Medical training model control method based on voice recognition

Abstract

The invention discloses a medical training model control method based on voice recognition, and the method comprises the steps: S1, constructing an instruction library, and dividing the types of instructions; s2, assigning a trigger level of the instruction category; s3, collecting a sound instruction sent by a user, and converting the sound instruction into an instruction text through voice recognition; s4, performing semantic analysis on the instruction text based on a large language model LLM, and matching the instruction text with instructions in the instruction library to obtain a target instruction corresponding to the instruction text; and S5, according to the instruction category and the trigger level corresponding to the target instruction, executing the instruction operation, through the application, the voice intention of the user can be quickly identified, and orderly execution of the instruction can be guaranteed.

Description

Technical Field

[0001] The present application relates to the field of medical simulation, and in particular to a medical training model control method based on speech recognition. Background Art

[0002] With the advancement of intelligent technology and the development of teaching modernization, technologies such as speech recognition have been widely used in the teaching field to improve the human-computer interaction experience during training.

[0003] Among them, in the field of medical teaching, due to the large number of types of diseases, different diseases have different clinical manifestations, treatment methods, intervention measures, etc., and the clinical manifestations during the treatment and intervention process will produce corresponding changes, resulting in the medical teaching interaction scenarios of voice recognition being more complex. Therefore, this application aims to improve the accuracy of voice commands issued during the interaction process of using medical training models and enhance the interactive experience. Summary of the invention

[0004] In order to solve the above problems, the present application provides a medical training model control method based on speech recognition, comprising the following steps: S1, build instruction library and divide instruction categories The vocabulary describing the categories of patients’ vital signs in the collected medical records is marked as vital sign parameter instructions; The vocabulary describing the patient's performance in the collected medical records was labeled as model performance instructions; The words describing the patient's condition in the medical records are marked as condition parameter instructions; The words describing the intervention and treatment methods implemented on the patients in the collected medical records are marked as intervention parameter instructions; The words describing the medications given to the patient in the medical records were collected and labeled as medication parameter instructions; S2, specify the trigger level of the instruction category Specifying the triggering level of the physical sign parameter instruction and the model performance instruction as level A; Specifying the trigger level of the condition parameter instruction as level B; Specifying the trigger level of the intervention parameter instruction as level C; Specifying the trigger level of the drug parameter instruction as level D; The trigger priority order of the trigger level is A level greater than B level greater than C level greater than D level; S3, collecting the voice commands issued by the user, and converting the voice commands into command text through speech recognition; S4, performing semantic analysis on the instruction text based on the large language model LLM, matching it with the instructions in the instruction library, and obtaining a target instruction corresponding to the instruction text; S5, executing the instruction operation according to the instruction type and trigger level corresponding to the target instruction.

[0005] Furthermore, The vital sign parameter instructions include heart rate, respiratory rate, blood oxygen saturation, end-tidal carbon dioxide, invasive blood pressure, non-invasive blood pressure, and body surface temperature; The model performance instructions include eye state, lip color, forehead sweating, foaming at the mouth, shivering, and difficult labor; The condition parameter instructions include bleeding, airway obstruction, ventricular coarse fibrillation, ventricular fine fibrillation, ventricular flutter, cardiac arrest, hypertension, death, asthma, and respiratory arrest; The intervention parameter instructions include fetal heart monitoring, endotracheal intubation, cesarean section, guided examination, umbilical cord bandaging, placental delivery, and digital rectal examination; The drug parameter instructions include aminophylline, oseltamivir, acetaminophen, dexamethasone, compound amino acids, fentanyl, epinephrine, normal saline, metronidazole, oxytocin, and labetalol.

[0006] Further, step S3 includes: S301, monitor the ambient sound in real time, determine whether the volume of the current ambient sound meets the threshold volume, if yes, mark the current ambient sound as valid voice input, collect the current ambient sound as the current voice command, if no, mark the current ambient sound as invalid voice input; S302, converting the current voice instruction into instruction text through voice recognition, and storing it in the instruction queue in time sequence.

[0007] Furthermore, step S3 also includes step S303, determining whether there is a time interval between the instruction texts in the instruction queue, and when the time interval is greater than a preset duration, segmenting the instruction text to obtain instruction segmented texts.

[0008] Further, step S4 includes: S401, performing semantic analysis on the instructions in the instruction library based on a large language model LLM to obtain a global vector value of the instruction library; S402, obtaining the instruction text, performing semantic analysis based on a large language model (LLM), and obtaining a vector value of the instruction text; S403, traverse and match the vector value of the instruction text with the global vector value of the instruction library to obtain the target instruction corresponding to the instruction text.

[0009] Further, step S403 specifically includes: determining the number of vector values ​​of the instruction text, When the number of vector values ​​of the instruction text is 0, return to step S3; When the number of vector values ​​in the instruction text is non-zero, the global vector values ​​of the instruction library are traversed and matched one by one to determine whether the match is successful. If the match is successful, the instruction in the instruction library that is consistent with the vector value is marked as the target instruction, and Num is counted. If the match is not successful, the instruction text corresponding to the vector value that is not successfully matched is marked as a redundant instruction.

[0010] Furthermore, step S403 further includes: When Num is 1, the target instruction is marked as a single instruction; When Num is greater than 1, the target instruction is marked as a multiple instruction.

[0011] Furthermore, it also includes: Determine whether the multiple instructions include a level 2 instruction, and if yes, mark the instruction category and trigger level of the level 2 instruction as being the same as the corresponding level 1 instruction.

[0012] Further, step S5 includes: S501, determine whether there is an instruction with a level A trigger mark, if yes, execute the instruction with the level A trigger mark, if no, execute step S502; S502, determine whether there is an instruction with a B-level trigger level mark, if yes, execute the instruction with the B-level trigger level mark, if no, execute step S503; S503, determining whether there is an instruction with a C-level trigger level mark, if yes, executing the instruction with the C-level trigger level mark, if no, executing step S504; S504, determine whether there is an instruction with a D-level trigger level mark, if yes, execute the instruction with the D-level trigger level mark.

[0013] The beneficial effects of this application are: By categorizing the command library, we can quickly identify voice commands during the interaction process and clarify the user's voice intention. At the same time, the setting of trigger levels also realizes the order of triggering multiple commands, ensuring the correct process of medical training model medical record demonstration and intervention treatment, and improving the experience during the interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of a medical training model control method based on speech recognition in an embodiment of the present application.

[0015] Figure 2 This is a flowchart of step S3 of the medical training model control method based on speech recognition in an embodiment of the present application.

[0016] Figure 3This is a flowchart of step S4 of the medical training model control method based on speech recognition in an embodiment of the present application.

[0017] Figure 4 This is a flowchart of step S403 of the medical training model control method based on speech recognition in an embodiment of the present application.

[0018] Figure 5 This is a flowchart of step S5 of the medical training model control method based on speech recognition in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below in conjunction with the drawings of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present application.

[0020] like Figure 1 As shown, the embodiment of the present application provides a medical training model control method based on speech recognition, through which the different intentions, different types of instructions and the order of execution of control instructions in the user's voice instructions can be accurately identified during the interaction process, thereby realizing the simulation control of the medical training model and improving the interactive experience, which specifically includes the following steps: S1, build instruction library and divide instruction categories The vocabulary describing the categories of patients’ vital signs in the collected medical records is marked as vital sign parameter instructions; The vocabulary describing the patient's performance in the collected medical records was labeled as model performance instructions; The words describing the patient's condition in the medical records are marked as condition parameter instructions; The words describing the intervention and treatment methods implemented on the patients in the collected medical records are marked as intervention parameter instructions; The words describing the medications given to the patient in the medical records were collected and labeled as medication parameter instructions; S2, specify the trigger level of the instruction category The trigger level of the specified physical sign parameter command and model performance command is A; The trigger level of the specified condition parameter command is B; The trigger level of the specified intervention parameter command is C level; The trigger level of the designated drug parameter instruction is D level; The trigger priority order of the trigger level is A level greater than B level greater than C level greater than D level; S3, collecting the voice commands issued by the user and converting the voice commands into command text through speech recognition; S4, performing semantic analysis on the instruction text based on the large language model LLM, matching it with the instructions in the instruction library, and obtaining the target instruction corresponding to the instruction text; S5, executing the instruction operation according to the instruction category and trigger level corresponding to the target instruction.

[0021] In this embodiment, the medical record is a clinical history record or is written by an experienced teacher. By collecting words with different attributes in the medical record to form a control instruction library for controlling the medical training model, the control instructions of the medical training model are more comprehensive and can provide more voice interaction scenarios for medical training.

[0022] By setting the above instruction categories, instructions of different attribute types are divided to facilitate the recognition and execution of user voice commands. At the same time, the trigger level of the instruction category is set so that there is a comprehensible sequence when executing instructions, ensuring the correct steps of the medical training model in the process of demonstrating medical records and diagnosis and treatment interventions.

[0023] Typically, vital sign parameter instructions include but are not limited to heart rate, respiratory rate, blood oxygen saturation, end-tidal carbon dioxide, invasive blood pressure, non-invasive blood pressure, body surface temperature, etc.; Model performance instructions, including but not limited to eye condition, lip color, forehead sweating, foaming at the mouth, shivering, difficult labor, etc.; Condition parameter instructions, including but not limited to bleeding (fast, medium, or slow), airway obstruction, ventricular coarse fibrillation, ventricular fine fibrillation, ventricular flutter, cardiac arrest, hypertension (grade 1, 2, or 3), death, asthma, respiratory arrest, etc.; Intervention parameter instructions, including but not limited to fetal heart rate monitoring, endotracheal intubation, cesarean section, guided examination, umbilical cord bandaging, placental delivery, digital rectal examination, etc.; Drug parameter instructions, including but not limited to aminophylline, oseltamivir, acetaminophen, dexamethasone, compound amino acids, fentanyl, epinephrine, normal saline, metronidazole, oxytocin, labetalol, etc.

[0024] In some embodiments, instruction vocabulary may also be collected by inputting medical records into a training model.

[0025] In this embodiment, Figure 2 As shown, step S3 includes: S301, monitor the ambient sound in real time, determine whether the volume of the current ambient sound meets the threshold volume, if yes, mark the current ambient sound as valid voice input, collect the current ambient sound as the current voice command, if no, mark the current ambient sound as invalid voice input; S302, converting the current voice command into command text through voice recognition, and storing it in a command queue in time sequence; S303, determining whether there is a time interval between the instruction texts in the instruction queue, and when the time interval is greater than a preset time length, segmenting the instruction text to obtain instruction segment texts.

[0026] In this embodiment, by setting the threshold volume, invalid ambient sound can be filtered out, the interference of ambient sound can be reduced, and the quality of collecting user voice can be improved. The threshold volume can be set and adjusted according to the on-site environment or the voice of the participant.

[0027] In this embodiment, the instruction text is stored in a time series, and the time interval of the collected voice instructions can be determined from the time interval of the instruction text, so that it can be selected whether to include all the instruction texts in the instructions to be executed, or the instruction texts can be segmented by the time interval, and the required instruction texts can be selectively intercepted as needed. For example, in some embodiments, the preset time length is set to 2S, and when the time interval is greater than 2S, it is determined to be a sentence break, and only the instruction text before the time interval is intercepted as the user's instruction text, and the instruction text that appears later is invalid.

[0028] In this embodiment, Figure 3 As shown, step S4 includes: S401, performing semantic analysis on the instructions in the instruction library based on the large language model LLM to obtain a global vector value of the instruction library; S402, obtaining the instruction text, performing semantic analysis based on the large language model LLM, and obtaining a vector value of the instruction text; S403, traverse and match the vector value of the instruction text with the global vector value of the instruction library to obtain the target instruction corresponding to the instruction text.

[0029] In this embodiment, a large language model LLM is used to perform semantic analysis on the instructions in the instruction library and the user's instruction text, and to perform approximate matching at the same time, so as to match the user's instruction text to the instructions in the instruction library.

[0030] Based on the medical training model, different vital signs are presented in the simulation of illness, intervention, diagnosis and treatment, and other operations, and the intervention, diagnosis and treatment operations have a specific order. At the same time, when the user issues a voice command, multiple commands may be issued at the same time, and there are multi-level commands in a sequence. Therefore, the instructions for controlling the medical training model are relatively complex, and there may be single instructions, multiple instructions, and multi-level instructions. Triggering.

[0031] In this embodiment, Figure 4 As shown, step S403 specifically includes: determining the number of vector values ​​of the instruction text, When the number of vector values ​​of the instruction text is 0, return to step S3; When the number of vector values ​​in the instruction text is non-zero, the global vector values ​​of the instruction library are traversed and matched one by one to determine whether the match is successful. If the match is successful, the instruction in the instruction library that is consistent with the vector value is marked as the target instruction, and Num is counted. If the match is not successful, the instruction text corresponding to the vector value is marked as a redundant instruction.

[0032] When Num is 1, the target instruction is marked as a single instruction; when Num is greater than 1, the target instruction is marked as a multiple instruction.

[0033] Further, it is necessary to determine whether the multiple instructions include a level 2 instruction. If yes, the instruction category and trigger level of the level 2 instruction are marked to be the same as the corresponding level 1 instruction.

[0034] Through this step, it is possible to identify whether the voice command issued by the user is valid or invalid, whether it is a single command or multiple commands. The number of vector values ​​in this step is the number of vector values ​​obtained after semantic analysis by the large language model LLM, which can also be understood as the number of commands recognized by the large language model LLM. When the number is 0, it means that the command is not recognized. When the number is not 0, the number is 1, which indicates a single command. If the number is greater than 1, it indicates multiple commands. By matching the commands in the command library, it is determined whether it is valid. If the match is successful, it is valid, and the number is accumulated. If the match fails, it is a redundant command. In some embodiments, when the number of vector values ​​is 0, and when there are redundant commands, feedback can be set, such as "no valid command is recognized", "the command is not executed", etc., to increase the interactive experience.

[0035] In this embodiment, Figure 5 As shown, step S5 includes: S501, determine whether there is an instruction with a level A trigger mark, if yes, execute the instruction with the level A trigger mark, if no, execute step S502; S502, determine whether there is an instruction with a B-level trigger level mark, if yes, execute the instruction with the B-level trigger level mark, if no, execute step S503; S503, determining whether there is an instruction with a C-level trigger level mark, if yes, executing the instruction with the C-level trigger level mark, if no, executing step S504; S504, determine whether there is an instruction with a D-level trigger level mark, if yes, execute the instruction with the D-level trigger level mark.

[0036] Through this step, the instructions can be triggered in sequence according to the priority order of the trigger level, ensuring the correct step flow of medical record simulation and intervention, diagnosis and treatment operations of the medical training model.

Claims

1. A medical training model control method based on speech recognition, characterized in that: include: S1, build instruction library and divide instruction categories The vocabulary describing the categories of patients’ vital signs in the collected medical records is marked as vital sign parameter instructions; The vocabulary describing the patient's performance in the collected medical records was labeled as model performance instructions; The words describing the patient's condition in the medical records are marked as condition parameter instructions; The words describing the intervention and treatment methods implemented on the patients in the collected medical records are marked as intervention parameter instructions; The words describing the medications given to the patient in the medical records were collected and labeled as medication parameter instructions; S2, specify the trigger level of the instruction category Specifying the triggering level of the physical sign parameter instruction and the model performance instruction as level A; Specifying the trigger level of the condition parameter instruction as level B; Specifying the trigger level of the intervention parameter instruction as level C; Specifying the trigger level of the drug parameter instruction as level D; The trigger priority order of the trigger level is A level greater than B level greater than C level greater than D level; S3, collecting the voice commands issued by the user, and converting the voice commands into command text through speech recognition; S4, performing semantic analysis on the instruction text based on the large language model LLM, matching it with the instructions in the instruction library, and obtaining a target instruction corresponding to the instruction text; S5, executing the instruction operation according to the instruction type and trigger level corresponding to the target instruction.

2. The method for controlling a medical training model based on speech recognition according to claim 1, characterized in that: The vital sign parameter instructions include heart rate, respiratory rate, blood oxygen saturation, end-tidal carbon dioxide, invasive blood pressure, non-invasive blood pressure, and body surface temperature; The model performance instructions include eye state, lip color, forehead sweating, foaming at the mouth, shivering, and difficult labor; The condition parameter instructions include bleeding, airway obstruction, ventricular coarse fibrillation, ventricular fine fibrillation, ventricular flutter, cardiac arrest, hypertension, death, asthma, and respiratory arrest; The intervention parameter instructions include fetal heart monitoring, endotracheal intubation, cesarean section, guided examination, umbilical cord bandaging, placental delivery, and digital rectal examination; The drug parameter instructions include aminophylline, oseltamivir, acetaminophen, dexamethasone, compound amino acids, fentanyl, epinephrine, normal saline, metronidazole, oxytocin, and labetalol.

3. The medical training model control method based on speech recognition according to claim 1, characterized in that: Step S3 includes: S301, monitor the ambient sound in real time, determine whether the volume of the current ambient sound meets the threshold volume, if yes, mark the current ambient sound as valid voice input, collect the current ambient sound as the current voice command, if no, mark the current ambient sound as invalid voice input; S302, converting the current voice instruction into instruction text through voice recognition, and storing it in the instruction queue in time sequence.

4. The method for controlling a medical training model based on speech recognition according to claim 3, characterized in that: The method further includes step S303 of determining whether there is a time interval between the instruction texts in the instruction queue, and when the time interval is greater than a preset time length, segmenting the instruction text to obtain instruction segmented texts.

5. The method for controlling a medical training model based on speech recognition according to claim 1, characterized in that: Step S4 includes: S401, performing semantic analysis on the instructions in the instruction library based on a large language model LLM to obtain a global vector value of the instruction library; S402, obtaining the instruction text, performing semantic analysis based on a large language model (LLM), and obtaining a vector value of the instruction text; S403, traverse and match the vector value of the instruction text with the global vector value of the instruction library to obtain the target instruction corresponding to the instruction text.

6. The medical training model control method based on speech recognition according to claim 5, characterized in that: Step S403 specifically includes: determining the number of vector values ​​of the instruction text, When the number of vector values ​​of the instruction text is 0, return to step S3; When the number of vector values ​​in the instruction text is non-zero, the global vector values ​​of the instruction library are traversed and matched one by one to determine whether the match is successful. If the match is successful, the instruction in the instruction library that is consistent with the vector value is marked as the target instruction, and Num is counted. If the match is not successful, the instruction text corresponding to the vector value that is not successfully matched is marked as a redundant instruction.

7. The method for controlling a medical training model based on speech recognition according to claim 6, characterized in that: Step S403 also includes: When Num is 1, the target instruction is marked as a single instruction; When Num is greater than 1, the target instruction is marked as a multiple instruction.

8. The method for controlling a medical training model based on speech recognition according to claim 7, characterized in that: Also includes: Determine whether the multiple instructions include a level 2 instruction, and if yes, mark the instruction category and trigger level of the level 2 instruction as being the same as the corresponding level 1 instruction.

9. The method for controlling a medical training model based on speech recognition according to claim 1, characterized in that: Step S5 includes: S501, determine whether there is an instruction with a level A trigger mark, if yes, execute the instruction with the level A trigger mark, if no, execute step S502; S502, determine whether there is an instruction with a B-level trigger level mark, if yes, execute the instruction with the B-level trigger level mark, if no, execute step S503; S503, determining whether there is an instruction with a C-level trigger level mark, if yes, executing the instruction with the C-level trigger level mark, if no, executing step S504; S504, determine whether there is an instruction with a D-level trigger level mark, if yes, execute the instruction with the D-level trigger level mark.

Images