Medical text classification method and equipment based on speech recognition and large language model

Through the method of speech recognition and large language model, the problem of delayed text generation and low classification accuracy in emergency medical events is solved, timely medical text classification and alarm processing is achieved, and the response efficiency and accuracy of emergency medical events is improved.

CN119025680AInactive Publication Date: 2024-11-26TASI TECH (TIANJIN) CO LTD +1
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411170447.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

When an emergency medical incident occurs, it is difficult to generate complete medical text in a timely manner, resulting in delayed text classification, which in turn affects timely alarms for emergency medical incidents. Moreover, the text classification model based on machine learning is difficult to accurately classify based on semantic information, reducing the accuracy of alarms Spend.

Method used

Using a medical text classification method based on speech recognition and large language models, the initial medical text sound signal is obtained for preprocessing and converting it into text information, and using the pre-trained speech recognition model and text feature extraction model to clean and classify the data , generate medical text category information to alert.

Benefits of technology

It has achieved timely acquisition and classification of medical texts in emergency medical events, improved the timely alarm ability of emergency medical events, and improved the accuracy of alarms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025680A_ABST
    Figure CN119025680A_ABST
Patent Text Reader

Abstract

The embodiment of the invention relates to the field of text classification, in particular to a medical text classification method and device based on voice recognition and a large language model. A specific embodiment of the method comprises the following steps: acquiring an initial medical text sound signal; preprocessing the initial medical text sound signal to obtain a medical text sound filtering signal; inputting the medical text sound filtering signal into a pre-trained speech recognition model to obtain initial medical text information; performing data cleaning processing on the initial medical text information to obtain target medical text information; inputting the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information; performing classification processing on the medical text feature information to obtain medical text category information; and sending the medical text category information to an alarm terminal for alarm processing. According to the embodiment, the emergency medical event can be alarmed in time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of text classification, and specifically to a medical text classification method and device based on speech recognition and a large language model. Background Art

[0002] Medical call centers need to obtain medical texts in real time and classify them so that they can issue alerts in time for treatment when emergency medical events (e.g., massive bleeding, cardiac arrest, major car accidents) occur. Currently, the common method for classifying medical texts is to classify the entire text using a machine learning-based classification model.

[0003] However, in practice, it is found that when the above method is used for medical text classification, the following technical problems often occur: First, when an emergency medical incident occurs, it is often difficult to generate complete medical text in a timely manner, which makes it difficult to classify the text in a timely manner, and thus makes it difficult to issue an emergency medical incident alert in a timely manner; Second, using only machine learning-based classification models for text classification makes it difficult to classify text based on the semantic information in the text, resulting in reduced accuracy in text classification and, in turn, reduced accuracy in emergency medical event alerts. Summary of the invention

[0004] The content of this application is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this application is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0005] Some embodiments of the present application propose a medical text classification method, a computer device, and a computer-readable storage medium based on speech recognition and a large language model to solve one or more of the technical problems mentioned in the above background technology section.

[0006] In a first aspect, some embodiments of the present application provide a medical text classification method based on speech recognition and a large language model, the method comprising: obtaining an initial medical text sound signal; preprocessing the initial medical text sound signal to obtain a medical text sound filter signal; inputting the medical text sound filter signal into a pre-trained speech recognition model to obtain initial medical text information, wherein the pre-trained speech recognition model comprises: a classification sub-model, an encoding sub-model, a mapping sub-model and a decoding sub-model; performing data cleaning processing on the initial medical text information to obtain target medical text information; inputting the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information; classifying the medical text feature information to obtain medical text category information; and sending the medical text category information to an alarm terminal for alarm processing.

[0007] In a second aspect, the present application also provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the method described in any implementation of the first aspect is implemented.

[0008] In a third aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, wherein when the computer program is executed by a processor, the method described in any implementation of the first aspect is implemented.

[0009] The above-mentioned various embodiments of the present application have the following beneficial effects: through the medical text classification method based on speech recognition and large language model of some embodiments of the present application, emergency medical events can be warned in time. Specifically, the reason why it is difficult to warn of emergency medical events in time is that when an emergency medical event occurs, it is often difficult to generate a complete medical text in time, which makes it difficult to classify the text in time, thereby making it difficult to warn of the emergency medical event in time. Based on this, the medical text classification method based on speech recognition and large language model of some embodiments of the present application first obtains the initial medical text sound signal. Thus, the sound information of the initial medical text can be obtained by inputting the speech signal. Secondly, the above-mentioned initial medical text sound signal is preprocessed to obtain the medical text sound filter signal. Thus, the features in the initial medical text sound signal can be preliminarily extracted. Then, the above-mentioned medical text sound filter signal is input into the pre-trained speech recognition model to obtain the initial medical text information, wherein the pre-trained speech recognition model includes: a classification sub-model, an encoding sub-model, a mapping sub-model and a decoding sub-model. Thus, the input speech signal can be converted into text information. Next, the above-mentioned initial medical text information is subjected to data cleaning processing to obtain the target medical text information. In this way, redundant information in the text information can be screened out. Subsequently, the above-mentioned target medical text information is input into a pre-trained text feature extraction large language model to obtain medical text feature information. In this way, text features can be extracted based on the semantic information of the text. Then, the above-mentioned medical text feature information is classified to obtain medical text category information. In this way, the text can be classified based on the text feature information. Finally, the above-mentioned medical text category information is sent to the alarm terminal for alarm processing. In this way, an alarm can be issued based on the text feature information. Therefore, some medical text classification methods based on speech recognition and large language models of the present application can obtain medical text information in a timely manner by receiving a speech signal and then converting the speech signal into text, so as to classify the medical text in a timely manner, thereby being able to issue an alarm in a timely manner for emergency medical events represented by the medical text. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0011] Figure 1 is a flowchart of some embodiments of the medical text classification method based on speech recognition and a large language model according to the present application; Figure 2It is a schematic diagram of the structure of a computer device suitable for implementing some embodiments of the present application. DETAILED DESCRIPTION

[0012] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0013] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0014] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0015] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0016] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0017] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0018] Figure 1 The process 100 of some embodiments of the medical text classification method based on speech recognition and a large language model according to the present application is shown. The medical text classification method based on speech recognition and a large language model comprises the following steps: Step 101, obtaining an initial medical text sound signal.

[0019] In some embodiments, the execution subject of the medical text classification method based on speech recognition and large language model can obtain the initial medical text sound signal from the storage terminal through a wired connection or a wireless connection. The storage terminal can be a terminal for storing the initial medical text sound signal. The initial medical text sound signal can be a signal representing the medical text information.

[0020] As an example, the above-mentioned medical text information can be but is not limited to at least one of the following: information representing "the patient has high blood pressure", information representing "the patient has upper respiratory tract infection", information representing "the patient's heartbeat slows down", and information representing "a major car accident".

[0021] It should be noted that the above wireless connection methods may include but are not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0022] Step 102, preprocessing the initial medical text sound signal to obtain a medical text sound filtering signal.

[0023] In some embodiments, the execution entity may pre-process the initial medical text sound signal to obtain a medical text sound filter signal.

[0024] In some optional implementations of some embodiments, the execution subject preprocesses the initial medical text sound signal to obtain a medical text sound filter signal, which may include the following steps: The first step is to perform pre-emphasis processing on the initial medical text sound signal to obtain a medical text sound emphasized signal. The initial medical text sound signal can be pre-emphasized using a preset pre-emphasis algorithm to obtain a medical text sound emphasized signal.

[0025] As an example, the above-mentioned preset pre-emphasis algorithm may be a high-pass filter algorithm.

[0026] The second step is to perform frame processing on the medical text sound emphasis signal to obtain a medical text sound frame signal. The medical text sound emphasis signal can be frame processed according to a preset time interval to obtain a medical text sound frame signal.

[0027] As an example, the preset time interval may be, but is not limited to, at least one of the following: 10 milliseconds, 20 milliseconds, or 30 milliseconds.

[0028] The third step is to perform windowing processing on the medical text sound frame signal to obtain a medical text sound window signal. The medical text sound frame signal can be windowed by a preset window function to obtain a medical text sound window signal.

[0029] As an example, the preset window function may be, but is not limited to, at least one of the following: a rectangular window function, a Hamming window function, or a Henning window function.

[0030] The fourth step is to transform the medical text sound window signal to obtain the medical text sound spectrum signal. The medical text sound window signal can be transformed by a preset transformation algorithm to obtain the medical text sound spectrum signal.

[0031] As an example, the above-mentioned preset transformation algorithm may be a Fourier transform algorithm.

[0032] The fifth step is to filter the medical text sound spectrum signal to obtain a medical text sound filter signal. The medical text sound spectrum signal can be filtered by a preset filter to obtain a medical text sound filter signal.

[0033] As an example, the above-mentioned preset filter may be a Mel filter.

[0034] Step 103, input the medical text sound filtering signal into a pre-trained speech recognition model to obtain initial medical text information.

[0035] In some embodiments, the above-mentioned execution entity can input the above-mentioned medical text sound filter signal into a pre-trained speech recognition model to obtain initial medical text information. Wherein, the above-mentioned pre-trained speech recognition model includes: a classification sub-model, an encoding sub-model, a mapping sub-model and a decoding sub-model. The above-mentioned pre-trained speech recognition model can be a pre-trained neural network model with a medical text sound filter signal as input and initial medical text information as output. Specifically, the above-mentioned classification sub-model can be used to classify the above-mentioned medical text sound filter signal. The above-mentioned encoding sub-model can be used to encode the classified medical text sound filter signal. The above-mentioned mapping sub-model can be used to map the encoded medical text sound filter signal. The above-mentioned decoding sub-model can be used to decode the mapped medical text sound filter signal to obtain the above-mentioned initial medical text information.

[0036] Optionally, before inputting the medical text sound filtering signal into a pre-trained speech recognition model to obtain initial medical text information, the execution subject may further perform the following steps: The first step is to obtain a sample speech signal set and a sample speech text set. The sample speech signals in the sample speech signal set correspond to the sample speech texts in the sample speech text set. The sample speech signals may be pre-processed filtered signals. Specifically, the specific implementation method of generating the sample speech signals and the technical effects thereof may refer to step 102 in the above embodiment, which will not be described in detail here.

[0037] In the second step, a sample speech signal is selected from the sample speech signal set as a target sample speech signal, and the following recognition training sub-steps are performed: In the first sub-step, the target sample speech signal is input into the classification sub-model included in the initial speech recognition model to obtain the initial speech classification information. The initial speech recognition model also includes a coding sub-model, a mapping sub-model and a decoding sub-model. Here, a sample speech signal can be randomly selected from the above-mentioned sample speech signal set as the target sample speech signal. The above-mentioned initial speech recognition model can be an untrained neural network model that takes the target sample speech signal as input and the initial speech text information as output. Specifically, the above-mentioned classification sub-model can be used to classify the target sample speech signal to obtain the initial speech classification information. The above-mentioned coding sub-model can be used to encode the initial speech classification information to obtain the initial speech coding information. The above-mentioned mapping sub-model can be used to map the initial speech coding information to obtain the initial speech mapping information. The above-mentioned decoding sub-model can be used to decode the initial speech mapping information to obtain the initial speech text information.

[0038] As an example, the classification sub-model may be an EfficientNet (lightweight network) model. The encoding sub-model may be a RetNet (residual network) model. The mapping sub-model may be a linear network model. The decoding sub-model may be a Beam Search (beam search) model.

[0039] The second sub-step is to input the initial speech classification information into the encoding sub-model included in the initial speech recognition model to obtain initial speech encoding information.

[0040] The third sub-step is to input the initial speech coding information into the mapping sub-model included in the initial speech recognition model to obtain the initial speech mapping information.

[0041] The fourth sub-step is to input the initial speech mapping information into the decoding sub-model included in the initial speech recognition model to obtain the initial speech text information.

[0042] The fifth sub-step is to determine the recognition difference value of the initial voice text information and the sample voice text in the sample voice text set corresponding to the target sample voice signal based on a preset recognition loss function.

[0043] As an example, the preset recognition loss function may be a CTC (connectionist temporal classification) function.

[0044] A sixth sub-step, in response to determining that the recognition difference value is less than a preset difference value, determines the initial speech recognition model as the speech recognition model.

[0045] As an example, the preset difference value may be 0.01.

[0046] Optionally, the execution subject may also adjust relevant parameters in the initial speech recognition model in response to determining that the recognition difference value is greater than or equal to a preset difference value, determine the adjusted initial speech recognition model as the initial speech recognition model, and select a sample speech signal from each sample speech signal that has not been selected in the sample speech signal set as the target sample speech signal for performing the recognition training step again. The relevant parameters in the initial speech recognition model may be adjusted by a preset adjustment algorithm.

[0047] As an example, the preset adjustment algorithm mentioned above may be but is not limited to one of the following: a back propagation algorithm or a stochastic gradient algorithm.

[0048] Thus, the filtered signal can be classified through the classification sub-model, and then the filtered signal can be further classified through the encoding sub-model. Then, the linear features of the speech signal can be obtained through the mapping sub-model, and finally, the text information corresponding to the speech signal can be obtained through the decoding sub-model. Thus, the speech signal can be converted into text in real time, so that the medical text can be classified in time. Furthermore, emergency medical events can be warned in time.

[0049] Step 104, performing data cleaning processing on the initial medical text information to obtain target medical text information.

[0050] In some embodiments, the execution subject may perform data cleaning on the initial medical text information to obtain target medical text information. Specifically, the initial medical text information may be cleaned using a preset data cleaning algorithm to obtain target medical text information.

[0051] As an example, the above-mentioned preset data cleaning algorithm may be a regular expression algorithm.

[0052] Step 105, input the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information.

[0053] In some embodiments, the execution subject may input the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information. The pre-trained text feature extraction large language model may be a pre-trained neural network model that takes the target medical text information as input and the medical text feature information as output.

[0054] Optionally, before inputting the target medical text information into the pre-trained text feature extraction large language model to obtain the medical text feature information, the execution subject may further perform the following steps: The first step is to obtain a first sample text information set and a second sample text set. The first sample text information in the first sample text information set includes: a first sample text and first sample text feature information. The first sample text may be a text representing medical information. The first sample text feature information may be information representing features of the first sample text.

[0055] As an example, the first sample text may be, but is not limited to, at least one of the following: text representing "the patient has elevated blood pressure", text representing "the patient has an upper respiratory tract infection", or text representing "the patient has a slowed heartbeat". The first sample text feature information may be, but is not limited to, at least one of the following: information representing "elevated blood pressure", information representing "upper respiratory tract infection", or information representing "slowed heartbeat".

[0056] The second step is to generate a first text feature extraction large language model based on the first sample text information set.

[0057] The third step is to extract a large language model based on the first text feature, generate second text label information corresponding to each second sample text in the second sample text set, and obtain a second text label information set.

[0058] The fourth step is to fuse the first sample text information set, the second sample text set and the second text label information set to obtain a fused sample text information set. The fused sample text information in the fused sample text information set includes: fused sample text and fused text label information. The above-mentioned fusion processing of the first sample text information set, the second sample text set and the second text label information set to obtain the fused sample text information set can be: first, for each second sample text in the above-mentioned second sample text set, the second text label information corresponding to the above-mentioned second sample text in the above-mentioned second text label information set and the above-mentioned second sample text are determined as the initial fused text label information and the initial fused sample text included in the initial fused sample text information. Then, each first sample text information in the above-mentioned first sample text information set and each initial fused sample text information determined above are determined as each fused sample text information included in the fused sample text information.

[0059] Step 5: Select fused sample text information from the above fused sample text information set as target fused sample text information, and perform the following second training sub-step: In the first sub-step, the fused sample text included in the target fused sample text information is input into the first text feature extraction large language model to obtain fused text feature information. The fused sample text information can be randomly selected from the above fused sample text information set as the target fused sample text information.

[0060] The second sub-step is to determine, based on a preset second loss function, a second feature difference value of the fused text feature information and the fused text label information included in the target fused sample text information.

[0061] As an example, the above-mentioned preset second loss function may be a dual-label correction loss function.

[0062] In a third sub-step, in response to determining that the second feature difference value is less than the target difference value, the first text feature extraction large language model is determined as the text feature extraction large language model.

[0063] As an example, the target difference value mentioned above may be 0.001.

[0064] Optionally, the execution subject can also adjust the relevant parameters in the first text feature extraction large language model in response to determining that the second feature difference value is greater than or equal to the target difference value, determine the adjusted first text feature extraction large language model as the first text feature extraction large language model, and select the fusion sample text information from each fusion sample text information that has not been selected in the above fusion sample text information set as the target fusion sample text information, so as to perform the above step of generating the second text label information again. The relevant parameters in the first text feature extraction large language model can be adjusted by the above preset adjustment algorithm.

[0065] In some optional implementations of some embodiments, the execution subject generates a first text feature extraction large language model based on the first sample text information set, and can select first sample text information from the first sample text information set as target sample text information to perform the following first training sub-step: In a first sub-step, the first sample text included in the target sample text information is input into the large language model for initial text feature extraction to obtain initial text feature information. The large language model for initial text feature extraction may be an untrained large language model that takes the first sample text as input and takes the initial text feature information as output.

[0066] As an example, the initial text feature extraction large language model may be an LLM (Large Language Model).

[0067] In a second sub-step, based on a preset first loss function, a first feature difference value of the first sample text feature information included in the initial text feature information and the target sample text information is determined.

[0068] As an example, the above-mentioned preset first loss function can be but is not limited to at least one of the following: a cross entropy loss function, a negative sampling loss function, a contrast loss function or an autoregressive loss function.

[0069] In a third sub-step, in response to determining that the first feature difference value is less than the target difference value, the initial text feature extraction large language model is determined as the first text feature extraction large language model.

[0070] As an example, the target difference value mentioned above may be 0.001.

[0071] Optionally, the execution subject may also adjust relevant parameters in the large language model for extracting initial text features in response to determining that the first feature difference value is greater than or equal to the target difference value, determine the adjusted large language model for extracting initial text features as the large language model for extracting initial text features, and select first sample text information from each first sample text information that has not been selected in the first sample text information set as the target sample text information for executing the first training sub-step again. The relevant parameters in the large language model for extracting initial text features may be adjusted by the preset adjustment algorithm.

[0072] In some optional implementations of some embodiments, the execution subject extracts a large language model based on the first text feature, generates second text label information corresponding to each second sample text in the second sample text set, and obtains the second text label information set, which may include the following steps: In the first step, each second sample text in the second sample text set is input into the first text feature extraction large language model to generate second text feature information, thereby obtaining a second text feature information set.

[0073] The second step is to cluster the second text feature information set to obtain a second text feature information group set. The second text feature information set can be clustered using a preset clustering algorithm to obtain the second text feature information group set.

[0074] As an example, the above-mentioned preset clustering algorithm may be but is not limited to at least one of the following: a K-Means clustering algorithm, a density clustering algorithm or an Infomap clustering algorithm.

[0075] The third step is to generate the second text label information corresponding to each second text feature information group in the second text feature information group set to obtain the second text label information set. The preset label information can be determined as the second text label information corresponding to each second text feature information group in the second text feature information group set.

[0076] As an example, the above-mentioned preset label information may be but is not limited to at least one of the following: information representing the "first category", information representing the "second category" or information representing the "third category".

[0077] The relevant content of step 105, as an inventive point of an embodiment of the present application, solves the technical problem 2 mentioned in the background technology, "the accuracy of emergency medical event warning is reduced". Among them, the factors that lead to the reduction of the accuracy of emergency medical event warning are often as follows: using only a classification model based on machine learning for text classification, it is difficult to classify the text according to the semantic information in the text, resulting in a reduction in the accuracy of text classification. If the above factors are solved, the effect of improving the accuracy of emergency medical event warning can be achieved. In order to achieve this effect, the present application can use part of the sample information containing annotation information to perform preliminary training on the initial large language model. Then, the unlabeled information can be input into the large language model after preliminary training, and the output of the large language model can be used as the annotation information of the unlabeled sample. Then, the large language model after preliminary training can be further trained according to the generated annotation information, thereby improving the accuracy of the model output. Therefore, the large language model can be used to extract feature information in the text according to the semantic information in the text, and the accuracy of the text feature information output by the model can be improved through a training method that combines unsupervised training with supervised training. Thereby, the accuracy of text classification can be improved, and further, the accuracy of emergency medical event warnings can be improved.

[0078] Step 106, classify the medical text feature information to obtain medical text category information.

[0079] In some embodiments, the execution subject may classify the medical text feature information to obtain medical text category information, wherein the medical text feature information may be classified by a preset classification algorithm to obtain the medical text category information.

[0080] As an example, the above-mentioned preset classification algorithm can be but is not limited to at least one of the following: a logistic regression algorithm, a support vector machine algorithm, a naive Bayes algorithm, a decision tree algorithm or a random forest algorithm.

[0081] Step 107, sending the medical text category information to the alarm terminal for alarm processing.

[0082] In some embodiments, the execution subject may send the medical text category information to an alarm terminal for alarm processing. The alarm terminal may be a terminal for displaying warning text corresponding to the medical text category information or emitting a prompt tone corresponding to the medical text category information in response to receiving the medical text category information.

[0083] The above-mentioned various embodiments of the present application have the following beneficial effects: through the medical text classification method based on speech recognition and large language model of some embodiments of the present application, emergency medical events can be warned in time. Specifically, the reason why it is difficult to warn of emergency medical events in time is that when an emergency medical event occurs, it is often difficult to generate a complete medical text in time, which makes it difficult to classify the text in time, thereby making it difficult to warn of the emergency medical event in time. Based on this, the medical text classification method based on speech recognition and large language model of some embodiments of the present application first obtains the initial medical text sound signal. Thus, the sound information of the initial medical text can be obtained by inputting the speech signal. Secondly, the above-mentioned initial medical text sound signal is preprocessed to obtain the medical text sound filter signal. Thus, the features in the initial medical text sound signal can be preliminarily extracted. Then, the above-mentioned medical text sound filter signal is input into the pre-trained speech recognition model to obtain the initial medical text information, wherein the pre-trained speech recognition model includes: a classification sub-model, an encoding sub-model, a mapping sub-model and a decoding sub-model. Thus, the input speech signal can be converted into text information. Next, the above-mentioned initial medical text information is subjected to data cleaning processing to obtain the target medical text information. In this way, redundant information in the text information can be screened out. Subsequently, the above-mentioned target medical text information is input into a pre-trained text feature extraction large language model to obtain medical text feature information. In this way, text features can be extracted based on the semantic information of the text. Then, the above-mentioned medical text feature information is classified to obtain medical text category information. In this way, the text can be classified based on the text feature information. Finally, the above-mentioned medical text category information is sent to the alarm terminal for alarm processing. In this way, an alarm can be issued based on the text feature information. Therefore, some medical text classification methods based on speech recognition and large language models of the present application can obtain medical text information in a timely manner by receiving a speech signal and then converting the speech signal into text, so as to classify the medical text in a timely manner, thereby being able to issue an alarm in a timely manner for emergency medical events represented by the medical text.

[0084] The present application also provides a computer device 200. Figure 2 As shown, the computer device 200 includes: a bus 201, a processor 202, a memory 203 and a communication interface 204. The processor 202, the memory 203 and the communication interface 204 communicate with each other through the bus 201. The computer device 200 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computer device 200.

[0085] The bus 201 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 The bus 201 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 201 may include a path for transmitting information between various components of the computer device 200 (for example, the memory 203, the processor 202, and the communication interface 204).

[0086] The processor 202 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0087] The memory 203 may include a volatile memory, such as a random access memory (RAM). The memory 203 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0088] The memory 203 stores executable program codes, and the processor 202 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module, sampling module, determination module, and mixing module, thereby implementing the aforementioned medical text classification method based on speech recognition and a large language model. That is, the memory 203 stores instructions for executing the aforementioned medical text classification method based on speech recognition and a large language model.

[0089] The communication interface 204 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computer device 200 and other devices or a communication network.

[0090] An embodiment of the present application also provides a chip, which includes a processor and a data interface. The processor reads instructions stored in a memory through the data interface to execute the above-mentioned medical text classification method based on speech recognition and a large language model.

[0091] The embodiment of the present application also provides a computer-readable storage medium. The above-mentioned computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The above-mentioned available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute the above-mentioned medical text classification method based on speech recognition and a large language model.

[0092] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the embodiments of the present application.

Claims

1. A medical text classification method based on speech recognition and a large language model, comprising: Acquire initial medical text sound signal; Preprocessing the initial medical text sound signal to obtain a medical text sound filter signal; Inputting the medical text sound filtering signal into a pre-trained speech recognition model to obtain initial medical text information, wherein the pre-trained speech recognition model includes: a classification sub-model, an encoding sub-model, a mapping sub-model and a decoding sub-model; Performing data cleaning on the initial medical text information to obtain target medical text information; Inputting the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information; Classifying the medical text feature information to obtain medical text category information; The medical text category information is sent to an alarm terminal for alarm processing.

2. The medical text classification method based on speech recognition and large language model according to claim 1, wherein: The preprocessing of the initial medical text sound signal to obtain a medical text sound filtering signal includes: Performing pre-emphasis processing on the initial medical text sound signal to obtain a medical text sound emphasized signal; Performing frame processing on the medical text sound emphasis signal to obtain a medical text sound frame signal; Performing windowing processing on the medical text sound frame signal to obtain a medical text sound window signal; Transforming the medical text sound window signal to obtain a medical text sound spectrum signal; The medical text sound spectrum signal is filtered to obtain a medical text sound filtered signal.

3. The medical text classification method based on speech recognition and large language model according to claim 1, wherein: Before inputting the target medical text information into a pre-trained text feature extraction large language model to obtain medical text feature information, the method further includes: Acquire a first sample text information set and a second sample text set, wherein the first sample text information in the first sample text information set includes: a first sample text and first sample text feature information; Based on the first sample text information set, generating a first text feature extraction large language model; Extracting a large language model based on the first text feature, generating second text label information corresponding to each second sample text in the second sample text set, and obtaining a second text label information set; The first sample text information set, the second sample text set and the second text label information set are fused to obtain a fused sample text information set, wherein the fused sample text information in the fused sample text information set includes: fused sample text and fused text label information; Select fused sample text information from the fused sample text information set as target fused sample text information, and perform the following second training step: Inputting the fused sample text included in the target fused sample text information into the first text feature extraction large language model to obtain fused text feature information; Based on a preset second loss function, determining a second feature difference value of the fused text feature information and the fused text label information included in the target fused sample text information; In response to determining that the second feature difference value is less than the target difference value, the first text feature extraction large language model is determined as the text feature extraction large language model.

4. The medical text classification method based on speech recognition and large language model according to claim 3, wherein: The method further comprises: In response to determining that the second feature difference value is greater than or equal to the target difference value, relevant parameters in the first text feature extraction large language model are adjusted, the adjusted first text feature extraction large language model is determined as the first text feature extraction large language model, and fused sample text information is selected from each of the fused sample text information that has not been selected in the fused sample text information set as the target fused sample text information, so as to perform the step of generating the second text label information again.

5. A computer device, wherein: The computer device comprises a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Citation Information

Patent Citations

  • Voice recognition device and method, electronic equipment and computer readable storage medium

    CN111862944A

  • Semantic recognition-based record generation method and device, equipment and storage medium

    CN113806492A

  • Text classification method and device, electronic equipment and storage medium

    CN114153975A

  • Text processing method and device, storage medium and electronic equipment

    CN115374782A

  • Medical text classification method and device, equipment and medium

    CN116450829A