Intelligent old-age service platform based on internet of things big data

Abstract

The application discloses a kind of wisdom old-age service platform based on big data of thing association, belong to the technical field of Internet of Things old-age, specifically include: voice instruction acquisition module, for real-time acquisition old man and the interactive voice data of smart home equipment;Voice feature extraction module is used to carry out feature separation processing to the original voice data collected, extracts quantized instruction index;Baseline model construction module is used to establish the voice health baseline model based on user characteristics, generates personalized voice index reference interval according to old man historical voice data;Voice instruction comparison module is used to dynamically compare real-time instruction index with baseline model, generate voice health deviation curve, when curve fluctuation exceeds preset morphological characteristics, trigger early warning;Interaction mode adjustment module is used to adjust the interaction mode of smart home equipment according to the voice health deviation;The application realizes the health of old man by voice index change judgment and improves the quality of smart home service.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) elderly care technology, specifically to a smart elderly care service platform based on IoT big data. Background Technology

[0002] With the accelerating aging of the global population, the number of elderly people is increasing daily, making elderly care a focal point of social concern. Traditional elderly care models face numerous challenges, including limited human and material resources, making it difficult to meet the diverse needs of the elderly. At the same time, the rapid development of emerging technologies such as the Internet of Things (IoT), big data, and artificial intelligence has brought new opportunities to elderly care services. Smart home devices are gradually entering the lives of the elderly, providing services such as environmental control, health monitoring, and safety alerts through interaction, greatly improving the convenience and safety of their lives. Integrating various elderly care service resources through IoT big data to build smart elderly care service platforms has become an important development direction in the field of elderly care.

[0003] In smart elderly care service scenarios, voice interaction, as a natural and convenient human-computer interaction method, is widely used in the interaction between smart home devices and the elderly. Currently, there are already some smart home control systems based on voice commands that can recognize the elderly's voice commands and execute corresponding operations.

[0004] However, existing voice interaction technologies still have many shortcomings in elderly care service applications. They merely meet the basic needs of voice recognition and fail to fully explore the information about the elderly's health status contained in the voice data. Furthermore, they cannot adjust the interaction method in a timely manner according to dynamic changes in the elderly's voice health status, which may cause inconvenience for the elderly when using smart home devices, affecting the quality and effectiveness of smart elderly care services. Summary of the Invention

[0005] The purpose of this invention is to provide a smart elderly care service platform based on IoT big data, and to solve the following technical problems:

[0006] Current voice interaction technologies merely meet the basic requirements of voice recognition, failing to fully extract information about the elderly's health status from the voice data. Furthermore, they cannot adjust the interaction method in a timely manner based on dynamic changes in the elderly's voice health status.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A smart elderly care service platform based on IoT big data includes:

[0009] The voice command acquisition module is used to collect real-time voice data of the interaction between the elderly and smart home devices;

[0010] The speech feature extraction module is used to perform feature separation processing on the collected raw speech data and extract quantified instruction indicators, including spoken clarity, language fluency, grammatical structure completeness and instruction logic rationality.

[0011] The baseline model building module is used to establish a voice health baseline model based on user characteristics and generate personalized voice indicator reference ranges based on the elderly’s historical voice data.

[0012] The voice command comparison module is used to dynamically compare real-time command indicators with the baseline model and generate a voice health deviation curve. When the curve fluctuates beyond the preset morphological characteristics, an early warning is triggered.

[0013] The interaction mode adjustment module is used to adjust the interaction mode of smart home devices according to the voice health deviation.

[0014] As a further aspect of the present invention: the voice command acquisition module is based on a multimodal sensor array, specifically including:

[0015] The multimodal sensor array is divided into three operating modes based on the acquisition frequency, from high to low: high-frequency mode in the morning and evening, normal mode during the day, and low-frequency mode at night. The multimodal sensor array includes a directional microphone array and an environmental noise suppressor.

[0016] The pickup angle of the directional microphone array is dynamically adjusted according to the elderly's daily activity trajectory. It is set to wide-angle coverage mode in the designated public activity area and switches to directional focusing mode in the designated rest area.

[0017] When the system detects that an elderly person has interrupted the voice command interaction process several times in a row, it automatically switches to emergency monitoring mode. At this time, the microphone array expands its pickup range to the entire living space and starts real-time location tracking based on sound wave reflection. The environmental noise suppressor activates the corresponding noise feature library to filter noise according to different home scenarios.

[0018] As a further aspect of the present invention: the specific process of feature separation processing in the speech feature extraction module is as follows:

[0019] For the spoken language clarity, a standard acoustic fingerprint database containing common syllables is pre-established. After the collected speech is segmented into independent syllables, waveform similarity matching is performed with the standard database to calculate the pronunciation offset score.

[0020] For the language fluency, the ratio of effective vocabulary to the number of pauses within a set time period is statistically analyzed, and a personal speech rate baseline is established by combining historical data to obtain the deviation of real-time speech rate from the baseline.

[0021] Regarding the grammatical structure integrity, the collected statements are checked for subject-verb-object structure integrity. When missing components are detected, the semantics of the associated context are used for secondary verification to obtain a semantic coherence score.

[0022] To assess the logical rationality of the instructions, a mapping rule base between environmental parameters and instruction types is established, and a fuzzy inference mechanism is used to calculate the matching score between the voice instructions and the current environmental state.

[0023] As a further aspect of the present invention: the process of establishing the voice health baseline model in the baseline model construction module includes:

[0024] Collect historical voice data of the elderly, divide the training set and validation set according to the time dimension, and use the sliding window method to calculate the standard deviation and mean of each indicator. The duration of the sliding window is dynamically adjusted according to the data collection density.

[0025] A personalized voice indicator reference range is constructed, and the regional phoneme combination characteristics in the user's historical pronunciation are analyzed. A dialect pronunciation error tolerance compensation value is added to the speech clarity assessment. The error tolerance compensation value is set according to the dialect type classification.

[0026] The reference range for personalized voice metrics is updated, and historical data is processed using a weighted moving average algorithm. The weight of the influence of voice data on the range boundary is adjusted exponentially.

[0027] When performing model validation periodically, the generalization ability of the baseline model is detected by cross-validation. When the model deviation is detected to exceed the threshold, the historical data backfeed training process is triggered.

[0028] As a further aspect of the present invention: the process by which the voice command comparison module generates the voice health deviation curve is as follows:

[0029] The four indicators of spoken clarity, language fluency, grammatical structure completeness, and logical rationality of instructions acquired in real time are dynamically compared with the corresponding personalized reference intervals in the baseline model.

[0030] For each indicator, the deviation between the real-time measurement value and the baseline reference value is calculated. The deviation of spoken clarity is calculated by the difference in syllable pronunciation accuracy, the deviation of language fluency is calculated by the fluctuation of speech rate, the deviation of grammatical structure integrity is calculated by the number of missing sentence components, and the deviation of instruction logic rationality is calculated by the frequency of abnormal instructions.

[0031] A weighted fusion algorithm is used to combine the deviation values ​​of the four indicators into a comprehensive speech health deviation, in which the weight coefficients of grammatical structure integrity and instruction logic rationality gradually increase with the usage time;

[0032] The value of the voice health deviation is a continuous value between 0 and 1. The larger the value, the higher the degree of deviation in health status. The calculation process adopts a sliding time window mechanism to ensure the temporal continuity of the output results. After each new voice command is obtained, the voice health deviation value is updated immediately and transmitted to the intelligent interaction adjustment module to adjust the interaction mode. At the same time, it is stored in the historical database for baseline model updates.

[0033] As a further aspect of the present invention: the working method of the interaction mode adjustment module includes:

[0034] The speech deviation is divided into three levels: Level 1, Level 2, and Level 3. The interaction modes include standard interaction mode, Level 1 interaction mode, Level 2 interaction mode, and Level 3 interaction mode.

[0035] When the voice health deviation is at the first level, the first interaction mode is adopted, the response waiting time of the smart device is extended, the voice broadcast speed is reduced to the set ratio of the baseline value, and a confirmation prompt is added after each instruction.

[0036] When the deviation level is upgraded to the second level, the second interaction mode is adopted, simplifying the device response statement to a subject-verb-object structure, automatically breaking down complex instructions into multiple single-step operations, and inserting repeated keywords in the voice prompts;

[0037] When the deviation reaches the third level, the third interaction mode is adopted, which simultaneously enhances the brightness of the light prompt, displays a magnified version of the text instruction on the display screen, and activates the haptic feedback device to generate a vibration reminder.

[0038] If no corresponding level of voice deviation is detected within a continuously set time period, the interaction mode will be adjusted to the next higher level.

[0039] As a further aspect of the present invention, the third interaction mode specifically includes:

[0040] During voice broadcasting, the smart lights are controlled simultaneously, and differentiated light signal codes are generated according to the type of instruction: a green breathing light is displayed when the device executes successfully, a yellow flashing light is switched when confirmation is required, and a solid red light is switched when execution fails.

[0041] When adjusting the display screen layout, enlarge the keyword of the currently pending instruction to the top of the screen, and number and highlight the subsequent steps according to the execution order;

[0042] The tactile feedback device generates different vibration modes according to the priority of the command. It generates long vibration reminders for operations involving water and electricity safety, and short vibration reminders for ordinary operations.

[0043] The adjustment parameters for all interaction modalities are related to the voice health deviation; the higher the deviation value, the greater the redundancy of multimodal prompts.

[0044] As a further aspect of the present invention: during the interaction mode adjustment period, the accuracy rate of the elderly's operation is continuously monitored, and an inverse correlation model between the number of correct operations and the deviation value is established; when the number of consecutive correct operations reaches a set threshold, the interaction mode is adjusted to the next level.

[0045] The beneficial effects of this invention are:

[0046] The voice command acquisition module of this invention utilizes a multimodal sensor array, employing a time-segmented working mode, dynamically adjusting the pickup angle and range, and filtering noise to ensure comprehensive and accurate acquisition of elderly people's voice data. The voice feature extraction module performs multi-dimensional feature separation processing on the voice data, obtaining quantified command indicators that provide detailed evidence for assessing the elderly's voice health. The baseline model construction module establishes a voice health baseline model based on user characteristics, combining the sliding window method and weighted moving average algorithm, considering regional dialect factors, making the personalized voice indicator reference range more accurate, flexible, and dynamically updated. The voice command comparison module dynamically compares real-time command indicators with the baseline model, generating a voice health deviation curve and issuing warnings, enabling timely detection of abnormal voice health in the elderly. The interaction mode adjustment module classifies and adjusts the interaction mode according to the voice health deviation, employing measures such as extending response time, simplifying sentences, and enhancing light prompts at different levels. Simultaneously, it establishes an inverse correlation model between operation accuracy and deviation, allowing for flexible adjustment of the interaction mode, improving the convenience and service quality of the elderly using smart home devices, and effectively achieving early warning of health problems such as Alzheimer's disease by judging the trend of voice indicator changes. Attached Figure Description

[0047] The invention will now be further described with reference to the accompanying drawings.

[0048] Figure 1 This is a schematic diagram of the modules of the present invention. Detailed Implementation

[0049] The technical solutions of 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.

[0050] Please see Figure 1 As shown, this invention is a smart elderly care service platform based on IoT big data, comprising:

[0051] Voice command acquisition module

[0052] This module serves as a data entry point, collecting real-time voice data from interactions between the elderly and smart home devices. Its design is based on IoT sensor technology, supporting full-time coverage and scenario-based adaptation: it continuously listens for voice commands during the elderly's daily activities to ensure no interaction data is missed; and it utilizes an indoor microphone array to achieve directional reception of voice signals, reducing environmental noise interference. The data collection process adheres to privacy protection principles, acquiring only valid voice segments from device interactions to avoid irrelevant data redundancy.

[0053] Speech feature extraction module

[0054] Multi-dimensional feature separation was performed on the raw speech data to extract four key instruction indicators:

[0055] Spoken speech clarity: By comparing acoustic models, the accuracy of pronunciation is quantified, and the degree to which syllables deviate from standard speech is identified;

[0056] Language fluency: Statistically measure the effective vocabulary and pause frequency per unit of time to assess sentence coherence;

[0057] Syntactic structural integrity: parse the subject, verb, and object components of a statement and detect whether there is any missing core semantic meaning;

[0058] Command logic rationality: Combine the current environmental state (such as indoor temperature, equipment operating status) to determine the degree of matching between the command and the scenario.

[0059] Each indicator outputs a standardized value, forming a structured speech feature vector, which provides a data foundation for subsequent analysis.

[0060] Baseline model building module

[0061] A personalized speech health baseline model is constructed based on the elderly's historical speech data: First, the training data is divided into time series segments, and the mean and fluctuation range of each indicator are calculated to generate an individual-specific speech indicator reference interval. The model has dynamic update capabilities, adjusting the reference interval boundaries according to the latest interaction data to adapt to the natural changes in the elderly's long-term speech habits. In particular, the model incorporates user characteristics (such as age and dialect habits) to avoid the adaptation bias of general models and ensure that the baseline data closely matches the individual's actual situation.

[0062] Voice command comparison module

[0063] The system dynamically compares real-time extracted instruction indicators with a baseline model, generating a speech health deviation curve through numerical fitting. The curve reflects the magnitude and trend of the difference between current speech characteristics and the historical baseline. When fluctuations exceed a preset threshold (e.g., a 20% drop in clarity indicators for three consecutive times), the system automatically triggers a health warning, indicating potential cognitive function abnormalities. The warning mechanism supports multi-level responses, sending different levels of alerts to guardians and community service centers based on the degree of deviation.

[0064] Interaction mode adjustment module

[0065] The system dynamically optimizes the interaction logic of smart home devices based on voice health deviations: when a slight deviation is detected, the device response time is extended and the number of repetitions of voice feedback is increased to help the elderly more clearly confirm the execution status of commands; if the deviation continues to increase, the system automatically simplifies the interaction process, breaking down complex commands into single-step operations (such as "turn on the living room light and adjust the brightness" into two steps: "turn on the living room light" and "adjust the light brightness"). At the same time, it strengthens information transmission through multimodal methods such as text display and light flashing. The interaction mode adjustment follows the principle of "minimal interference," ensuring the efficiency of elderly operation while continuously collecting voice data to verify the adjustment effect, forming a closed-loop optimization of "collection-analysis-adaptation."

[0066] In a preferred embodiment of the present invention, the voice command acquisition module is based on a multimodal sensor array and specifically includes:

[0067] The voice command acquisition module uses a multimodal sensor array to acquire data, and ensures the efficiency and quality of voice data acquisition by dividing the data into modes, dynamically adjusting the pickup strategy, and suppressing noise.

[0068] The multimodal sensor array operates in three modes based on time: a high-frequency mode in the morning and evening, where the sampling frequency is increased to 5 times per second to meet the voice interaction needs of elderly people who are more active in the morning and before going to bed; a regular daytime mode, which maintains a sampling frequency of 2 times per second to balance data integrity and system resource consumption; and a low-frequency nighttime mode, which reduces the frequency to 0.5 times per second to maintain basic monitoring while reducing interference. The array includes a directional microphone array responsible for capturing voice signals and an environmental noise suppressor for filtering noise.

[0069] The directional microphone array features dynamic pickup angle adjustment. By analyzing the elderly person's historical activity trajectory, the system automatically expands the pickup angle to a 120° wide-angle coverage mode in public activity areas such as the kitchen and living room; in resting areas such as the bedroom, it switches to a 30° directional focusing mode to accurately capture the elderly person's voice. When the system detects three consecutive interruptions in the elderly person's voice commands (such as stopping before the sentence is fully uttered), the array immediately switches to emergency monitoring mode, expanding the pickup range to the entire house, and simultaneously initiating real-time location tracking based on the principle of sound wave reflection to pinpoint the elderly person's location. The environmental noise suppressor pre-stores noise feature libraries for different scenarios such as the living room, kitchen, and bedroom. Based on the scene labels identified by the sensors, it automatically calls the corresponding library files and eliminates environmental noise through frequency domain analysis and filtering algorithms.

[0070] In another preferred embodiment of the present invention, the specific process of feature separation processing in the speech feature extraction module is as follows:

[0071] To address spoken language clarity, the system pre-constructs a standard acoustic fingerprint database containing over 400 common syllables in Mandarin Chinese. The collected speech is divided into independent syllable units at 0.1-second intervals. The waveform similarity between each syllable and the standard database is calculated using the DTW (Dynamic Time Warping) algorithm. If the similarity is below a threshold, it is determined to be a pronunciation deviation, and a pronunciation deviation score of 0-10 is assigned based on the degree of deviation.

[0072] During the language fluency assessment, the system counts the number of effective words within a 10-second time period and simultaneously records the number of pauses (an interval exceeding 0.5 seconds is considered a pause), calculating the ratio of vocabulary size to the number of pauses. Combined with the elderly person's historical data from the past week, a personal speech rate baseline is calculated. The real-time ratio is compared with the baseline, and deviations are converted into deviation values ​​proportionally.

[0073] In terms of grammatical integrity verification, the system performs part-of-speech tagging and component analysis on the sentences, checking whether the core subject-verb-object structure is complete. If a missing component is detected, the system automatically associates the semantics of the speech within 10 seconds before and after the sentence, performs secondary verification through semantic dependency analysis, and gives a semantic coherence score of 0-100 based on the degree of semantic coherence.

[0074] The rationality assessment of command logic relies on a rule library that maps environmental parameters to command types, with over 200 pre-set environment-command matching rules. The system acquires over 10 environmental parameters in real time, such as indoor temperature and humidity, and equipment status, and uses a fuzzy inference mechanism to calculate the matching degree between voice commands and the current environmental status, outputting a matching degree score from 0 to 100.

[0075] In another preferred embodiment of the present invention, the process of establishing a voice health baseline model in the baseline model construction module includes:

[0076] First, the system collects nearly three months of historical speech data from elderly users, dividing it into training and validation sets in a 7:3 ratio. A sliding window method is used to process the data, with the window duration dynamically adjusted based on data collection density: a 1-hour window for high-frequency collection periods (e.g., morning and evening) and a 3-hour window for regular periods. Within the window, the mean and standard deviation of four indicators—speech clarity, fluency, etc.—are calculated to characterize the central tendency and dispersion of speech features within that time period.

[0077] When constructing personalized speech metric reference intervals, the system analyzes regional phoneme combinations in users' historical pronunciation. For example, for users in the Wu dialect region, a 10% error tolerance compensation value is added to the speech clarity assessment; for users in the Cantonese region, it is increased by 15%. The compensation value is pre-classified and set according to dialect type and similarity. Through this mechanism, the interference of dialect pronunciation differences on the baseline model is eliminated.

[0078] The reference interval is updated using a weighted moving average algorithm, assigning higher weights to recent data. When new data is added, the weights are adjusted exponentially, for example, data from 1 day prior to the current time has a weight of 0.8, data from 2-3 days prior has a weight of 0.5, and data from 4-7 days prior has a weight of 0.2. Model validation is performed weekly, using five-fold cross-validation to evaluate the model's generalization ability. When the validation error exceeds 15%, historical data is fed back in to retrain the model and optimize the parameters.

[0079] In another preferred embodiment of the present invention, the process by which the voice command comparison module generates the voice health deviation curve is as follows:

[0080] The system dynamically compares four indicators—speaking clarity, fluency, etc.—collected each time with the personalized reference interval in the baseline model. Specifically, the deviation in speaking clarity is represented by the difference between the real-time syllable pronunciation accuracy and the baseline accuracy; the deviation in fluency is measured by the fluctuation range between the real-time speech rate and the baseline speech rate; the deviation in grammatical structure integrity is calculated based on the number of missing subject-verb-object components in the sentence; and the deviation in the logical rationality of instructions is determined by the frequency of abnormal instructions (such as issuing a "raise the temperature" instruction when the ambient temperature is 25℃).

[0081] A weighted fusion algorithm was used to integrate the deviation of four indicators. The initial weights were set as follows: spoken clarity 0.2, language fluency 0.2, grammatical structure integrity 0.3, and instruction logic rationality 0.3. As the system usage time increased, the weights of grammatical structure integrity and instruction logic rationality were each increased by 0.05 every 30 days to highlight the importance of these two indicators in the health assessment. After fusion, a comprehensive speech health deviation score in the range of 0-1 was generated, with the value closer to 1 indicating a higher degree of deviation.

[0082] The calculation process uses a 10-minute sliding time window, and the deviation value is updated immediately after each new instruction is acquired. The results are synchronously transmitted to the interactive mode adjustment module to drive the adjustment of the device's interactive strategy; at the same time, they are stored in the historical database to provide data support for the periodic updates of the baseline model, forming a closed loop of "data acquisition - model comparison - strategy adjustment - model optimization".

[0083] In another preferred embodiment of the present invention, the operation method of the interaction mode adjustment module includes:

[0084] The system divides the speech deviation quantification range [0,1] into three levels: the first level (0-0.3) represents mild deviation, the second level (0.3-0.6) represents moderate deviation, and the third level (0.6-1) represents severe deviation. Four corresponding interaction modes are set: the standard mode is used for daily interaction; the first to third interaction modes are matched to different deviation levels, improving the elderly's operating efficiency by adjusting the response speed and information presentation method.

[0085] When the voice health deviation is detected to be at the first level, the system activates the first interaction mode: the device response waiting time is extended from the default 1 second to 3 seconds, the voice broadcast speed is reduced to 70% of the baseline value, and a voice confirmation prompt of "confirm operation?" is added after each command is executed.

[0086] If the deviation level is upgraded to the second level, switch to the second interaction mode: the device response statement is simplified to a subject-verb-object structure (e.g., "The living room ceiling light has been turned on for you and the brightness has been adjusted to 50%" is simplified to "Turn on the ceiling light, brightness 50%"). Complex instructions are automatically broken down into single-step tasks (e.g., "Turn on the air conditioner and set it to 26 degrees" is broken down into two steps: "Turn on the air conditioner" and "Set the temperature to 26 degrees"). At the same time, keywords are repeated in the voice prompts (e.g., "Turn on the ceiling light, turn on the ceiling light").

[0087] When the deviation reaches the third level, the third interaction mode is activated, integrating visual, auditory, and tactile multimodal prompts: the brightness of the smart lamp is increased to 100%, the display screen enlarges the instruction keywords to twice the original size and displays them on top, and the tactile feedback device outputs different vibration waveforms according to the instruction priority (such as a long vibration lasting 2 seconds triggered by water and electricity safety instructions, and a short vibration of 0.5 seconds for ordinary instructions).

[0088] If no corresponding level of voice deviation is detected within a continuously set time period, the interaction mode will be adjusted to the next higher level.

[0089] In a preferred embodiment, the third interaction mode specifically includes:

[0090] The third interaction mode uses multimodal signal encoding to provide intuitive feedback on command status: the smart lights present different light signals based on the command execution result; successful execution is indicated by a 0.5Hz green breathing flash, confirmation is required by a 1Hz rapid yellow flash, and failure is indicated by a solid red light; the display screen uses a "keyword at the top + step number" layout, enlarging "close" and "gas valve" from the "close kitchen gas valve" command to the top of the screen, with subsequent operation steps highlighted in red numbers in sequence; the haptic feedback device has a built-in vibration waveform library, outputting a 120Hz low-frequency long vibration for safety-related commands (such as turning off the power), and a 200Hz high-frequency short vibration for ordinary operations (such as playing music). All interaction parameters are dynamically correlated with the deviation; for every 0.1 increase in deviation, the light brightness increases by 10%, the voice repetition count increases by 1, and the vibration intensity increases by 20%.

[0091] In another preferred embodiment, the system monitors the elderly person's operational behavior in real time during the interaction mode adjustment period and counts the number of consecutive correct operations. When there are 10 consecutive correct operations under the first level of deviation, 8 consecutive correct operations under the second level, and 5 consecutive correct operations under the third level, an interaction mode downgrade is triggered: such as switching from the third interaction mode to the second interaction mode, gradually reducing the redundancy of multimodal prompts to avoid excessive assistance leading to elderly dependence. At the same time, the operation data is synchronously transmitted back to the baseline model construction module for optimizing the deviation threshold and interaction strategy parameters.

[0092] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1.A smart elderly care service platform based on Internet of Things big data, characterized in that, The method comprises the following steps: a voice instruction acquisition module for real-time acquisition of the interaction voice data between the old people and the smart home devices; a voice feature extraction module for feature separation processing of the collected raw voice data, extracting quantized instruction indicators, including oral clarity, language fluency, grammatical structure integrity and instruction logic rationality; a baseline model construction module for establishing a voice health baseline model based on user characteristics, generating personalized voice indicator reference intervals from the old people's historical voice data; a voice instruction comparison module for dynamically comparing real-time instruction indicators with the baseline model to generate a voice health deviation curve, and triggering an early warning when the curve fluctuation exceeds the preset morphological characteristics; an interaction mode adjustment module for adjusting the interaction mode of the smart home devices according to the voice health deviation; The process of generating a voice health deviation curve by the voice instruction comparison module is as follows: Dynamically compare the real-time acquired oral clarity, language fluency, grammatical structure integrity and instruction logic rationality with the corresponding personalized reference intervals in the baseline model; Calculate the deviation degree of each indicator between the real-time measurement value and the baseline reference value, wherein the oral clarity deviation degree is calculated by the syllable pronunciation accuracy difference, the language fluency deviation degree is calculated by the speech speed fluctuation amplitude, the grammatical structure integrity deviation degree is calculated by the number of missing sentence components, and the instruction logic rationality deviation degree is calculated by the frequency of abnormal instructions; Merge the deviation values of the four indicators into a comprehensive voice health deviation using a weighted fusion algorithm, wherein the weight coefficients of grammatical structure integrity and instruction logic rationality gradually increase with use time; The voice health deviation value ranges from 0 to 1, and the larger the value, the higher the health state deviation degree. The calculation process uses a sliding time window mechanism to ensure the time sequence continuity of the output results; After each new voice instruction is acquired, the voice health deviation value is immediately updated and transmitted to the intelligent interaction adjustment module for adjusting the interaction mode, and stored in the historical database for baseline model updating. 2.The Internet of Things big data-based wisdom pension service platform according to claim 1, characterized in that, The voice instruction acquisition module is based on a multi-modal sensor array, which specifically includes: The multi-modal sensor array working mode is divided into early and late high-frequency mode, daytime routine mode and night low-frequency mode according to the acquisition frequency from high to low; the multi-modal sensor array includes a directional microphone array and an environmental noise suppressor; The pickup angle of the directional microphone array is dynamically adjusted according to the old people's daily activity trajectory, set to wide-angle coverage mode in the set public activity area, and switched to directional focusing mode in the set rest area; When it is detected that the old people interrupt the voice instruction interaction process for several times in a row, automatically switch to the emergency monitoring mode, at this time the pickup range of the microphone array is expanded to the whole living space, and the real-time location tracking based on sound wave reflection is started; the environmental noise suppressor activates the corresponding noise feature library to filter noise according to different home scenes. 3.The Internet of Things big data-based wisdom pension service platform according to claim 1, characterized in that, In the voice feature extraction module, the specific process of feature separation processing is as follows: For the oral clarity, a standard acoustic fingerprint library containing common syllables is pre-established, and the collected speech is segmented into independent syllables for waveform similarity matching with the standard library to calculate the pronunciation deviation score; For the language fluency, the ratio of effective vocabulary to pause frequency within a set time period is counted, and a personal speech speed baseline is established based on historical data to obtain the deviation value of real-time speech speed relative to the baseline; For the grammatical structure integrity, the subject-predicate-object structure integrity of the collected sentence is checked, and when a component is missing, the context semantics are associated for secondary verification to obtain the semantic coherence score; For the instruction logic rationality, a mapping rule library of environmental parameters and instruction types is established, and a fuzzy reasoning mechanism is used to calculate the matching degree score of the voice instruction and the current environmental state. 4.The Internet of Things big data-based wisdom pension service platform according to claim 1, wherein, In the baseline model construction module, the establishment process of the voice health baseline model includes: Collecting historical voice data of the elderly, dividing the training set and the validation set according to the time dimension, and using the sliding window method to calculate the standard deviation and mean of each index, the length of the sliding window is dynamically adjusted according to the data collection density; Constructing a personalized voice index reference interval, analyzing the regional phoneme combination characteristics in the user's historical pronunciation, and increasing a dialect pronunciation fault tolerance compensation value in the voice clarity evaluation, the fault tolerance compensation value is set according to the dialect type classification; Updating the personalized voice index reference interval, using the weighted moving average algorithm to process historical data, and adjusting the influence weight of the voice data on the interval boundary according to the exponential law; When periodically performing model verification, the generalization ability of the baseline model is detected by cross-validation method, and when the model deviation exceeds the threshold value, the historical data backfill training process is triggered. 5.The Internet of Things (IoT) big data-based smart aged care service platform of claim 1, wherein, The working method of the interaction mode adjustment module includes: Dividing the voice deviation degree into a first level, a second level and a third level, and the interaction mode includes a standard interaction mode, a first interaction mode, a second interaction mode and a third interaction mode; When the voice health deviation degree is in the first level, the first interaction mode is used, the intelligent device response waiting time is prolonged, the voice broadcast speed is reduced to a set proportion of the baseline value, and a confirmation prompt is added after each instruction; When the deviation degree upgrades to the second level, the second interaction mode is used, the device response sentence is simplified to a subject-predicate-object structure, complex instructions are automatically split into multiple single-step operations, and key words are inserted in the voice prompt; When the deviation degree reaches the third level, the third interaction mode is used, the light prompt brightness is simultaneously enhanced, the enlarged version of the text instruction is displayed on the display screen, and the tactile feedback device is activated to generate vibration reminders; When no voice deviation degree of the corresponding level is detected within a continuous set time period, the interaction mode is adjusted to the previous level. 6.The Internet of Things (IoT) big data-based smart aged care service platform of claim 5, wherein, The third interaction mode specifically includes: Synchronously controlling the intelligent lamp during the voice broadcast process, generating differential light signal codes according to the instruction type: displaying a green breathing lamp when the device executes successfully, switching to a yellow flashing lamp when confirmation is needed, and turning to a red long bright lamp when execution fails; When adjusting the display screen interface layout, the keywords of the current to-be-executed instruction are enlarged to the top of the interface, and the subsequent steps are numbered and highlighted according to the execution order; The haptic feedback device generates different vibration patterns according to the instruction priority, and generates long vibration reminders for operations related to water and electricity safety, and short vibration reminders for ordinary operations; The adjustment parameters of all interaction modes are associated with the speech health deviation degree, and the higher the deviation degree value, the greater the redundancy of multi-modal prompts. 7.The Internet of Things (IoT) big data-based smart aged care service platform of claim 5, wherein, During the adjustment of the interaction mode, the correct operation rate of the old people is continuously monitored, and a reverse correlation model of the correct operation times and the deviation degree value is established; when the continuous correct operation reaches a set number threshold, the interaction mode is adjusted to the previous level.

Images