, which include, for example:

[0149] (1) In some embodiments, the audio coding layer in the trained extraction module is extracted as the target audio coding layer of the feature extraction module.

[0150] (2) In some embodiments, the spectrum correlation encoder in the audio coding layer in the extraction module after training is extracted as the target audio coding layer of the feature extraction module.

[0151] 205. Perform extraction processing based on the text encoding layer of the trained extraction module to obtain a target text encoding layer of the feature extraction module.

[0152] There are many specific implementations of step 205, which include, for example:

[0153] (1) In some embodiments, the text encoding layer in the trained extraction module is extracted as the target text encoding layer of the feature extraction module.

[0154] (2) In some embodiments, the text association encoder in the text encoding layer in the extraction module after training is extracted as the target text encoding layer of the feature extraction module.

[0155] In an embodiment of the present application, by utilizing the feature loss between text and speech to train a preset extraction module, the text-related features extracted from the sample text of the sample medical guidance order and the spectrum-related features extracted from the sample speech of the sample medical guidance order by the preset extraction module during the training process become increasingly similar, so as to mine the shared content information of the sample text and the sample speech. Therefore, the target audio coding layer and the target text coding layer of the feature extraction module extracted by the trained extraction module can extract the potential correlation information of the speech data and text data in the target medical guidance order, thereby obtaining more accurate intelligent recommendation results, which is more advantageous for dealing with problems in complex scenarios.

[0156] From the above content, it can be seen that the present application conducts a more comprehensive and rich analysis of the target medical guidance order data from the two perspectives of text domain and audio domain. Since there is potential shared information between the voice data and text data of the same target medical guidance order, the present application trains a preset extraction module by utilizing the feature loss between the text and audio of the sample medical guidance order, so that there is more potential shared information between the features extracted from the sample text of the sample medical guidance order and the features extracted from the sample voice of the sample medical guidance order by the trained extraction module, and then obtains a feature extraction module by extracting and processing the trained extraction module. Therefore, the present application uses the feature extraction module to more specifically mine the correlation information between the target voice and the target text, obtains decoupled audio features and decoupled text features, effectively reduces the impact of different data types on feature fusion between text data and voice data, and more efficiently extracts potential correlation features between the text domain and the audio domain, which is beneficial to improving the accuracy of the classification recommendation results of the target medical guidance order, helps to more efficiently assist medical staff, provide better quality medical guidance services, and improve work efficiency.

[0157] In addition, in order to better implement the method for classifying and processing medical guide orders in the embodiment of the present application, on the basis of the method for classifying and processing medical guide orders, a medical guide order classification processing device is also provided in the embodiment of the present application, such as Figure 6 As shown, it is a schematic diagram of the structure of an embodiment of a medical guide order classification processing device provided in an embodiment of the present application. The medical guide order classification processing device 600 includes: an acquisition unit 601, a decoupling unit 602, and a classification unit 603:

[0158] An acquisition unit 601 is used to acquire a target voice of a target medical guidance order and a target text of the target medical guidance order;

[0159] A decoupling unit 602 is used to decouple the target speech through a target audio coding layer in a feature extraction module to obtain a decoupled audio feature of the target medical guidance order, wherein the target audio coding layer is obtained by extracting the audio coding layer of a trained extraction module, and the trained extraction module is obtained by training a preset extraction module based on a feature loss between text and speech of a sample medical guidance order;

[0160] The decoupling unit 602 is further used to perform a decoupling process on the target text through a target text encoding layer in the feature extraction module to obtain a decoupled text feature of the target medical guidance order, wherein the target text encoding layer is obtained by extracting the text encoding layer of the trained extraction module;

[0161] The classification unit 603 is used to perform classification processing based on the decoupled audio features and the decoupled text features through a trained classification module to obtain a classification recommendation result of the target medical guidance order.

[0162] In some embodiments, the medical guidance order classification processing device 600 further includes a training unit (not shown in the figure), which is used to:

[0163] Through the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the sample text of the sample medical guidance order to obtain a text feature group, wherein the text encoding layer in the preset extraction module includes a preset text association encoder and a preset text private encoder;

[0164] Through the audio coding layer in the preset extraction module, feature decoupling processing is performed based on the sample voice of the sample medical guidance order to obtain a spectrum feature group, wherein the audio coding layer in the preset extraction module includes a preset spectrum association encoder and a preset spectrum private encoder;

[0165] Based on the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss, the preset extraction module is trained to obtain a trained extraction module;

[0166] Performing extraction processing based on the audio coding layer of the trained extraction module to obtain the target audio coding layer;

[0167] The target text encoding layer is obtained by performing extraction processing based on the text encoding layer of the trained extraction module.

[0168] In some embodiments, the training unit is specifically used for:

[0169] Obtaining the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss;

[0170] By using the generation layer in the preset extraction module, feature cross processing is performed based on the text feature group and the spectrum feature group to obtain comparison text and comparison audio;

[0171] Based on the text similarity between the sample text of the sample medical consultation order and the comparison text, obtaining the text cycle consistency loss of the preset extraction module;

[0172] Based on the audio similarity between the sample voice of the sample medical guidance order and the comparison audio, obtaining the audio cycle consistency loss of the preset extraction module;

[0173] Based on the text-speech feature loss, the text cycle consistency loss, and the audio cycle consistency loss, the preset extraction module is trained to obtain the trained extraction module.

[0174] In some embodiments, the training unit is further configured to:

[0175] By using the generation layer in the preset extraction module, a generation process is performed based on the text feature group and the spectrum feature group to obtain the intermediate text of the sample medical guidance order and the intermediate audio of the sample medical guidance order;

[0176] By using the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the intermediate text to obtain an intermediate text feature group;

[0177] Performing feature decoupling processing based on the intermediate audio through the audio coding layer in the preset extraction module to obtain an intermediate spectrum feature group;

[0178] The comparison text and the comparison audio are obtained by performing generation processing based on the intermediate text feature group and the intermediate spectrum feature group through the generation layer in the preset extraction module.

[0179] In some embodiments, the training unit is further configured to:

[0180] Generate an intermediate text of the sample medical guidance order based on the text private features of the text feature group and the spectrum association features of the spectrum feature group by using a text generator in a generation layer in the preset extraction module;

[0181] The intermediate audio of the sample medical guidance order is generated based on the spectrum-private features of the spectrum feature group and the text-associated features of the text feature group through the audio generator in the generation layer in the preset extraction module.

[0182] In some embodiments, the decoupling unit 602 is specifically used for:

[0183] The target speech is decoupled by a spectrum correlation encoder in the target audio coding layer to obtain decoupled audio features of the target medical guidance order;

[0184] In some embodiments, the decoupling unit 602 is specifically used for:

[0185] The target text is decoupled by the text association encoder in the target text encoding layer to obtain the decoupled text features of the target medical guidance order.

[0186] In some embodiments, the decoupling unit 602 is further configured to:

[0187] Generate auxiliary text and auxiliary audio based on the decoupled audio feature and the decoupled text feature through the auxiliary information generation layer in the feature extraction module, wherein the auxiliary information generation layer in the feature extraction module is obtained by extracting the generation layer of the trained extraction module;

[0188] Through the target text encoding layer in the feature extraction module, feature decoupling processing is performed based on the auxiliary text to obtain auxiliary text features;

[0189] Performing feature decoupling processing based on the auxiliary audio through the target audio coding layer in the feature extraction module to obtain auxiliary audio features;

[0190] In some embodiments, the classification unit 603 is specifically used to:

[0191] Performing feature fusion on the decoupled audio feature and the decoupled text feature to obtain a fusion feature of the target medical guidance order;

[0192] Through the trained classification module, classification processing is performed based on the fusion features, the auxiliary text features, and the auxiliary audio features to obtain a classification recommendation result for the target medical guidance order.

[0193] In specific implementation, the above-mentioned units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above-mentioned units can refer to the above-mentioned implementation example of the classification processing method for medical guide orders, which will not be repeated here.

[0194] See also Figure 7 , Figure 7 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. The computer device may be a server, or a terminal device such as a smart phone, a laptop, or a portable calculator.

[0195] like Figure 7 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0196] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any one of the medical guidance order classification processing methods.

[0197] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0198] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any method for classifying and processing medical guide orders.

[0199] The network interface is used for network communication, such as receiving / sending assigned tasks, etc. Those skilled in the art will appreciate that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0200] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0201] In some embodiments, the processor is used to run a computer program stored in the memory to implement the following steps:

[0202] Obtain a target voice of a target medical guide order and a target text of the target medical guide order; perform decoupling processing on the target voice through a target audio coding layer in a feature extraction module to obtain decoupled audio features of the target medical guide order, wherein the target audio coding layer is obtained by extracting and processing the audio coding layer of a trained extraction module, and the trained extraction module is obtained by training a preset extraction module based on the text-speech feature loss of a sample medical guide order; perform decoupling processing on the target text through a target text coding layer in the feature extraction module to obtain decoupled text features of the target medical guide order, wherein the target text coding layer is obtained by extracting and processing the text coding layer of the trained extraction module; perform classification processing based on the decoupled audio features and the decoupled text features through a trained classification module to obtain a classification recommendation result of the target medical guide order.

[0203] A person of ordinary skill in the art can understand that all or part of the steps in the method for classifying and processing medical guide orders can be completed by instructions, or by controlling related hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0204] To this end, an embodiment of the present application provides a computer-readable storage medium, in which multiple computer programs are stored. The computer programs can be loaded by a processor to execute any one of the medical guide order classification processing methods provided in the embodiments of the present application.

[0205] The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0206] In the embodiments of the medical guide order classification processing device and the computer-readable storage medium, the description of each embodiment has its own emphasis. For the parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process and beneficial effects of the medical guide order classification processing device, computer-readable storage medium and its corresponding units described above can refer to the description of the medical guide order classification processing method in the above embodiment, and will not be repeated here.

[0207] The above is a detailed introduction to a method, device, computer equipment and computer-readable storage medium for classifying and processing medical guide orders provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and scope of application. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for classifying and processing medical guidance orders, characterized in that: The method comprises: Obtaining a target voice of a target medical guidance order and a target text of the target medical guidance order; The target speech is decoupled through the target audio coding layer in the feature extraction module to obtain the decoupled audio features of the target medical guidance order, wherein the target audio coding layer is obtained by extracting the audio coding layer of the trained extraction module, and the trained extraction module is obtained by training the preset extraction module based on the feature loss between text and speech of the sample medical guidance order; The target text is decoupled through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order, wherein the target text encoding layer is obtained by extracting the text encoding layer of the trained extraction module; Through the trained classification module, classification processing is performed based on the decoupled audio features and the decoupled text features to obtain a classification recommendation result for the target medical guidance order.

2. A method for classifying and processing medical guidance orders according to claim 1, characterized in that: The target audio coding layer of the feature extraction module and the target text coding layer of the feature extraction module are trained in the following manner: Through the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the sample text of the sample medical guidance order to obtain a text feature group, wherein the text encoding layer in the preset extraction module includes a preset text association encoder and a preset text private encoder; Through the audio coding layer in the preset extraction module, feature decoupling processing is performed based on the sample voice of the sample medical guidance order to obtain a spectrum feature group, wherein the audio coding layer in the preset extraction module includes a preset spectrum association encoder and a preset spectrum private encoder; Based on the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss, the preset extraction module is trained to obtain a trained extraction module; Performing extraction processing based on the audio coding layer of the trained extraction module to obtain the target audio coding layer; The target text encoding layer is obtained by performing extraction processing based on the text encoding layer of the trained extraction module.

3. A method for classifying and processing medical guidance orders according to claim 2, characterized in that: The method of training the preset extraction module based on the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss to obtain the trained extraction module comprises: Obtaining the loss between the text-related features in the text feature group and the spectrum-related features in the spectrum feature group as the text-speech feature loss; By using the generation layer in the preset extraction module, feature cross processing is performed based on the text feature group and the spectrum feature group to obtain comparison text and comparison audio; Based on the text similarity between the sample text of the sample medical consultation order and the comparison text, obtaining the text cycle consistency loss of the preset extraction module; Based on the audio similarity between the sample voice of the sample medical guidance order and the comparison audio, obtaining the audio cycle consistency loss of the preset extraction module; Based on the text-speech feature loss, the text cycle consistency loss, and the audio cycle consistency loss, the preset extraction module is trained to obtain the trained extraction module.

4. A method for classifying and processing medical guidance orders according to claim 3, characterized in that: The generating layer in the preset extraction module performs feature cross processing based on the text feature group and the spectrum feature group to obtain the comparison text and the comparison audio, including: By using the generation layer in the preset extraction module, a generation process is performed based on the text feature group and the spectrum feature group to obtain the intermediate text of the sample medical guidance order and the intermediate audio of the sample medical guidance order; By using the text encoding layer in the preset extraction module, feature decoupling processing is performed based on the intermediate text to obtain an intermediate text feature group; Performing feature decoupling processing based on the intermediate audio through the audio coding layer in the preset extraction module to obtain an intermediate spectrum feature group; The comparison text and the comparison audio are obtained by performing generation processing based on the intermediate text feature group and the intermediate spectrum feature group through the generation layer in the preset extraction module.

5. A method for classifying and processing medical guidance orders according to claim 4, characterized in that: The generating layer in the preset extraction module generates and processes based on the text feature group and the spectrum feature group to obtain the intermediate text of the sample medical guidance order and the intermediate audio of the sample medical guidance order, including: Generate an intermediate text of the sample medical guidance order based on the text private features of the text feature group and the spectrum association features of the spectrum feature group by using a text generator in a generation layer in the preset extraction module; The intermediate audio of the sample medical guidance order is generated based on the spectrum-private features of the spectrum feature group and the text-associated features of the text feature group through the audio generator in the generation layer in the preset extraction module.

6. A method for classifying and processing medical guidance orders according to claim 1, characterized in that: The target audio coding layer in the feature extraction module performs decoupling processing on the target voice to obtain the decoupled audio features of the target medical guidance order, including: The target speech is decoupled by a spectrum correlation encoder in the target audio coding layer to obtain decoupled audio features of the target medical guidance order; The target text is decoupled through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order, including: The target text is decoupled by the text association encoder in the target text encoding layer to obtain the decoupled text features of the target medical guidance order.

7. A method for classifying and processing medical guidance orders according to claim 1, characterized in that: The method further comprises: Generate auxiliary text and auxiliary audio based on the decoupled audio feature and the decoupled text feature through the auxiliary information generation layer in the feature extraction module, wherein the auxiliary information generation layer in the feature extraction module is obtained by extracting the generation layer of the trained extraction module; Through the target text encoding layer in the feature extraction module, feature decoupling processing is performed based on the auxiliary text to obtain auxiliary text features; Performing feature decoupling processing based on the auxiliary audio through the target audio coding layer in the feature extraction module to obtain auxiliary audio features; The trained classification module performs classification processing based on the decoupled audio features and the decoupled text features to obtain the classification recommendation result of the target medical guidance order, including: Performing feature fusion on the decoupled audio feature and the decoupled text feature to obtain a fusion feature of the target medical guidance order; Through the trained classification module, classification processing is performed based on the fusion features, the auxiliary text features, and the auxiliary audio features to obtain a classification recommendation result for the target medical guidance order.

8. A device for classifying and processing medical guidance orders, characterized in that: The medical guidance order classification processing device comprises: An acquisition unit, used to acquire a target voice of a target medical guidance order and a target text of the target medical guidance order; A decoupling unit is used to decouple the target speech through a target audio coding layer in a feature extraction module to obtain a decoupled audio feature of the target medical guidance order, wherein the target audio coding layer is obtained by extracting the audio coding layer of a trained extraction module, and the trained extraction module is obtained by training a preset extraction module based on a feature loss between text and speech of a sample medical guidance order; The decoupling unit is further used to perform decoupling processing on the target text through the target text encoding layer in the feature extraction module to obtain the decoupled text features of the target medical guidance order, wherein the target text encoding layer is obtained based on the extraction processing of the text encoding layer of the trained extraction module; A classification unit is used to perform classification processing based on the decoupled audio features and the decoupled text features through a trained classification module to obtain a classification recommendation result of the target medical guidance order.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the medical guide order classification processing method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the medical guide order classification processing method described in any one of claims 1 to 7.

Images