Voice recognition-based clinical nursing information checking method and system

Abstract

This invention relates to the field of smart healthcare technology and discloses a method and system for verifying clinical nursing information based on speech recognition. The method includes the following steps: an edge device acquires a nursing item script; responds to a voice wake-up command, driving the terminal to read the script item by item; collects real-time voice recordings and performs beamforming noise reduction, using semantic vectors to accurately extract assessment information and fill in the slots; intelligently identifies missing items and triggers dynamic follow-up questions; summarizes and generates structured confirmation information for repetition; if a modification command is detected, a pointer callback is triggered to jump to re-record, and after confirmation, the information is encrypted and uploaded to the cloud. This invention, through a cloud-edge-device collaborative architecture and breakpoint protection technology, achieves contactless entry and closed-loop verification of nursing records. This solution supports interrupted resume transmission and non-linear logic jumps, effectively solving the pain points of easily interfered clinical operations and cumbersome manual recording. While ensuring aseptic operation, it significantly improves the efficiency and accuracy of clinical document generation.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, specifically to a method and system for verifying clinical nursing information based on voice recognition. Background Technology

[0002] With the rapid development of smart healthcare and "Internet Plus" nursing services, the level of informatization in clinical nursing is constantly improving. Nursing records, as a crucial component of medical documentation, are directly related to medical quality and patient safety due to their timeliness, accuracy, and completeness. Traditional methods of handwriting or touch-screen entry using mobile nursing terminals or tablets are not only cumbersome and time-consuming, consuming a significant amount of nursing staff's energy, but more seriously, frequent contact with electronic screens during aseptic procedures or home visits while wearing gloves can easily break the aseptic barrier, significantly increasing the risk of cross-infection.

[0003] To free up nurses' hands, the industry has gradually begun to introduce input technology based on basic speech recognition. However, existing voice-assisted input solutions have revealed many deep-seated limitations in real and complex clinical applications. Real wards or home care environments are often accompanied by highly dynamic background noise, such as medical instrument alarms and conversations between patients and their families. This noisy environment causes a sharp drop in the signal-to-noise ratio of conventional speech-to-text engines, resulting in extremely high error rates. Furthermore, most existing systems can only achieve simple speech-to-text conversion, often outputting unstructured long texts lacking logical coherence. The system cannot actively and intelligently map and extract these scattered, colloquial descriptions from standardized assessment elements in electronic health records, significantly increasing the subsequent manual processing costs.

[0004] A more significant technical shortcoming lies in the fact that real clinical nursing is a highly dynamic and easily interrupted process. Nurses frequently encounter unexpected situations such as sudden patient inquiries, family interruptions, or emergency treatments during procedures and verbal recording. Existing voice systems generally employ rigid linear recording logic. Once the recording process is unexpectedly interrupted by external factors, the system often crashes or corrupts due to timeouts or picking up irrelevant ambient noise. When nurses attempt to resume recording after handling the emergency, existing systems completely lack effective breakpoint protection mechanisms and contextual memory capabilities, failing to automatically determine which core nursing assessment elements have been recorded and which have been omitted. Furthermore, during the review phase after recording, if an error is found in a specific data point, current technology cannot support non-linear voice rollback and correction for that error. This forces nurses to abandon their current progress and restart the entire recording session, or ultimately still have to manually click on the screen to check and correct the error, violating aseptic principles. This extremely high error correction cost means that existing voice systems not only fail to truly reduce the clinical burden but also reduce the consistency of standardized operations.

[0005] Therefore, there is an urgent need in this field for a clinical nursing information verification solution that can adapt to complex acoustic environments, support dynamic extraction of structured semantics, and possess intelligent anti-interruption and nonlinear logic error correction capabilities, so as to truly achieve a highly efficient and safe closed loop with no contact throughout the entire process. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a clinical nursing information verification method and system based on speech recognition. It solves the technical problems of existing clinical nursing voice recording systems in noisy and easily interrupted real operating environments, which make it difficult to achieve high-precision structured semantic extraction, seamless resume transmission from breakpoints, and nonlinear speech error correction, thus leading to easy damage to aseptic operations and low efficiency in document recording.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a clinical nursing information verification method based on speech recognition, applied in an architecture including a terminal interaction device, an edge device, and a cloud system, the method comprising: The edge device retrieves the nursing project script corresponding to the current nursing task from the cloud system. The nursing project script contains multiple nursing assessment elements. Receive a wake-up voice command collected through the terminal interaction device, and trigger the edge device to broadcast voice prompts item by item according to the nursing project script; Receive real-time voice data based on the voice prompt information, extract assessment information from the real-time voice data, and match it to the corresponding nursing assessment elements; Determine whether there are any missing items among the multiple nursing assessment elements that have not been matched. If there are missing items, generate follow-up question voice prompts and broadcast them through the terminal interaction device, and receive supplementary voice real-time data to fill in the missing items. Structured confirmation information is generated based on the fully matched nursing assessment elements, and is then repeated and broadcast through the terminal interaction device. Upon receiving a confirmation voice command, the complete clinical nursing record is uploaded to the cloud system.

[0008] As a further refinement of the technical solution and explanation of the technical principle of the present invention, the method includes the following specific mechanisms during implementation: In the data synchronization and verification pre-stage, the system first establishes an encrypted channel between the edge device and the cloud system to synchronize the patient's electronic health record and the latest version of the nursing project script. Responding to the wake-up voice command collected through the terminal interaction device, the system initiates the pre-operation verification process according to the script, generating and broadcasting verification voice prompts containing basic patient information (such as bed number, name, etc.). Upon receiving a confirmation of correct verification based on these voice prompts, the system records the start time and geographical location information of the current nursing task, and then begins executing the step-by-step broadcasting of voice prompts, thereby ensuring the correct time, correct location, and correct patient.

[0009] In the complex environment speech acquisition and channel activation stage, to overcome clinical environmental noise, this invention uses a beamforming module built into the terminal interaction device to perform directional beamforming and background noise filtering on the acquired environmental audio. Assume the microphone array includes... The array element, the first The ambient audio signal received by each array element is The output signal after directional beamforming Represented as: in, For the corresponding amplitude weighting coefficient, This is the phase delay used to compensate for the time difference in arrival of the sound source at each array element. When a wake-up voice command containing a preset wake-up word is detected in the filtered high signal-to-noise ratio audio, the system immediately activates the low-power Bluetooth data transmission channel between the terminal interaction device and the edge device to trigger subsequent item-by-item broadcasting and data transfer.

[0010] During the semantic extraction and slot matching stage, the edge device's built-in local speech recognition engine performs local semantic recognition on the incoming live speech data and converts it into text data. The business process rule engine then parses this text data. The system maps the extracted text data into a multi-dimensional semantic feature vector. And map each pre-set nursing assessment element in the nursing project script into a target feature vector. By calculating the cosine similarity between the two... : When similarity When the value exceeds the preset confidence threshold, the corresponding nursing observation indicators are accurately extracted as evaluation information, filled into the corresponding preset slots in the script, and previewed in real time on the screen of the edge device.

[0011] During the missing item follow-up and breakpoint protection phase, the system dynamically scans the preset slots corresponding to nursing assessment elements in the nursing project script to detect any empty slots without filled assessment information. If an empty slot exists, it is marked as a missing item, and a natural language generation template is invoked to dynamically generate a follow-up voice prompt based on the assessment element corresponding to the missing item (e.g., "Blood pressure vasopressors were not recorded; please supplement."). After this prompt is broadcast via a terminal interactive device, the system receives the supplementary voice data and performs the same semantic recognition and feature matching extraction on it, accurately filling the empty slot with the extracted supplementary assessment information to complete the filling. Simultaneously, throughout the entire data reception process, the system calculates the short-time energy of the processed audio. : in For discretized audio signals, This is a sliding window function. If short-time energy is detected... If the ambient noise threshold is exceeded, or a preset global control voice command (such as pausing recording) is detected, an interruption and suspension operation is triggered, saving the recording progress status (i.e., fill indicator and pointer) of the current preset slot as a breakpoint node. When a wake-up command is received again, the breakpoint node status is read, a recovery voice prompt containing matched evaluation information is generated and played, and recording can be seamlessly resumed from that breakpoint node upon receiving a continue recording command.

[0012] During the data verification and closed-loop upload phase, once all nursing assessment elements are matched, the system combines and splices all assessment information to generate structured confirmation information for repetition and confirmation, and issues a confirmation request via terminal broadcast. The system parses the voice feedback data in response to this request: if it recognizes a modification instruction for a specific nursing assessment element (such as changing the heart rate to 80), the system pointer returns to execute the broadcast operation corresponding to that specific element, re-receives data and extracts assessment information, until updated structured confirmation information is generated again for repetition and broadcast; if it recognizes a confirmation voice instruction (such as confirming that there is no error), the data containing all confirmed assessment information is packaged, encrypted, and uploaded to the cloud system as a complete clinical nursing record.

[0013] A second aspect of the present invention provides a clinical nursing information verification system based on speech recognition, applied in an architecture including a terminal interaction device, an edge device, and a cloud system, the system comprising: The script acquisition module is used to acquire the nursing project script corresponding to the current nursing task from the cloud system through the edge device. The nursing project script contains multiple nursing assessment elements. The voice wake-up module is used to receive wake-up voice commands collected through the terminal interaction device and trigger the edge device to broadcast voice prompts item by item according to the nursing project script; The data matching module is used to receive real-time voice data based on the voice prompt information, extract assessment information from the real-time voice data, and match it to the corresponding nursing assessment elements. The follow-up questioning module is used to determine whether there are any missing items in the multiple nursing assessment elements based on the nursing project script. If there are missing items, a follow-up questioning voice prompt is generated and broadcast through the terminal interaction device, and supplementary voice real-time data is received to fill in the missing items. The closed-loop confirmation module is used to generate structured confirmation information based on the fully matched nursing assessment elements, repeat and broadcast it through the terminal interaction device, and parse the voice feedback data for the structured confirmation information. After recognizing the confirmation voice command, the module uploads the complete clinical nursing record to the cloud system.

[0014] A third aspect of the present invention provides an electronic device, comprising: a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program to implement the clinical nursing information verification method based on speech recognition as described in the first aspect of the present invention.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the clinical nursing information verification method based on speech recognition as described in the first aspect of the present invention.

[0016] This invention provides a method and system for verifying clinical nursing information based on speech recognition. It has the following beneficial effects: 1. This invention constructs a cloud-edge-device collaborative architecture, distributing nursing scripts to edge devices and relying on voice commands to drive item-by-item playback and semantic feature extraction and matching. This contactless interaction mode completely frees the hands of nursing staff, enabling them to perform highly aseptic operations such as intravenous puncture and catheterization without touching paper, pen, or tablet devices. This not only significantly shortens the time required for a single task but also physically cuts off the transmission route of cross-infection within the hospital.

[0017] 2. Addressing the pain point of traditional oral record-keeping easily overlooking details, this solution introduces a dynamic follow-up questioning mechanism based on empty slot scanning. The system compares the extracted speech content with preset nursing assessment elements in real time in the background. Once an unfilled item is detected, it immediately uses a natural language template to generate targeted follow-up questions and proactively asks them. This process moves post-event document quality control to the operational site, ensuring the absolute integrity of clinical data and medical compliance from the source.

[0018] 3. Considering the unpredictable nature of real emergency or ICU environments, this invention specifically designs a breakpoint protection strategy based on environmental acoustic analysis. By calculating short-term audio energy in real time, the system can instantly freeze the current slot pointer and recording progress when encountering sudden high-decibel noise or receiving a suspension command; after the environment recovers and the system is reactivated, the user can listen to the recorded information summary and seamlessly resume the transmission, perfectly matching the fragmented and easily interrupted work rhythm in high-pressure clinical scenarios.

[0019] 4. To overcome the physical limitations of noisy hospital rooms on speech recognition accuracy, this technology integrates array microphones and directional beamforming algorithms into the underlying interactive terminal. By performing phase delay compensation and amplitude weighting on multi-channel audio signals, the system can accurately locate the target sound source in complex sound field environments, strongly suppressing background noise and mechanical background noise. This hardware-level signal-to-noise ratio improvement eliminates the operational frustration caused by frequent misidentification, making subsequent semantic parsing more sensitive and reliable.

[0020] 5. This solution constructs a rigorous data verification loop through pointer callback technology. When generating structured information retelling after element matching, if a nursing staff member discovers a verbal error in a value and issues a correction instruction, the system pointer will precisely backtrack to that single element node to re-collect the data until it is finally confirmed to be correct before packaging, encrypting, and uploading it to the cloud. This error-proofing design, which allows for immediate correction of errors and provides feedback at every step, minimizes human memory bias and reduces the medical error rate in clinical information entry to an extremely low level. Attached Figure Description

[0021] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is the main flowchart of the method of the present invention; Figure 3 This is a schematic diagram of the entire intelligent interactive scenario of the present invention; Figure 4 This is a schematic diagram of the semantic feature vector mapping and slot dynamic matching logic of the present invention; Figure 5 This is a schematic diagram of the breakpoint protection and pointer callback state machine of the present invention; Figure 6 This is a block diagram of the hardware structure of the electronic device of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figures 1-6 This invention provides a method and system for verifying clinical nursing information based on speech recognition. The method for verifying clinical nursing information based on speech recognition is applied to an architecture that includes terminal interaction devices, edge devices, and cloud systems.

[0024] The cloud system serves as the core data hub, responsible for storing and maintaining global electronic health records for patients, as well as various standardized nursing project scripts.

[0025] Edge devices are deployed at the front end of clinical operations. They integrate a local speech recognition engine and a business process rule engine, which are responsible for core semantic parsing, logical judgment, and real-time preview of the screen UI.

[0026] The terminal interaction device is worn or configured on the caregiver's end, with a built-in beamforming noise reduction module and audio acquisition array, used to capture high signal-to-noise ratio live voice data in complex physical environments, and to interact with edge devices through wireless communication links.

[0027] The speech recognition-based clinical nursing information verification method may include the following steps: The nursing project script corresponding to the current nursing task is obtained from the cloud system through the edge device. The nursing project script contains multiple nursing assessment elements.

[0028] During this execution phase, an encrypted channel is established between the edge device and the cloud system to synchronize the patient's electronic health record and the latest version of the nursing project script in encrypted form. This script predefines all key observation indicators and operational procedures required for the current nursing task.

[0029] It receives wake-up voice commands collected through terminal interaction devices, triggering edge devices to broadcast voice prompts item by item according to the nursing project script.

[0030] The terminal interaction device is in real-time monitoring mode. When it collects and recognizes a preset wake word, it activates the data transmission channel with the edge device. The edge device then initiates the pre-operation verification process, sequentially generating verification voice prompts containing the patient's basic information and project requirements, and broadcasting them aloud through the terminal interaction device.

[0031] Receive real-time voice data based on voice prompts, extract assessment information from the real-time voice data, and match it to the corresponding nursing assessment elements.

[0032] The local speech recognition engine converts the received live speech data into text data, and the business process rule engine performs logical parsing and feature extraction on the text data. Specifically, the system maps the extracted text data into multi-dimensional semantic feature vectors. Simultaneously, the pre-defined nursing assessment elements in the nursing project script are mapped to target feature vectors. The system calculates the cosine similarity between the two. To determine the relationship between data and elements: in, This represents the text feature vector after the actual recorded speech has been converted. The first character in the script Standard feature vectors of each nursing assessment element This represents the numerical similarity between the two in a multidimensional semantic space. and These represent the magnitudes of the corresponding eigenvectors.

[0033] When the calculated similarity When the value exceeds the system's preset confidence threshold, the corresponding nursing observation indicators are accurately extracted as evaluation information and automatically filled into the corresponding preset slots in the script.

[0034] It determines whether there are any missing items among multiple nursing assessment elements that have not been matched. If there are missing items, it generates follow-up question voice prompts and broadcasts them through the terminal interactive device, and receives supplementary voice real-time data to fill in the missing items.

[0035] The system dynamically scans the preset slot matrix in the current nursing project script. When an unfilled empty slot is detected, the system marks it as a missing item and calls the underlying natural language generation template to dynamically generate targeted follow-up question voice prompts based on the missing item attributes. The terminal interactive device broadcasts the prompt and collects supplementary voice data. After the same semantic parsing and feature vector matching process, the supplemented and extracted assessment information is accurately written into the corresponding empty slot.

[0036] Structured confirmation information is generated based on the complete matching nursing assessment elements, which is then repeated and broadcast through the terminal interactive device. Upon receiving the confirmation voice command, the complete clinical nursing record is uploaded to the cloud system.

[0037] After all preset slots are filled with data, the edge device splices the full assessment information according to a preset structured protocol. The terminal interaction device then generates structured confirmation information and sends a final confirmation request. The system parses the voice feedback data in response to this request. If the final confirmation voice command is extracted and recognized from the voice feedback data, the data packet containing all confirmed assessment information is encrypted and encapsulated, and uploaded to the cloud system as a complete clinical nursing record, completing the data loop for the current nursing task.

[0038] In one embodiment of the present invention, to ensure the absolute security of clinical data interaction and the accuracy of object verification, the edge device first establishes a two-way encrypted channel with the cloud system during the acquisition of nursing project scripts. This channel is based on a preset secure transmission protocol to ensure the confidentiality and integrity of the patient's electronic health record and the latest version of the nursing project script during synchronization.

[0039] In this invention, after the terminal interaction device receives the wake-up voice command, it does not directly enter the nursing data entry stage. Instead, the edge device responds to the command and prioritizes initiating the pre-operation verification process. The edge device extracts the patient's electronic health record synchronized from the cloud, generates a verification voice prompt containing the patient's basic information, such as bed number and name, and broadcasts it through the terminal interaction device.

[0040] After receiving a verification instruction from the nursing staff based on voice prompts, the system automatically acquires and records the start timestamp of the current nursing task via edge devices, while simultaneously using positioning components to obtain the current geographical location information. This mechanism ensures the traceability of nursing actions in both temporal and spatial dimensions. Only after completing the aforementioned identity verification and spatiotemporal evidence storage is the edge device allowed to execute the subsequent item-by-item broadcast process according to the nursing project script.

[0041] In one specific embodiment of the present invention, in response to the common medical device beeping sounds, footsteps, and other background noises in clinical environments, the terminal interaction device uses a built-in microphone array and beamforming noise reduction module to perform front-end processing of the environmental audio.

[0042] In this invention, the beamforming module employs directional beamforming technology to process the acquired ambient audio. Assume the microphone array includes... The array element, the first Individual elements The ambient audio signal received at all times is The output signal after directional beamforming by the beam noise reduction module Represented as: In the above formula, This represents the target audio output signal after enhancement processing. This represents the total number of elements in the microphone array. Representing the The raw environmental audio signal captured by each array element This represents the amplitude weighting coefficient set to suppress sidelobe interference. This represents the preset phase delay used to compensate for the time difference between the speaker's voice arriving at each array element.

[0043] Through the aforementioned directional beamforming process, the beam denoising module can significantly suppress background noise in non-target areas, thereby improving the signal-to-noise ratio of the target speech.

[0044] In this invention, the terminal interaction device continuously monitors the filtered audio stream in real time. When a wake-up voice command containing a preset wake-up word is detected in the processed audio, the terminal interaction device immediately activates its Bluetooth data transmission channel with the edge device.

[0045] The activation of this channel signifies that the terminal interactive device has switched from a low-power standby state to a full-duplex communication state, thereby triggering the edge device to begin broadcasting voice prompts item by item according to the nursing project script. This mechanism, which triggers physical channel activation based on voice wake-up, effectively reduces the device's power consumption in non-operating states while ensuring the real-time nature of voice data transmission.

[0046] In one embodiment of the present invention, the activation of the Bluetooth data transmission channel is also accompanied by a handshake protocol verification to ensure that the pairing relationship between the terminal interaction device and the edge device is correct. This hardware-level logical association, together with the aforementioned software logic verification, constitutes multiple layers of security protection, ensuring that the process of collecting nursing information not only complies with clinical aseptic operation standards but also meets stringent medical data quality control requirements.

[0047] This spatial beamforming-based directional acquisition technology, combined with a rigorous pre-verification mechanism, enables the invention to maintain an extremely high speech recognition trigger rate and task initiation compliance even in noisy and highly dynamic clinical scenarios, providing a clean and reliable data source for subsequent structured information extraction.

[0048] In one embodiment of the present invention, after receiving the live audio data processed by the terminal interaction device through noise reduction, the edge device uses its built-in local speech recognition engine to process the audio signal in real time. In this invention, the use of a local speech recognition engine effectively avoids latency caused by network fluctuations, ensuring high real-time performance and data security in the conversion of speech to text data in a medical setting.

[0049] In this invention, the converted text data is input into a business process rule engine for deep analysis. This rule engine does not simply perform keyword retrieval, but rather uses pre-defined semantic parsing logic to map unstructured natural language text into observational indicators with medical business implications. Specifically, the system transforms the text data into multi-dimensional semantic feature vectors. The pre-defined nursing assessment elements in the nursing project script are mapped to corresponding target feature vectors. .

[0050] In this invention, the system measures the degree of matching between the text to be identified and each evaluation element in a multidimensional semantic space by calculating the cosine similarity between the two. To perform feature matching, the calculation formula is expressed as follows: In the above formula, This represents the converted text data and the first one in the script. The cosine similarity between nursing assessment elements is such that the closer the value is to 1, the more consistent the semantic intent of the two elements is. This represents the semantic feature vector generated after mapping the currently input text data; The first pre-set in the nursing project script The target feature vector corresponding to each nursing assessment element; Represents the dot product of two eigenvectors; and These represent the magnitudes of the two feature vectors in the multidimensional semantic space.

[0051] In one embodiment of the present invention, the business process rule engine calculates the cosine similarity between the current text feature vector and the feature vectors of all candidate options in the script, and extracts the item with the highest similarity value as the potential matching result. If the maximum similarity... If the confidence threshold is exceeded, the content of the current voice feedback is determined to correspond to the first... Nursing assessment elements.

[0052] In this invention, once the corresponding nursing assessment element is determined, the business process rule engine further extracts specific numerical values ​​or status descriptions from the text data as assessment information. Subsequently, the system accurately fills the extracted assessment information into the preset slot corresponding to the nursing assessment element in the nursing project script. This slot is a logical unit that establishes a mapping relationship with medical indicators during the script initialization phase and is used to store temporarily entered structured data.

[0053] In this invention, to provide real-time visual feedback and assist nursing staff in on-site verification, the screen of the edge device synchronously displays the execution progress of the current nursing project script in real time. As voice recognition and slot matching alternate, the preview interface on the screen dynamically presents the filling status of each slot, instantly displaying the extracted assessment information. This instant-on-speech interaction allows nursing staff to confirm whether the system has correctly interpreted their dictated clinical indicators while keeping their hands on the machine, thus establishing the first line of defense for accuracy during the data entry stage.

[0054] In one specific embodiment of the present invention, the construction process of the semantic feature vector combines a set of clinical nursing terminology with contextual features. This means that the system can identify different colloquial expressions of the same observation indicator. By uniformly projecting these heterogeneous texts onto the same semantic space coordinate system, the robustness of the slot matching algorithm to different nurses' pronunciation habits and expression styles is ensured.

[0055] This slot-filling mechanism based on the vector space model frees the invention from dependence on a fixed instruction set, greatly improving the naturalness and intelligence of clinical nursing records during the collection process, and laying a precise data foundation for subsequent automated quality control and complete medical record generation.

[0056] In one embodiment of the present invention, during the execution of the nursing project script, the system dynamically scans and monitors the logical status of each preset slot in real time. In this invention, after the edge device completes one round of semantic matching, it automatically retrieves the preset slot matrix corresponding to each nursing assessment element in the nursing project script and detects whether there are empty slots without assessment information. If the system identifies a specific slot as empty, it immediately marks it as a missing item and calls the underlying preset natural language generation template.

[0057] In this invention, the system dynamically generates targeted follow-up question voice prompts based on the nursing assessment element attributes associated with the missing item. After the follow-up question voice prompt is broadcast through the terminal interaction device, the system enters the supplementary data collection state. Upon receiving supplementary voice real-time data in response to the follow-up question, the edge device restarts the aforementioned semantic recognition, feature vector mapping, and cosine similarity calculation process to accurately backtrack the extracted supplementary assessment information and fill it into the corresponding empty slot. This dynamic follow-up questioning mechanism based on slot status awareness ensures the integrity of clinical nursing records during the collection process and avoids medical document defects caused by omissions.

[0058] In one specific embodiment of the present invention, to ensure the reliability of data entry in clinical emergencies or noisy environments, the terminal interaction device and the edge device implement real-time acoustic monitoring and interruption protection strategies throughout the data interaction process. In this invention, the system continuously performs discrete sampling on the audio stream after front-end noise reduction processing and calculates the short-time energy of the audio signal. The formula for calculating short-time energy is expressed as follows: In the above formula, The short-time energy value representing the audio frame at time t reflects the intensity of energy fluctuations in the audio signal in the time domain; Represents the discrete audio sample signal of the input; Representative of the first A sliding window function centered at a specific time point is used to determine the time window span for energy calculation; This represents the index value of a discrete sampling point in the time series.

[0059] In this invention, the edge device is preset with an environmental noise energy threshold, and the calculated short-term energy is displayed in real time. The data is compared with this threshold. If the system detects that the short-term energy of multiple consecutive sampling frames continuously exceeds the preset environmental noise energy threshold, or if the local speech recognition engine recognizes a global control voice command issued by the caregiver, the system will immediately respond and trigger an interruption and suspension operation.

[0060] In this invention, after triggering an interruption suspension operation, the system automatically saves the current entry progress status of the nursing project script. This progress status specifically includes the identifier matrix of filled slots, the current broadcast sequence pointer, and the extracted assessment information data. This information is stored as breakpoint nodes in the system's non-volatile memory or cache, ensuring that the current verification progress is not lost in abnormal conditions or under human intervention.

[0061] In one embodiment of the present invention, once environmental interference is eliminated or the caregiver has completed handling an emergency medical task, the recovery process can be triggered by issuing a wake-up voice command to the terminal interaction device again. In this invention, the edge device responds to the wake-up command by automatically reading and parsing the recording progress status in the previously saved breakpoint nodes. The system then generates a recovery voice prompt containing matched assessment information and broadcasts it, informing the caregiver of the current recorded progress summary. Upon receiving a continue recording command, the system points the broadcast sequence pointer to the specific element node where the interruption occurred, seamlessly resuming subsequent process execution from that breakpoint node.

[0062] This adaptive control mechanism, based on short-term energy detection and breakpoint state saving, enables the present invention to maintain precise control over the progress of nursing records in complex and ever-changing clinical scenarios such as emergency scenes and ICU alarms. This breakpoint-resumption recording method greatly reduces the burden of repetitive verification caused by environmental interference or task interruptions, providing a fundamental technical guarantee for the efficient continuity of medical procedures.

[0063] In one embodiment of the present invention, when the edge device detects that all preset nursing assessment elements in the nursing project script have been matched and filled, the system automatically triggers a data aggregation and structured processing flow. In the present invention, the business process rule engine extracts the assessment information from each preset slot and performs string concatenation and formatted encapsulation according to a preset clinical document logic protocol to generate structured confirmation information for restatement and confirmation.

[0064] In this invention, the terminal interaction device receives and broadcasts the structured confirmation information. During the broadcast, the system maintains real-time monitoring and parsing of the voice feedback data. If the system identifies the voice feedback data for the structured confirmation information as a modification instruction, the edge device initiates a pointer callback mechanism.

[0065] In this invention, the pointer callback mechanism maintains a logical pointer. To locate the nursing assessment element node for the current operation. Assume the nursing project script contains... The nursing assessment elements constitute the set of elements. When the system interprets voice feedback containing a specific intention to modify, it determines the index value of the element to be modified using a semantic extraction algorithm. At this point, the system executes the pointer jump logic, setting the current pointer... Update to target index value : In the above formula, This represents the position of the target element pointer after the callback; This represents the identifier of the element to be modified identified from the voice feedback data; A mapping function between feature identifiers and script index sequences.

[0066] In this invention, once the pointer executes the callback, the system will immediately return to execute the indexing of that specific feature. The corresponding broadcast operation prompts the nursing staff to re-enter the data. Subsequently, the system re-receives the voice data of that node and extracts and overwrites the assessment information. After updating the data of that specific slot, the system will regenerate structured confirmation information based on the updated set of elements and broadcast it again through the terminal interaction device, forming a dynamic error correction loop.

[0067] In one embodiment of the present invention, if the voice feedback data parsed by the system is a preset confirmation voice command, it is determined that the data entry for the current nursing task has passed manual review. In this invention, the edge device digitally signs and encrypts data packets containing all confirmed assessment information. This digital signature mechanism ensures the originality and non-repudiation of clinical records during the upload to the cloud system, while the encryption ensures the confidentiality of medical data during network transmission.

[0068] In this invention, after a complete clinical nursing record is uploaded to the cloud system, the cloud system sends a successful upload receipt to the edge device. Upon receiving the receipt, the edge device clears the breakpoint nodes and temporary slot data in its local cache for this task, and then officially closes the execution thread of the current nursing project script.

[0069] This pointer-based closed-loop confirmation mechanism allows nurses to quickly correct specific items at the end of a task using simple natural language commands, without having to go through the entire verification process again. This non-linear logical interaction method ensures extremely high accuracy in medical data entry while greatly improving the flexibility of clinical document generation, effectively solving the technical pain point of traditional systems where modifications to correct verbal slips or input errors are cumbersome.

[0070] In one specific embodiment of the present invention, the system also logs the frequency and content of modification instructions, serving as reference data for subsequent clinical operation quality assessment. Through this closed-loop design of restatement-correction-confirmation, the present invention constructs a rigorous error-proofing and omission-proofing system at the technical level, ensuring that every nursing record uploaded to the cloud system undergoes dual verification by human-machine collaboration.

[0071] In one embodiment of the present invention, a clinical nursing information verification system based on speech recognition is provided. This system serves as the logical carrier for implementing the aforementioned verification method and is applied in a physical architecture including terminal interaction devices, edge devices, and cloud systems. In this invention, the system achieves fully automated processing from raw audio acquisition to the uploading of structured nursing records through the collaborative interaction of functional modules.

[0072] In this invention, the system includes a script acquisition module, which is configured to access the database of the cloud system via an edge device to synchronously acquire nursing project scripts that match the current nursing staff's identity and shift schedule. These nursing project scripts logically consist of multiple nursing assessment elements, defining the characteristic indicators that must be collected during clinical operations.

[0073] In this invention, the system also includes a voice wake-up module, which is used to monitor the ambient audio stream collected by the terminal interactive device in real time. When a preset wake-up semantic feature is recognized, the voice wake-up module triggers the edge device to enter the working state and drives the speech synthesis engine to broadcast guiding voice prompts item by item according to the preset sequence of the nursing project script, guiding the nursing staff to provide data feedback.

[0074] In this invention, the system further includes a data matching module, the core function of which is to perform semantic-level parsing and classification of the received live audio data. The data matching module calls the local speech recognition engine to convert audio into text and uses a vector space model to perform matching tasks. In a specific embodiment of this invention, this module calculates text feature vectors. With target feature vector Similarities between them: In the above formula, The score represents the degree of matching between the text and the evaluation elements. This is the semantic vector of the input text. This refers to the feature vectors of pre-defined elements in the script. When... When the preset judgment conditions are met, the data matching module will accurately write the extracted evaluation information into the corresponding logical slot.

[0075] In this invention, the system also integrates a follow-up questioning module, which has the ability to scan the execution status of nursing project scripts in real time. The follow-up questioning module traverses all preset slots to identify any missing items that have not been matched. Once a missing item is found, the module immediately triggers natural language generation logic to construct targeted follow-up question prompts and delivers them to the terminal interaction device for broadcast, thereby ensuring the completeness of clinical records through a supplementary data entry mechanism.

[0076] In this invention, the system also includes a closed-loop confirmation module, which is configured to execute verification logic after all assessment elements have been matched. The closed-loop confirmation module aggregates the assessment information scattered in each slot into structured confirmation information and parses the voice commands fed back by the nursing staff. If a modification intention is detected, the module triggers pointer callback logic to control the system state machine to fall back to a specific broadcast node for re-recording; if a confirmation command is detected, the encrypted record package is uploaded to the cloud system.

[0077] In one embodiment of the present invention, an electronic device is also provided, which can be implemented as a hardware form of an edge device or a terminal interaction device. In the present invention, the electronic device includes at least a processor and a memory. The memory stores computer program instructions capable of implementing the above-described clinical nursing information verification method; the processor is coupled to the memory and configured to execute these program instructions, thereby controlling the hardware circuitry to implement core technical aspects such as beamforming, cosine similarity calculation, and breakpoint node saving.

[0078] In one specific embodiment of the present invention, the internal architecture of the electronic device interconnects various hardware components through a bus system. During program execution, the processor exchanges full-duplex data with an external cloud server or portable acquisition terminal through a communication interface. This hardware configuration ensures parallel processing capabilities for large-scale semantic operations, supporting real-time speech conversion and feature matching requirements in high-sampling-rate environments.

[0079] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by the processor of an electronic device, it can drive the hardware system to implement the various steps of the speech recognition-based clinical nursing information verification method described in the first aspect. In this invention, the storage medium can employ non-volatile storage technology to ensure that the stored nursing item scripts and system operating logic are not lost in the event of a power outage or sudden failure, providing reliable media support for the continuous collection of clinical nursing records.

[0080] The system architecture, electronic devices, and storage media of this invention together constitute a complete technical system from underlying computing power to high-level logic. By combining functional modularity with hardware execution efficiency, this invention not only solves the robustness problem of voice acquisition in noisy clinical environments, but also significantly reduces the error rate of medical document entry through a rigorous logical error correction mechanism, providing a solid technical guarantee for the construction of digital wards.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for verifying clinical nursing information based on speech recognition, characterized in that, Includes the following steps: The method, applied in an architecture including a terminal interaction device, an edge device, and a cloud system, includes: obtaining a nursing project script corresponding to the current nursing task from the cloud system via the edge device, the nursing project script containing multiple nursing assessment elements; receiving a wake-up voice command collected via the terminal interaction device, triggering the edge device to play voice prompts item by item according to the nursing project script; receiving voice real-time data based on the voice prompts, extracting assessment information from the voice real-time data, and matching it to the corresponding nursing assessment elements; determining whether there are any missing items among the multiple nursing assessment elements that have not been matched, and if there are missing items, generating follow-up voice prompts and playing them via the terminal interaction device, and receiving supplementary voice real-time data to fill in the missing items; generating structured confirmation information based on the fully matched nursing assessment elements, repeating and playing it via the terminal interaction device, and uploading the complete clinical nursing record to the cloud system after receiving a confirmation voice command.

2. The clinical nursing information verification method based on speech recognition according to claim 1, characterized in that, The step of obtaining the nursing project script corresponding to the current nursing task from the cloud system through the edge device includes: establishing an encrypted channel between the edge device and the cloud system, and synchronizing the patient's electronic health record and the latest version of the nursing project script; the step of receiving a wake-up voice command collected through the terminal interaction device and triggering the edge device to broadcast voice prompts item by item according to the nursing project script includes: responding to the wake-up voice command collected through the terminal interaction device, starting the pre-operation verification process according to the nursing project script, generating and broadcasting verification voice prompts containing basic patient information; after receiving a verification correct instruction based on the verification voice prompt feedback, recording the start time and geographical location information of the current nursing task, and starting to execute the step of broadcasting voice prompts item by item according to the nursing project script.

3. The clinical nursing information verification method based on speech recognition according to claim 1, characterized in that, The step of receiving real-time voice data based on the voice prompt information, extracting evaluation information from the real-time voice data, and matching it to the corresponding nursing evaluation elements includes: performing local semantic recognition on the real-time voice data using the local speech recognition engine built into the edge device, converting the real-time voice data into text data; calling the business process rule engine to parse the text data, extracting the content of nursing observation indicators corresponding to the nursing evaluation elements as the evaluation information; filling the extracted evaluation information into the preset slots in the nursing project script corresponding to the nursing evaluation elements, and previewing it in real time on the screen of the edge device.

4. The clinical nursing information verification method based on speech recognition according to claim 3, characterized in that, The process of determining whether there are any missing items among the multiple nursing assessment elements that have not been matched, and if such missing items exist, generating a follow-up question voice prompt and broadcasting it through the terminal interaction device, and receiving supplementary voice data to fill the missing items, includes: scanning the preset slots in the nursing project script corresponding to the nursing assessment elements, detecting whether there are empty slots that have not been filled with assessment information; when an empty slot is detected, marking the empty slot as the missing item, and dynamically generating the follow-up question voice prompt according to the nursing assessment element corresponding to the missing item; broadcasting the follow-up question voice prompt through the terminal interaction device, and after receiving the supplementary voice data, performing semantic recognition and extraction on the supplementary voice data, and filling the empty slot with the extracted assessment information to fill the missing items.

5. The clinical nursing information verification method based on speech recognition according to claim 1, characterized in that, The process of generating structured confirmation information based on the fully matched nursing assessment elements, repeating and broadcasting it through the terminal interaction device, and uploading the complete clinical nursing record to the cloud system after receiving a confirmation voice command includes: after all the nursing assessment elements are matched, combining and splicing all the assessment information to generate structured confirmation information for repetition and confirmation; repeating and broadcasting the structured confirmation information through the terminal interaction device and issuing a confirmation request; parsing the voice feedback data in response to the confirmation request, and if a modification instruction for a specific nursing assessment element is identified, returning to the step of repetitively broadcasting the voice prompt information to execute the broadcast operation corresponding to the specific nursing assessment element, so as to re-receive data and extract assessment information until matching is completed and structured confirmation information is generated again for repetition and broadcast; if the confirmation voice command is identified from the voice feedback data, packaging and encrypting the data containing all the confirmed assessment information, and uploading it to the cloud system as the complete clinical nursing record.

6. The clinical nursing information verification method based on speech recognition according to claim 1, characterized in that, The steps of receiving voice real-time data based on the voice prompt information, extracting assessment information from the voice real-time data, and matching it to the corresponding nursing assessment elements further include: during the process of receiving the voice real-time data, if a preset global control voice command is detected or environmental noise exceeding a preset threshold is identified, an interruption and suspension operation is triggered, and the current input progress status of the nursing project script is saved as a breakpoint node; when a wake-up voice command is received again, the input progress status at the breakpoint node is read, a recovery voice prompt containing the matched assessment information is generated and played, and after receiving a continue input command, the subsequent process continues from the breakpoint node.

7. The clinical nursing information verification method based on speech recognition according to claim 1, characterized in that, The step of receiving the wake-up voice command collected through the terminal interaction device and triggering the edge device to broadcast voice prompts item by item according to the nursing item script includes: performing directional beamforming and background noise filtering on the collected ambient audio through the beam noise reduction module built into the terminal interaction device; when the wake-up voice command containing a preset wake-up word is detected in the filtered audio, activating the Bluetooth data transmission channel between the terminal interaction device and the edge device to trigger the edge device to broadcast the voice prompts item by item according to the nursing item script.

8. A clinical nursing information verification system based on speech recognition, applied in an architecture including terminal interaction devices, edge devices, and cloud systems, characterized in that, The system includes: a script acquisition module, used to acquire the nursing project script corresponding to the current nursing task from the cloud system via the edge device, the nursing project script containing multiple nursing assessment elements; a voice wake-up module, used to receive a wake-up voice command collected via the terminal interaction device, triggering the edge device to play voice prompts item by item according to the nursing project script; a data matching module, used to receive voice real-time data based on the voice prompts, extract assessment information from the voice real-time data, and match it to the corresponding nursing assessment elements; a follow-up questioning module, used to determine whether there are any missing items in the multiple nursing assessment elements that have not been matched based on the nursing project script, and if there are missing items, generate follow-up questioning voice prompts to be played via the terminal interaction device, and receive supplementary voice real-time data to fill in the missing items; and a closed-loop confirmation module, used to generate structured confirmation information based on the fully matched nursing assessment elements, repeat and play it via the terminal interaction device, and parse the voice feedback data for the structured confirmation information, and upload the complete clinical nursing record to the cloud system after recognizing the confirmation voice command.

9. An electronic device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program to implement the clinical nursing information verification method based on speech recognition as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the clinical nursing information verification method based on speech recognition as described in any one of claims 1-7.

Images