Search

CN-121983032-A - Auxiliary processing method and system for mother and infant room voice recognition based on knowledge graph

CN121983032ACN 121983032 ACN121983032 ACN 121983032ACN-121983032-A

Abstract

The invention provides a mother and infant room voice recognition auxiliary processing method and system based on a knowledge graph, which are used for constructing and optimizing a light-weight mother and infant care knowledge graph, developing voice interaction and visual assistance in stages, constructing voice question-answering, authorization verification, visual guidance and feedback optimization flows, constructing an occupied state monitoring and service area linkage mechanism, constructing a real-time monitoring, multi-terminal feedback and intelligent scheduling system, constructing a simulation environment test for 72 hours of stability, inviting a user to test and optimizing details, and constructing a full-link privacy protection and safety guarantee mechanism.

Inventors

  • TANG QING
  • WEI XUEXIA
  • TANG ZHU

Assignees

  • 湖南贝林母婴服务有限公司

Dates

Publication Date
20260505
Application Date
20251212

Claims (10)

  1. 1. A mother and infant room voice recognition auxiliary processing method based on a knowledge graph is characterized by comprising the following steps: Constructing and optimizing a light-weight care knowledge graph of the mother and infant; developing voice interaction and visual assistance in stages, and constructing a voice question-answering, authorization verification, visual guidance and feedback optimization flow; constructing an occupancy state monitoring and service area linkage mechanism, and constructing a real-time monitoring, multi-terminal feedback and intelligent scheduling system; building a simulation environment to test the stability for 72 hours, inviting a user to test and optimizing details; and constructing a full-link privacy protection and safety guarantee mechanism.
  2. 2. The knowledge-based maternal and infant room voice recognition auxiliary processing method according to claim 1, wherein the constructing and optimizing the light-weight maternal and infant care knowledge graph comprises: Defining knowledge dimensions of a knowledge graph, the knowledge dimensions comprising four dimensions of basic care, emergency treatment, scene adaptation and equipment use, wherein the basic care covers feeding, changing urine, not wetting, sleeping, the emergency treatment covers choking, slight knocking, body temperature abnormality treatment, the scene adaptation covers highway milk powder brewing water temperature control scenerization content, and the equipment use covers a milk warmer, a wet towel heater operation; The method comprises the steps of obtaining knowledge data of a knowledge graph, wherein the knowledge data is obtained by adopting an authoritative data source and a scene adaptation mode, wherein core data is derived from pediatric society child care guidelines and trimethyl hospital neonatal care manuals, and supplementary data is derived from service area maternal and infant room equipment specifications and experience summaries of senior child nurses; Designing a framework of a knowledge graph, wherein the knowledge graph adopts a three-layer framework of an entity, a relationship and an attribute, the entity comprises a baby month age, a care scene, an operation step and an emergency symptom, the relationship defines an applicable scene, an operation pre-condition and a risk prompt association, and the attribute marks operation difficulty, required duration and notice; And carrying out scene optimization and incremental updating on the knowledge graph, aiming at a expressway scene, adding a vehicle-mounted milk powder storage and brewing connection knowledge point for the knowledge graph, and establishing an incremental updating mechanism based on user consultation data for updating the knowledge graph every month.
  3. 3. The knowledge-based maternal and infant room voice recognition auxiliary processing method according to claim 1, wherein the developing voice interaction and visual assistance in stages constructs a voice question-answer, authorization verification, visual guidance, feedback optimization flow, comprising: The method comprises the steps that a local awakening and cloud semantic analysis mixed mode is adopted, a high-frequency awakening word is locally preset, voice data are transmitted to the cloud through an edge computing gateway after awakening, and text and voice guidance is returned by combining a knowledge graph; Designing a voice verification and touch screen confirmation dual authorization mechanism, starting a camera after authorization, training a care behavior recognition model based on a computer vision technology, recognizing operation actions in real time, early warning and guiding nonstandard behaviors, and monitoring abnormal choking signs; recording care process data to form a log, providing a problem feedback inlet, and combining feedback optimization knowledge graph and an identification model.
  4. 4. The auxiliary processing method for mother and infant room voice recognition based on the knowledge graph according to claim 3, wherein the method for adopting a mixed mode of local wake-up and cloud semantic parsing, locally presetting a high-frequency wake-up word, transmitting voice data to the cloud through an edge computing gateway after wake-up, and returning text and voice guidance in combination with the knowledge graph comprises the following steps: Setting a system core performance index to meet the interaction response requirement of a mother and infant care scene, extracting a high-frequency wake-up word under the scene, and adopting a low-power consumption voice wake-up module supporting offline wake-up, an edge computing gateway with data processing and edge caching capabilities, a light semantic understanding model subjected to knowledge fine adjustment in the mother and infant care field and a mother and infant care knowledge map database supporting associated inquiry; Building a development environment based on a selected voice awakening module, importing the high-frequency awakening words, building a lightweight awakening model through a voice feature extraction and model training method, deploying the model to the awakening module, configuring an adjustable awakening threshold value, building an audio acquisition link with a noise reduction function, and enabling the awakening module to match audio features with the model in real time, triggering an awakening signal when the matching reaches the standard and transmitting the awakening signal to an edge computing gateway; Deploying a system and developing a data preprocessing module at an edge computing gateway, processing voice data acquired after awakening through a noise reduction algorithm, converting the voice data into a standard audio format, constructing an encrypted communication link, encrypting and packaging the preprocessed data, configuring a high-frequency knowledge node local caching strategy, and completing data processing and pushing to cloud service after the gateway receives an awakening signal; Deploying a semantic analysis engine at a cloud end, constructing a semantic understanding module based on the lightweight semantic understanding model, receiving data transmitted by a gateway, decrypting and converting the data into a text, extracting entities and intentions in the text through semantic analysis, generating a knowledge graph query instruction, responding to the query by a knowledge graph database, returning a result, packaging the result into a natural language text and synthesizing voice by the engine, generating text and voice double-format guiding data, and transmitting the text and voice double-format guiding data back to the gateway; The edge computing gateway pushes the guiding data to the mother and infant care terminal for synchronous output, a context cache module is developed to assist subsequent associated query analysis, and interactive adaptation of the main stream dialect is realized through adaptation of a wake-up module and fine adjustment of a semantic model; And simulating noise, dialect and network fluctuation environment in a mother and infant care scene to develop joint debugging test, optimizing audio acquisition link parameters, gateway cache range and model performance aiming at wake-up, delay and recognition problems in the test, and feeding back and optimizing terminal interaction details in combination with field scene test.
  5. 5. The knowledge-graph-based maternal and infant room voice recognition auxiliary processing method according to claim 4, wherein the building a development environment based on a selected voice wake-up module, importing the high-frequency wake-up word, building a lightweight wake-up model through a voice feature extraction and model training method, deploying the model to the wake-up module, configuring an adjustable wake-up threshold, building an audio acquisition link with a noise reduction function, wherein the wake-up module matches audio features with the model in real time, triggers a wake-up signal when the matching reaches the standard, and transmits the wake-up signal to an edge computing gateway, and the method comprises the following steps: Selecting hardware interface specifications and software adaptation requirements of a voice awakening module, constructing a development environment consisting of a module development tool chain, an adapted embedded system development environment and a voice signal processing dependency library, ensuring that the development environment is compatible with a hardware architecture and an instruction set of the voice awakening module, synchronously combing core interaction requirements of a mother and infant care scene, refining high-frequency awakening words conforming to use habits of carers, and importing the high-frequency awakening words into the voice awakening module after finishing according to a awakening model training tool format matched with the voice awakening module; Based on the upper limit of the computing resource of the voice awakening module, a lightweight voice characteristic extraction algorithm is selected to extract characteristic parameters of an imported high-frequency awakening word sample, an initial awakening model is built by adopting the characteristic parameters by adopting a training method adapting to lightweight requirements, and the lightweight awakening model matched with the hardware capacity of the voice awakening module is obtained through model pruning and parameter quantization optimization; The lightweight wake-up model is deployed to the voice wake-up module in a deployment mode supported by the voice wake-up module, driving adaptation is completed, an initial wake-up threshold is preset in combination with noise characteristics of a care scene of a mother and an infant, and a wake-up threshold adjusting interface is configured for the voice wake-up module to balance false triggering and missed triggering problems; Selecting an active noise reduction audio acquisition component, selecting proper components to complete physical connection according to the specification of an audio input interface of the voice awakening module, configuring noise reduction parameters of the audio acquisition component to adapt to the voice awakening module, and transmitting acquired audio to the voice awakening module after noise reduction; the voice wake-up module receives noise reduction audio transmitted by the audio acquisition component, invokes the lightweight wake-up model to extract features and matches the features of wake-up words to obtain a matching degree value, generates a wake-up signal when the matching degree value exceeds the wake-up threshold, and transmits the wake-up signal to the edge computing gateway through a communication interface matched with the edge computing gateway by the voice wake-up module to trigger a subsequent data processing flow.
  6. 6. The knowledge-graph-based maternal and infant room voice recognition auxiliary processing method according to claim 4, wherein the deploying the system at the edge computing gateway and developing a data preprocessing module processes voice data collected after awakening through a noise reduction algorithm and converts the voice data into a standard audio format, constructing an encrypted communication link, encrypting and packaging the preprocessed data, configuring a high-frequency knowledge node local caching strategy, and completing data processing and pushing to cloud service after the gateway receives an awakening signal, wherein the method comprises the following steps: The method comprises the steps of adapting the edge computing gateway hardware architecture to install an embedded operating system, configuring kernel parameters, installing basic dependency packages such as audio processing, encryption and communication, constructing a modularized development framework, dividing preprocessing, communication, caching and scheduling modules, setting resource thresholds and completing basic operation environment deployment; Selecting a noise reduction algorithm for the noise characteristics of the care scene of the mother and infant to integrate into the preprocessing module, developing a format conversion sub-module based on an audio processing tool chain, converting nonstandard audio into a preset standard format and checking, binding the preprocessing module with a wake-up signal triggering logic, starting processing after only receiving the wake-up signal, and designing an oversized audio slicing rule; Selecting a communication protocol to combine a symmetric encryption algorithm to construct the encrypted communication link, developing the encryption packaging submodule, defining a data packet structure, encrypting to generate a data packet with a check code, configuring a gateway and cloud bidirectional authentication channel, developing a link monitoring submodule, temporarily storing the data packet when abnormal, and retransmitting the data packet after recovery; Developing the cache management module, selecting a local cache carrier, defining a storage structure, configuring a timing and triggering updating rule and a capacity threshold value eliminating rule, developing a cache calling logic, and preferentially inquiring the local cache after receiving a wake-up signal; the wake-up signal receiving and analyzing module is developed, an analyzing signal is received through an adaptive interface, a process is triggered to collect voice data, the preprocessing module is invoked to process the voice data, an encrypted data packet is generated through the sealing sub-module, the pushing scheduling module is developed, the pushing scheduling module is pushed to the cloud end in normal time, temporary storage is carried out in abnormal time, a log is recorded after pushing, a confirmation signal is monitored, and retransmission is carried out in overtime.
  7. 7. The knowledge-graph-based maternal and infant room voice recognition auxiliary processing method of claim 1, wherein the building of an occupancy state monitoring and service area linkage mechanism, the building of a real-time monitoring, multi-terminal feedback and intelligent scheduling system, comprises: Collecting on-site data of the mother and infant room through a human body infrared sensor and a door magnetic sensor, transmitting the data to an edge computing gateway, and determining three states of the mother and infant room, namely the idle state, the occupied state and the to-be-cleaned state, after the edge computing gateway analyzes the data; constructing a multi-terminal data linkage network, uploading the judged state data to a service area management platform by the edge computing gateway, synchronizing the data to a service area guide screen, a mother and infant room gate indicator lamp and a navigation APP by the management platform, and adding a mother and infant room reservation queuing function in the linkage network; And constructing an intelligent scheduling module, wherein the intelligent scheduling module schedules cleaning staff to carry out cleaning work based on historical occupation data of the maternal and infant rooms, and when all the maternal and infant rooms in the service area are in an occupied state, the intelligent scheduling module pushes predicted idle time of the maternal and infant rooms to waiting users and recommends real-time states of the maternal and infant rooms in adjacent service areas so as to realize transregional resource scheduling.
  8. 8. The knowledge-graph-based maternal and infant room voice recognition auxiliary processing method of claim 7, wherein the intelligent scheduling module schedules cleaning staff to perform cleaning work based on maternal and infant room historical occupation data, when all maternal and infant rooms in a service area are in an occupied state, the intelligent scheduling module pushes predicted idle time of the maternal and infant rooms to waiting users and recommends real-time states of maternal and infant rooms in adjacent service areas to realize transregional resource scheduling, and the method comprises the following steps: The method comprises the steps of carding core data dimensions required by intelligent scheduling, including mother and infant room historical occupation data, cleaning personnel basic data, adjacent service area associated data and user waiting data, constructing a layered data storage architecture, storing real-time data and local data in a service area edge computing gateway local database, storing historical statistical data and cross-area data in a cloud service area management platform database, and establishing a local and cloud data synchronization mechanism to ensure timeliness of the data; Constructing a time sequence prediction model based on historical occupation data and cleaning time consumption data of a mother and infant room aiming at a cleaning scheduling scene, setting a cleaning scheduling trigger threshold, constructing a linear regression prediction model aiming at an idle time prediction scene by fusing real-time occupation data, historical duration data and current waiting data, constructing a multi-dimensional weight scoring model aiming at an adjacent service area recommendation scene, and setting a scoring rule; The splitting module is three sub-modules, including a cleaning scheduling sub-module, a prediction and pushing sub-module, a cross-region resource recommending sub-module and a weight scoring module, wherein the cleaning scheduling sub-module is used for reading the cleaning state of the mother and infant room and the data of cleaning personnel, matching target cleaning personnel and issuing scheduling instructions, the idle time predicting and pushing sub-module is used for calling a prediction model to calculate predicted idle time when the total occupied state of the mother and infant room is monitored, the waiting queue is combined to generate pushing information, and the cross-region resource recommending sub-module is used for calling a weight scoring model to screen adjacent service regions and integrating recommendation information after the local mother and infant room is fully occupied and the waiting time is predicted to be overtime; Integrating the three sub-modules into an intelligent scheduling module main frame, and operating a triggering rule, a message pushing rule and a cross-region data calling rule by a configuration module; Developing an interactive interface between the intelligent scheduling module and an external system, wherein the interactive interface comprises an uplink interface between the intelligent scheduling module and a service area management platform, a downlink pushing interface between the intelligent scheduling module and a user terminal, a cross-region data interface between the intelligent scheduling module and an adjacent service area and an instruction interactive interface between the intelligent scheduling module and a cleaning staff terminal, so that the real-time performance and the safety of data interaction are ensured; After the module is online, the algorithm model and the module rule are optimized based on actual operation data iteration, an exception handling mechanism is developed, and a spam strategy is configured for data loss, cross-region data interface interruption and cleaning personnel on-duty scene.
  9. 9. The knowledge-graph-based auxiliary processing method for mother and infant room voice recognition according to claim 1, wherein the building the simulation environment to test the stability for 72 hours invites the user to test and optimize the details comprises: Building a test environment simulating a mother and infant care scene, developing a continuous system stability test for 72 hours in the environment, inviting a user to participate in the simulated care test, collecting feedback information of the user on system guidance content and operation convenience, and optimizing system interaction details based on the feedback information; Selecting expressway service areas of different scale types as test points, deploying a system in the test point area, performing field test for 1 month, and optimizing a system local cache mechanism to ensure normal operation of basic functions according to the actual problem of network fluctuation in the test process; establishing a monthly knowledge graph updating mechanism, updating the system knowledge graph by combining user use data and field expert suggestions, establishing a quarterly hardware and function optimizing mechanism, and periodically optimizing and upgrading the system hardware equipment and the function module.
  10. 10. A mother and infant room voice recognition auxiliary processing system based on a knowledge graph is characterized by comprising: A processor; a machine-readable storage medium storing machine-executable instructions for the processor; wherein the processor is configured to perform the knowledge-graph-based maternal-infant room speech recognition assisted processing method of any one of claims 1 to 9 via execution of the machine-executable instructions.

Description

Auxiliary processing method and system for mother and infant room voice recognition based on knowledge graph Technical Field The invention relates to the technical field of voice recognition auxiliary processing, in particular to a mother and infant room voice recognition auxiliary processing method and system based on a knowledge graph. Background The traditional mother and infant room only provides infrastructure support, lacks the instruction of taking care of the refinement of different ages of infants, especially in mobile scenes such as highway service areas, the parents often face the sudden problems such as on-vehicle milk powder brewing, emergent choking treatment, strange milk warming device/wet towel heater operation, and the like, and the existing support channel depends on paper handbooks or on-line scattered inquiry, has lagged response and insufficient information pertinence, and is difficult to rapidly solve the instant care requirement. The existing voice recognition auxiliary technology has obvious short plates in adaptability of the maternal and infant care field, low wake-up success rate in a noise environment, insufficient recognition precision of a main flow dialect, lack of scene optimization of a knowledge system, and incapability of accurately matching mobile care characteristics of a high-speed service area, and meanwhile, the occupancy state of a maternal and infant room is lack of real-time monitoring and multi-terminal linkage feedback, so that the coordination efficiency of sanitation scheduling and transregional resources is low, and the safety guarantee mechanism of privacy data acquisition and storage is imperfect, so that the use experience of users is influenced, and the large-scale application of the technology in maternal and infant service scenes is restricted. Disclosure of Invention In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for assisting in processing speech recognition of a mother and infant room based on a knowledge graph, the method comprising: Constructing and optimizing a light-weight care knowledge graph of the mother and infant; developing voice interaction and visual assistance in stages, and constructing a voice question-answering, authorization verification, visual guidance and feedback optimization flow; constructing an occupancy state monitoring and service area linkage mechanism, and constructing a real-time monitoring, multi-terminal feedback and intelligent scheduling system; building a simulation environment to test the stability for 72 hours, inviting a user to test and optimizing details; and constructing a full-link privacy protection and safety guarantee mechanism. In still another aspect, an embodiment of the present invention further provides a system for assisting in voice recognition of a mother and infant room based on a knowledge graph, including: The system comprises a processor, a machine-readable storage medium, and a machine-executable instruction of the processor, wherein the processor is configured to execute the knowledge-graph-based mother-infant room voice recognition auxiliary processing method by executing the machine-executable instruction. In yet another aspect, an embodiment of the present invention further provides a computer program product, where the computer program product includes machine executable instructions, where the machine executable instructions are stored in a computer readable storage medium, and a processor of a computer device reads the machine executable instructions from the computer readable storage medium, and the processor executes the machine executable instructions, so that the computer device performs the above-mentioned knowledge-graph-based maternal-infant room voice recognition auxiliary processing method. Based on the above aspects, through constructing a light-weight maternal and infant care knowledge graph integrating authoritative data sources and scene adaptation, and combining a mixed interaction mode of local awakening and cloud semantic analysis, the efficient awakening, the accurate recognition of the mainstream dialect and millisecond-level guiding response under a noise environment are realized, the computer vision behavior recognition and abnormal early warning function is matched, the complete-flow accurate guiding covering basic operation, emergency treatment and equipment use is provided for a maternal and infant care person, and pain points such as lag in acquisition of care knowledge, non-standard operation and the like under a mobile scene are effectively solved. Meanwhile, the state of the mother and infant room is monitored through multi-sensor linkage and is synchronized to a multi-terminal, reservation queuing and cross-region resource recommending functions are added, and the intelligent scheduling module is combined to optimize clean distribution and idle tim