Sound stimulation adaptive regulation and control system based on abnormal state recognition

Abstract

The invention discloses a sound stimulation adaptive regulation and control system based on abnormal state recognition, and the system is characterized in that a signal collection and calculation module calculates an anxiety index through obtaining various physiological and motion sensing data of a user in real time; the panic recognition alarm module firstly performs motion state correction according to the motion sensing data, then runs a fast threshold path and an unsupervised model path in parallel, and outputs a panic attack alarm signal and an abnormal intensity score; the sound stimulation control module selects a sound stimulation mode according to the output of the panic recognition alarm module and the anxiety index, and outputs corresponding sound stimulation according to the individualized file of the user; the parameter optimization learning module evaluates the effect of sound stimulation in real time in each stimulation intervention process of the sound stimulation control module, and optimizes stimulation parameters of the next step; and meanwhile, archiving and learning the whole-course data after the intervention is finished. According to the invention, early accurate identification and timely intervention of the panic attack state can be realized.

Description

Technical Field

[0001] This invention belongs to the field of mental health monitoring and intervention technology, and in particular relates to an adaptive control system for sound stimulation based on abnormal state recognition. Background Technology

[0002] Panic attacks are a common manifestation of acute autonomic dysfunction, often occurring suddenly and reaching their peak within 10 minutes. Typical symptoms include strong palpitations, shortness of breath, chest tightness, dizziness, and other physical reactions, as well as intense fear (such as a feeling of impending death or loss of control). Currently, cognitive behavioral therapy (CBT) and drug therapy are the main clinical interventions for panic attacks, but these methods have significant limitations: (1) Drugs have a slow onset of action (usually several tens of minutes), making it difficult to control the intense symptoms in the early stages of an attack in a timely manner, and long-term use may lead to side effects such as addiction, drowsiness, and tolerance; (2) CBT requires patients to learn coping strategies when not having an attack, but patients often find it difficult to calmly implement the plan during an actual attack; (3) Existing protocols lack a real-time feedback and regulation mechanism for individual physiological states, and cannot dynamically adjust intervention strategies according to the real-time changes in physiological parameters of different patients. The lack of individualization leads to controversy regarding the effectiveness of the intervention.

[0003] Sound stimulation, as a non-pharmacological intervention, has been shown to have a certain regulatory effect on autonomic nervous system function and anxiety. For example, soothing music and natural environmental sounds (such as ocean waves and rain) can improve parasympathetic nerve activity, but they usually take 5–8 minutes to take effect, which is insufficient to meet the need for rapid relief in the early stages of a panic attack. In contrast, some spontaneous sensory meridian response (ASMR) sounds (such as whispers, brush strokes, and finger tapping) can induce relaxation responses in a short time: studies have shown that ASMR stimulation can enhance brain electrical activity. Wave, reduction ASMR sounds promote the activation of the parasympathetic nervous system and inhibit the sympathetic nervous system, thereby significantly relieving tension and anxiety within one minute. Therefore, using ASMR sounds for early intervention in panic attacks has the potential for rapid effectiveness.

[0004] However, current research also indicates that the long-term sustainability of the relaxation effect following such brief sensory stimulation remains uncertain, and the relaxation triggered by a single ASMR session is often transient. Without follow-up consolidation measures, the individual's autonomic nervous system may rebound to sympathetic dominance within minutes, thus reducing the overall intervention effect. Therefore, it is necessary to supplement ASMR with secondary interventions that can stabilize the rhythm and prolong the parasympathetic dominance state after rapid ASMR relief. Slow-paced sounds with a wave rhythm (4-8 Hz) are believed to help with emotional stability. The "slow beat" sound (e.g. low-frequency rhythm simulating heartbeat or soft drum beat) employed by the present invention aims to tune the user's respiratory and cardiovascular rhythms: the slow undulating sound at 4-8 times per minute can gradually guide the breathing and heart rate down to a resonant frequency of around 0.1 Hz, further enhancing parasympathetic activity and suppressing excessive sympathetic excitation, allowing the initial relief to last longer. In addition, there are studies that found low-frequency pulsed sound synchronized with the heartbeat rhythm can induce significant heart rate reduction and autonomic nervous balance adjustment within one cardiac cycle (<1 second) through cardiovascular reflex when suddenly stopped. This mechanism is faster than most central nervous regulation processes, equivalent to directly acting on the acute reflex pathway of the autonomic nervous system. If the heartbeat-synchronized pulse is introduced at the peak of a panic attack and interrupted in time, it is expected to instantly break the excessive activation of the sympathetic nervous system, creating favorable physiological conditions for subsequent rhythm stabilization interventions. In summary, the multi-stage sound intervention strategy of fast ASMR induction + slow beat consolidation + heartbeat-synchronized pulse is expected to achieve both fast-acting and sustained stability, with complementary advantages.

[0005] Although sound stimulation intervention shows the above potential, existing sound-based panic attack intervention methods still have deficiencies. First, most schemes use fixed sound types and stimulation parameters, lacking a mechanism for hierarchical dynamic adjustment according to real-time physiological feedback. For example, the patent document with publication number CN110743077A proposes an acoustic wave relaxation device that collects user physiological signals and sets individualized target parameters, then monitors physiological signal changes during relaxation training, and controls music switching between multiple tracks in real time to guide users to gradually relax. This method uses physiological feedback to adjust music output, avoiding reliance on human psychologists. However, such schemes are mainly used for psychological relaxation training in daily life, not for instant emergency intervention for panic attacks, and the sound stimulation used is single, lacking a hierarchical stimulation strategy for different stages of an attack. Second, some wearable devices on the market can collect multiple physiological signals, but the fusion accuracy and real-time application of multi-modal data are still insufficient, and their role in automatic intervention triggering decision-making is limited. Many products only monitor and alarm a single indicator, which can easily lead to delayed intervention timing or frequent false alarms. For example, existing systems can use wearable devices to analyze heart rate, skin conductance, breathing patterns, and other physiological data to predict panic attacks and issue warnings to users before an attack. However, such systems usually only provide warning prompts or guide users to relax on their own, and do not integrate automatic stimulation intervention functions. Third, the lack of effective recognition of user motion state makes it difficult for the system to distinguish between physiological changes caused by intense physical activity and autonomic nervous responses during a real panic attack, which can easily lead to false triggering and reduce user experience and trust. Single physiological signals often cannot reliably distinguish between these differences, and must be combined with multi-sensor information. Literature shows that the fusion of different modal signals can significantly improve the accuracy of emotion state detection: for example, the single-modal emotion recognition accuracy based on speech or electrocardiogram is only 70-80%, while the recognition rate can be improved to more than 90% after fusing multiple modalities such as speech + electrocardiogram. Therefore, systems lacking multi-parameter fusion and motion pseudo-signal correction often struggle to balance sensitivity and specificity, either missing reports or false alarms. The above deficiencies make existing sound intervention systems still unsatisfactory in quickly and accurately identifying panic attacks and implementing effective intervention.

[0006] To solve the above problems, the industry has begun to explore new technical solutions for closed-loop detection-intervention integration. For example, the patent document with publication number US11235156B2 proposes a wearable earphone device that continuously monitors user state with built-in physiological sensors, and automatically performs electrical stimulation intervention on the vagus nerve in the ear once an abnormal event such as anxiety or panic attack is detected. However, there is currently a lack of a multi-modal high-precision monitoring and individualized sound stimulation closed-loop regulation system designed for panic attacks. SUMMARY

[0007] The application provides a sound stimulation adaptive regulation system based on abnormal state recognition, which can evaluate the autonomic nervous state of a user in real time by using multi-source physiological signals, and dynamically output graded sound stimulation based on the evaluation results, so as to realize early and accurate identification and timely intervention of panic attack state.

[0008] A sound stimulation adaptive regulation system based on abnormal state recognition comprises a signal acquisition and calculation module, a panic recognition and alarm module, a sound stimulation control module and a parameter optimization and learning module. The signal acquisition and calculation module acquires the physiological data and motion sensing data of a user in real time through wearable sensors, and synchronously fuses and calculates the physiological data to obtain an anxiety index. The panic recognition and alarm module first corrects the motion state according to the motion sensing data, and then runs a rapid threshold path and an unsupervised model path in parallel to output a panic attack alarm signal and an abnormal intensity score ; wherein the panic attack alarm signal is a Boolean quantity, = 1 indicates that the user is currently in a panic attack, = 0 indicates no attack alarm or that the alarm has been lifted; the abnormal intensity score subdivides the risk level into low risk, high risk and high risk. The sound stimulation control module selects a sound stimulation mode according to the output of the panic recognition and alarm module and the anxiety index, and outputs corresponding sound stimulation according to the individualized profile of the user in the selected sound stimulation mode. The parameter optimization and learning module evaluates the effect of sound stimulation in real time during the stimulation intervention process of the sound stimulation control module, optimizes the stimulation parameters for the next step, and archives and learns the whole process data after the intervention ends.

[0009] The application uses wearable sensors as an entry point, fuses multi-source physiological and motion information, and forms an anxiety index and an abnormal intensity score that can directly reflect the autonomic nervous tension in real time; sound intervention is triggered under the recognition mechanism of “motion state correction + dual-path abnormality detection”; the intervention side adopts a three-level strategy of “rapid – rhythm – maintenance”, and realizes safe, comfortable and stable closed-loop regulation through online optimization and self-learning driven by individual acoustic profile; after the intervention ends, the system encrypts and archives the whole process data and incrementally updates the model to continuously improve the individualized effect.

[0010] Further, in the signal acquisition and calculation module, the physiological signals collected include electrocardiogram (ECG), captured heart rate (HR), heart rate variability (HRV), respiratory waveform and electrodermal activity (EDA). The motion sensing data collected include three-axis accelerometer data and gyroscope data.

[0011] Further, the signal acquisition and calculation module needs to calculate the signal quality index of each channel signal after collecting various physiological data and motion sensing data , the formula is as follows: ; In the formula, is the signal-to-noise ratio, is the data missing rate, is the motion tail degree; , , is the weight coefficient; The value range of , when , it is considered that the signal quality meets the standard, and the panic recognition alarm module works after the signal quality meets the standard; is the preset signal quality threshold.

[0012] Further, the specific process of calculating the anxiety index in the signal acquisition and calculation module is as follows: The physiological signals collected by different sensors are preprocessed to obtain a synchronous signal sequence, and each second in the synchronous signal sequence contains a set of synchronous physiological signal values for feature extraction; With 1 second as the update step, a sliding time window is selected to calculate various physiological features on the synchronous signal sequence; Different weights are calculated for each physiological feature , and the standardized deviations of all physiological features are summed according to the corresponding weights to obtain a comprehensive deviation score ; The comprehensive deviation score Score is mapped to an anxiety index in the 0-100 interval through an S-shaped Sigmoid function ; when mapping, the midpoint θ and the slope k of the Sigmoid curve are adjusted according to the individual situation of the user, so that 50 points correspond to the user's resting state; The instantaneous anxiety index calculated every second is smoothed by using an exponential moving average, and the final output time corresponds to the smoothed anxiety index ; ; In the formula, is the smoothing coefficient.

[0013] Further, the panic recognition alarm module according to the motion sensing data performs motion state correction, specifically including: Based on the motion sensing data, the average acceleration per unit time of the user is calculated and step frequency ; if Exceeding the preset threshold and lasting for more than 5 seconds, or step frequency If the user reaches the brisk walking / running level, the user's exercise status flag is set to TRUE; otherwise, the exercise status flag is set to FALSE. When the motion state flag is set to TRUE, the anomaly detection conditions for the fast thresholding path and the unsupervised model path are corrected; when the internal motion state flag is set to FALSE, no correction is made.

[0014] Furthermore, the working process of the fast thresholding pathway is as follows: Direct threshold judgment rules are set for physiological features extracted from physiological data, for each physiological feature. Set an abnormal threshold ; Whenever there is a deviation in physiological characteristics Exceeding the abnormal threshold If the direction of change is consistent with the pattern of a panic attack, then this physiological characteristic is marked as a suspicious abnormality at the current moment. ; Multiple characteristic deviations After normalizing each MDC, the weighted summation yields the joint effect size. If the combined effect size Exceeding the set threshold for combined effect size Then the suspicious anomaly will be considered. Perform a count to obtain an anomaly count. ; Based on joint effect size With joint effect size threshold The relationship between the numbers and the anomaly count The value, the output boolean exception flag. .

[0015] Furthermore, the working process of the unsupervised model pathway is as follows: Anomaly detection of physiological feature patterns is performed using pre-trained single-class support vector machines and isolated forest models, respectively; the normalized anomaly probabilities output are respectively... and .

[0016] Furthermore, the panic detection alarm module outputs a panic attack alarm signal. and abnormal intensity score The specific process is as follows: The comprehensive anomaly strength score is calculated based on the outputs of the fast thresholding pathway and the unsupervised model pathway. The formula is: ; wherein, , , is a weight coefficient, satisfying ; The trigger condition is set by m-of-k logic, and is set in a determination window of the last k seconds. If at least m time points satisfy , an alarm is triggered, and a panic attack alarm signal is output ; is a preset alarm threshold.

[0017] Further, the sound stimulation mode of the sound stimulation control module comprises a rapid intervention mode, a rhythm relaxation mode and a maintenance consolidation mode. The entering condition of the rapid intervention mode is that the user state is in a high-risk interval, and the panic recognition alarm module outputs a panic attack alarm signal = 1 and the comprehensive abnormal intensity score , or the anxiety index ; in this mode, rapid relaxation type sound is played; the duration is not more than 60 seconds. The entering condition of the rhythm relaxation mode is that the user state is in a medium-risk interval, and or ; in this mode, θ rhythm slow beat sound is played, and the duration is 3-5 minutes. The entering condition of the maintenance consolidation mode is that the user state is in a low-risk level, and <0.30 and ; in this mode, soothing background environmental sound or light music is played, and the duration is 3-5 minutes. In each sound stimulation mode, the instantaneous change of the user's physiological indicators is continuously monitored during the sound playing process, and the comprehensive abnormal intensity score and the anxiety index are obtained to evaluate the intervention effect; after each sound stimulation mode ends, the next mode is selected based on the intervention effect, or the sound stimulation is exited.

[0018] Further, the individualized profile of the user comprises individualized sound stimulation threshold, sound preference and safety limit.

[0019] Compared with the prior art, the present application has the following beneficial effects: The application realizes early, accurate and low false alarm detection of panic attack by a "double-channel" recognition mechanism of multi-modal physiological signal fusion and motion state correction; relies on a hierarchical sound stimulation strategy to realize second-level intervention by fast-acting acoustic stimulation at the early stage of attack, and then to realize steady-state transition by rhythmic sound and to maintain and consolidate by environmental sound, thereby significantly shortening the relief time and prolonging the relaxation effect; establishes an individualized acoustic archive and adjusts parameters according to real-time physiological feedback during intervention, so as to achieve accurate regulation and comfortable experience according to different people and different times; introduces self-learning and historical optimal scheme hot start, so that the system is continuously optimized and becomes more effective with use; sets smooth transition and safety boundary in the whole process, controls the intensity and switching without abruptness, and balances effectiveness and safety; provides explainable state quantities such as anxiety index and intervention log, which is convenient for effect tracking and long-term management in clinical or home scenarios; adopts wearable and mobile terminal integration, and is non-drug, non-invasive, low threshold and easy to deploy, and through local processing and encryption archiving, privacy and data security are considered; has robustness and redundancy strategy for missing sensors and complex environment, and ensures stable and reliable closed-loop regulation ability under daily use conditions. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 The working flow chart of the sound stimulation adaptive regulation system based on abnormal state recognition according to an embodiment of the present application. DETAILED DESCRIPTION

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

[0023] It should be noted that the features in the following embodiments and implementation manners can be combined with each other without conflict.

[0024] As Figure 1As shown, a sound stimulation adaptive regulation system based on abnormal state recognition, the main working process is: with wearable sensor as the entrance, first through the "signal acquisition-signal quality" link to complete data access and quality control: unqualified return to collection, qualified into "anxiety index calculation", continuous output state for downstream use. Subsequently, "motion state correction" uses motion information to remove body motion artifacts, and the corrected features are sent to "panic recognition (double channel)" to complete rapid rules and unsupervised model determination in parallel; if the result is "no", return to the anxiety index channel for continuous monitoring, if "yes", trigger "sound stimulation". Sound stimulation is executed according to the three-level strategy of "rapid intervention-rhythm mitigation-maintenance consolidation", and is supervised by "termination judgment": continue intervention if not stable, and end and transfer to "parameter optimization and self-learning" if recovery is up to standard. This unit archives the current data, updates the model and improves the individual acoustic file, and at the same time sends the real-time parameter adjustment instruction back to the "sound stimulation" for online fine tuning, forming a closed loop cooperation of "detection-intervention-evaluation-optimization", so that the subsequent trigger is faster, more accurate and more individualized.

[0025] The system of the present application mainly comprises a signal acquisition and calculation module, a panic recognition and alarm module, a sound stimulation control module and a parameter optimization and learning module. The four modules will be described in detail below.

[0026] Module 1: Signal acquisition and calculation module Through the wearable sensor network, the user's various physiological and behavioral data are acquired in real time, and the multi-source data are synchronously fused to calculate the anxiety index (AI). This module ensures that various signals are synchronously collected and have sufficient sampling accuracy to capture the subtle changes in autonomic nervous activity during panic attacks, and converts complex multi-channel physiological changes into intuitive and quantitative anxiety index for use by subsequent modules.

[0027] 1.1 Signal acquisition sub-module The signal acquisition sub-module synchronously controls various sensors through a unified time reference, continuously and in parallel acquires raw signals output by various physiological sensing devices worn by the user, including but not limited to: electrocardiogram (ECG), capturing heart rate (HR) and heart rate variability (HRV) key data, sampling rate set to be greater than or equal to 250 Hz; respiratory waveform monitoring respiratory frequency and depth changes, sampling rate greater than or equal to 25 Hz; skin conductance (EDA) reflecting the small changes in skin sweat gland activity, sampling rate greater than or equal to 4 Hz; skin temperature – indicating peripheral blood flow changes, optional parameters; electromyography (EMG) capturing muscle tension changes, optional parameters; blood oxygen saturation (SpO2) reflecting blood oxygen content, optional parameters; electroencephalogram (EEG) recording brain activity as an optional reference signal; motion sensing signals including three-axis accelerometer and gyroscope data for monitoring user activity status and posture changes; auxiliary information input (optional) such as facial expressions / posture changes captured by a camera, voice features collected by a microphone, and subjective reports obtained through a mobile phone App, etc.

[0028] The signal acquisition sub-module ensures that all sensor data is recorded synchronously according to a unified clock: to meet real-time evaluation requirements, the main physiological signals use a specified high sampling rate (as described above), and a unified timestamp or serial number is attached to each data sample to achieve accurate alignment across channels. The module also performs basic data quality monitoring during operation: it performs preliminary filtering on the channel signals (such as removing power frequency interference from electrocardiogram, identifying and marking motion artifacts), and removes obviously distorted data segments; and it detects sensor status (such as poor electrode contact, device disconnection) to indicate signal quality indicators. During the acquisition process, specific sensors can be dynamically activated / deactivated as needed according to the situation (for example, temporarily turning off the microphone in a noisy environment to save energy).

[0029] On this basis, the signal acquisition sub-module further calculates signal quality indicators for each channel signal to measure the reliability of the current physiological signal. The indicator is based on multi-factor weighted calculation, including signal-to-noise ratio , data loss rate , and motion artifact degree , and the calculation formula is as follows: ; The value range is [0, 1], and the larger the value, the more reliable the signal. When (as ), it is considered that the signal quality meets the standard, and the system will trigger the alarm logic.

[0030] The signal acquisition sub-module outputs time-synchronized and verified multi-modal raw data streams. Specifically, it includes physiological signal data, motion state data, and possible auxiliary information input arranged in time sequence (all with unified timestamps). These data will serve as the basis for subsequent fusion and identification module inputs. The main physiological signals are used to extract features and calculate the anxiety index; motion sensor data is directly provided to the panic recognition and alarm module to identify motion artifacts and dynamically adjust the threshold; additional visual, audio, and subjective feedback information can be used to corroborate the identification results or for manual verification. The signal acquisition sub-module runs throughout the entire system startup process, providing uninterrupted raw data support for continuous monitoring and closed-loop control.

[0031] 1.2 Anxiety Index Calculation Sub-Module The anxiety index calculation sub-module synchronously fuses multi-source physiological data from the signal acquisition sub-module and extracts features to calculate a comprehensive indicator of the user's current anxiety / tension level, the anxiety index (AI), to achieve real-time quantitative assessment of the user's autonomic nervous state. The anxiety index ranges from 0 to 100 points, with 50 points corresponding to the user's normal calm state, 0 indicating extreme relaxation, and 100 indicating extreme anxiety. This indicator will serve as an important basis for the identification and control modules.

[0032] The data sources for the anxiety index calculation sub-module are the synchronized multi-channel physiological signal sequences provided by the signal acquisition sub-module, as well as reference parameters such as baseline statistical values (resting mean, standard deviation) and channel weight initial values for the user in a resting state.

[0033] The processing process for calculating the anxiety index includes the following steps: Multi-channel alignment and preprocessing: Due to different sampling rates and possible data packet loss, asynchronous signals need to be aligned and interpolated. A multi-modal Kalman filter algorithm is used, with each channel signal as an observation, to estimate the "best" signal value on a unified time axis in real time, thereby outputting a multi-channel signal sample aligned every second. At the same time, obvious noise is filtered out (such as power frequency interference in ECG signals, motion artifacts), improving the reliability of the signals. After this step, a set of synchronized physiological signal values is obtained every second for feature extraction.

[0034] Sliding window feature extraction: With 1 second as the update step, a certain length of sliding time window (e.g. 30 seconds) is selected to calculate physiological features on the synchronous signal sequence. Typical features include: current heart rate HR (calculated from electrocardiogram R-R interval), respiratory rate RR, time-domain indicators of heart rate variability (such as RMSSD, the root mean square of successive differences between adjacent heartbeats), frequency-domain indicators of HRV (such as low / high frequency power ratio LF / HF), mean skin conductance level SCL, and skin conductance transient peak response count SCR, etc. In addition, shorter windows (e.g. 5 seconds or 10 seconds) can be used to calculate transient features such as heart rate transient rise rate, EDA surge, etc. The series of calculated feature values are then compared with the user's resting baseline, converted into standardized deviation values (Z-score): the difference between the current value of each feature and the resting mean divided by the resting standard deviation, representing the degree of deviation of the feature relative to the individual's calm state.

[0035] Feature threshold discrimination and weighted fusion: To improve robustness, a change threshold dead zone is set: when the deviation amplitude of a certain feature does not exceed its minimum detectable change (MDC), it is considered that the change may be submerged in noise and can be considered as "no contribution" to the overall anxiety state, and the feature deviation is recorded as 0; when the feature deviation exceeds MDC, it is considered that there is a significant change. Given that different physiological indicators have different sensitivity and reliability to anxiety state, the module assigns different weights to each feature . First, the instantaneous signal reliability is evaluated according to the observation error covariance of the channel obtained by Kalman filtering (the smaller the error, the higher the reliability), and then the feature importance coefficient obtained by prior setting or online learning , the unnormalized weight is calculated and normalized to obtain , ensuring that the weight distribution vector sums to 1. The calculation formula is as follows: ; Then, the standardized deviations of all features are summed according to the corresponding weights to obtain the comprehensive deviation score: ; This Score value represents the overall deviation of the current autonomic nervous activation level relative to the resting baseline. The more significant the deviation and the more features from high-reliability channels, the greater their contribution.

[0036] Mapping to anxiety index: The comprehensive deviation score Score is mapped to the anxiety index in the 0-100 interval by the S-shaped Sigmoid function . When mapping, the midpoint θ and slope k of the Sigmoid curve are adjusted according to the individual user situation, so that 50 points correspond to the user's resting state, and the sensitivity of the score to physiological changes conforms to the experience judgment. The mapping formula is as follows: ; where θ is the midpoint parameter (usually the mean or median of Score in resting state), k is the slope parameter (usually between 0.5-2.0, can be set according to individual resting fluctuation, to control the sensitivity of the exponential to physiological deviation).

[0037] Smoothed output anxiety index: Exponential moving average is used to smooth the instantaneous anxiety index calculated per second, to reduce the misjudgment caused by instantaneous fluctuations. Set the smoothing coefficient α ≈ 0.2, then the final output of the smoothed anxiety index is: ; This smoothing process is equivalent to allowing the past few seconds of history to still have some influence on the current index, so as to obtain a stable anxiety index .

[0038] Output result: Anxiety index AI (0-100). This index is updated at a frequency close to real time (delay in seconds) as an objective measure of the user's current autonomic nervous tension state. In addition to outputting the AI value, the module also retains information such as component features and Z scores, which can be referred to by other modules for reference.

[0039] The signal acquisition and calculation module provides two main outputs to the downstream: one is the anxiety index time series and key physiological features for the panic recognition and alarm module, which are used for abnormal state determination; the second is the real-time state feedback for the sound stimulus control module and parameter optimization learning module, which is used to decide the intervention trigger and evaluate the intervention effect (for example, the high and low of AI can determine whether emergency intervention is needed, and the change of AI is used to evaluate the effectiveness of sound stimulation, etc.).

[0040] 2. Panic recognition and alarm module The panic recognition and alarm module is the core decision-making unit of the system, responsible for detecting panic attacks from real-time physiological data and triggering alarm signals in a timely manner to start intervention. This module realizes high sensitivity and low false alarm through the combination of "double-channel anomaly detection + motion state correction" mechanism: first, it detects whether the user is in intense exercise and dynamically corrects the judgment threshold to exclude the interference of physiological changes caused by exercise; then it runs two sets of algorithms, rapid feature threshold filtering and unsupervised anomaly detection model, in parallel to identify autonomic nervous abnormal patterns. After the fusion of the results of the two channels, the final alarm signal is obtained through m-of-k rolling decision and hysteresis logic. This module is not only used for detection and triggering in the early stage of attack, but also continuously monitors the intensity of abnormality during intervention to determine when to end the intervention or whether a secondary attack has occurred.

[0041] The main inputs of the panic recognition alarm module come from the time series of multiple physiological features (HR, RMSSD, EDA, etc.), motion sensor data (accelerometer, gyroscope signals) and signal quality indicators collected by the signal acquisition and computation module (which indicates the reliability of the current data). The module also refers to the threshold parameters of each feature (such as feature MDC, resting mean) of the user at rest, etc.

[0042] 2.1 Motion state correction sub-module The motion state correction sub-module first calculates the user's motion intensity indicators, such as the average acceleration per unit time , step frequency , etc., to determine whether the user is currently in a state of intense exercise. For example: The calculation formula of the average acceleration is as follows, where is the average acceleration in the time window.

[0043] ; The calculation formula of the step frequency is as follows, where is the step frequency per minute converted from the number of steps monitored in the time window.

[0044] ; If exceeds the preset threshold and lasts for more than 5 seconds, or the step frequency reaches the level of fast walking / running, the internal motion state flag is set to TRUE, indicating that the current physiological signals may be affected by intense exercise.

[0045] In the case of the motion flag TRUE, the system considers that the increase in heart rate, respiration, etc. at this time may be due to exercise rather than a panic attack, so the panic attack judgment conditions are dynamically corrected: (1) increase the abnormality judgment threshold: for example, require the heart rate to rise above the calm baseline by more than "+5 bpm" from "+2 bpm" to be counted as abnormal; HRV needs to be more significant (such as RMSSD decreasing by more than 25% instead of 10%) to be counted as abnormal; (2) increase the requirement for multiple indicators to be abnormal simultaneously: in the case of multiple channel data available, the alarm must be triggered only when other physiological indicators such as EDA show a significant surge and respiratory variability show abnormal changes; if a certain auxiliary channel is missing, the threshold for the remaining channels is further increased to compensate for the lack of information and ensure that false alarms do not occur due to fluctuations in a single indicator during exercise.

[0046] ​​If the motion state flag is FALSE (user is at rest or with light activity), the system uses the individual resting state sensitivity threshold for decision making. The motion state detection itself does not directly output an alarm, but serves as a dynamic gating mechanism to adjust the sensitivity of the subsequent anomaly recognition in real time. To avoid false alarms caused by the high heart rate residual after the user has just finished exercising, the system sets a threshold recovery period: when the motion state is detected to switch from TRUE to FALSE, the decision threshold is gradually recovered from the increased value in the motion state to the resting level within about 1-2 minutes. For example, the heart rate anomaly threshold is gradually transitioned from the motion threshold back to the resting threshold over time t: ; where is the moment when the exercise ends, is the threshold recovery duration. This avoids the alarm decision jitter caused by the sharp drop in the threshold.

[0047] 2.2 Dual-path anomaly detection submodule After removing the interference of motion factors, the dual-path anomaly detection submodule performs true panic attack anomaly detection on multiple physiological features. Two parallel paths are used for simultaneous analysis, and each path independently gives an anomaly indication.

[0048] (1) Fast threshold filtering path: direct threshold decision rules are set for several key physiological features, characterized by simple calculation and extremely low delay (<1 second). Typically, the 4 most sensitive indicators for panic attacks are selected, such as: heart rate HR, HRV time domain indicators (such as RMSSD), skin conductance level SCL, and heart rate instantaneous change rate, which form . For each feature , an anomaly threshold is set, which is an empirical value ; set the expected direction symbol , for example, HR: , HRV: . Take the maximum value of the empirical threshold of the index and the minimum detectable change value MDC of the index to ensure that the threshold variation is statistically detectable: ; Whenever a feature deviates from the threshold and the change direction is consistent with the typical pattern of panic attacks (e.g. HR rises, RMSSD falls), the feature at the current time is marked as "suspected anomaly": ; where is an indicator function, which takes 1 if the condition is true, and 0 otherwise.

[0049] However, single feature abnormality flagging can be disturbed by noise, for robustness, introduce a weighted effect size similar to the aforementioned effective reaction judgment: normalize and weight the multiple feature deviations by their respective MDCs and accumulate them: where the weight is set according to the feature SNR, reliability (ICC) and historical stability. If the joint effect size exceeds the threshold (taking 1.0, corresponding to the overall deviation reaching the minimum detectable level; usually taking 1.0, but also can be individualized adjustment in the range of 0.8-1.5), it is considered that the current significant abnormality occurs, and the abnormality count is recorded .

[0050] Under the premise of signal quality up to standard, the final judgment of the fast channel adopts the "voting + fusion" parallel criterion: ; where, is the minimum number of abnormal features threshold (e.g. ), is the joint effect size threshold (e.g. , corresponding to the overall deviation reaching the MDC level).

[0051] (2) Unsupervised model channel: use pre-trained machine learning models to detect feature patterns for abnormality. The module runs two kinds of unsupervised anomaly detection models in parallel: One-Class SVM (single class support vector machine) and IsolationForest (Isolation Forest). These models have been trained on system initialization or user historical normal data to represent the "normal state" distribution.

[0052] One-Class SVM: using RBF kernel, mapping multi-dimensional physiological features to high-dimensional space, learning decision boundary around normal data, its discriminant function can output the distance of sample to boundary . The form is: where, is the support vector weight, is the kernel function output, representing the similarity of the input sample and the normal sample in the training set, is the boundary constant. When , it is judged as abnormal.

[0053] Isolation Forest: Randomly select feature subsets to construct a large number of decision trees, and measure the degree of anomaly by judging the difficulty of sample separation in the tree. The average path length of a given sample (the number of branches needed to find x on each tree in the forest) , according to the size of the training data N, calculate the normalized anomaly score : ; where is the normalization constant for the size of the data. The closer to 1 means the easier to isolate (the more abnormal), the closer to 0 means the more similar to the normal mode. The system sets the threshold to be the 99% quantile of the anomaly score distribution of the training set (which can be adjusted in the 95%-99% interval according to the actual application false positive / false negative cost), if the module output anomaly score is greater than the quantile, it is determined to be abnormal.

[0054] Unsupervised model path outputs two values at each time: SVM discriminant function (can be converted to probability or distance) and Isolation Forest anomaly score . For easy fusion, they are normalized to the 0-1 interval respectively.

[0055] 2.3 Result fusion and alarm decision sub-module The results of the fast path and the model path need to be evaluated comprehensively to give the final alarm decision. When fusing, the reliability of each is assigned a weight, and the signal quality index Q(t) is used to control the start and stop of the alarm. When (for example ), it is considered that the signal quality meets the standard, and the system will trigger the alarm logic.

[0056] Comprehensive anomaly intensity calculation: Let the Boolean anomaly label (or effect quantity probability) output by the fast path be , and the normalized anomaly probability output by the unsupervised model path be and . Then the comprehensive anomaly intensity score can be expressed as: ; where , , satisfy . The default can take the average allocation, or it can be dynamically adjusted according to the signal quality, for example, when the channel quality relied on by a certain path becomes poor, reduce its weight. The calculated , the larger the value represents the more significant the abnormal mode.

[0057] Trigger condition (m-of-k logic): To improve robustness, a sliding window count method is used to determine the alarm. Set in the determination window of the last k seconds, if there are at least m times that meet (the integrated abnormal intensity exceeds the alarm threshold, such as = 0.5), an alarm is triggered. Taking a typical value k = 5, m = 3 as an example, if there are 3 seconds of abnormalities in 5 seconds, it is determined that there is a continuous abnormality, and an alarm trigger signal is output. This m-of-k logic can filter out individual isolated peak interference. When , an alarm is triggered.

[0058] Alarm hysteresis retention: Once the alarm trigger is TRUE, the system starts the hysteresis mechanism: requires to drop below the threshold and remain at least seconds (such as 5 seconds) before the alarm is lifted. This mechanism can avoid frequent triggering / elimination caused by repeated jitter at the threshold edge.

[0059] Model adaptive update: To maintain detection accuracy, the module periodically updates the anomaly detection model using new data in the background. Specifically, it includes: append the normal physiological data accumulated during the intervention interval to the training set of One-Class SVM and Isolation Forest, retrain or fine-tune the model boundary, so that it reflects the user's latest normal physiological range (ensure that the model learns the normal envelope over time, reduce false positives). At the same time, for the feature threshold in the fast threshold path, ; where is the update coefficient (0.7-0.9), represents the average change amplitude of the new baseline measurement. This update ensures that the threshold adjustment is smooth and not affected by short-term fluctuations. When the system detects that some channels are missing for a long time (such as the user stopping wearing a certain sensor), the module will switch to the pre-trained dimensionality reduction model (reassign weights after removing missing channels) to maintain the continuity of the recognition function.

[0060] The output results of the result fusion and alarm decision submodule are as follows: Panic attack alarm signal : Boolean quantity, indicating whether a high-probability panic attack event is detected. = 1 indicates that the user is currently most likely in a panic attack, triggering the intervention process with high priority; = 0 indicates that there is no attack alarm or the alarm has been lifted. In implementation, Determined by the m-of-k decision logic described above, and considering signal quality control, only triggered when data is reliable.

[0061] Abnormality intensity score : Continuous numerical value (0-1 interval) quantifying the severity of the current panic state. This score is actually a combination of the abnormality intensity score and the model confidence score. Combined with the model confidence score, the result can be subdivided into risk levels: e.g. 0.0-0.3 low risk (only slight abnormality signs), 0.3-0.6 medium risk (moderate abnormality), 0.6-1.0 high risk (significant abnormality). This score will be used in the sound stimulus control module as a basis for deciding the intensity of the intervention (higher score outputs stronger stimulus), and also for evaluating the effectiveness of the intervention (the value should decrease after the attack is alleviated). In addition, the module can also output abnormality type information (which indicators triggered the alarm) in the background of the system for record analysis, and support providing auxiliary judgment basis (such as combining the typical symptom subjective report of DSM-5 as auxiliary verification).

[0062] Alternatives: The dual-channel mechanism adopted by this module has fully combined the advantages of real-time evaluation and unsupervised learning, but in specific scenarios or limited data situations, some simplified alternatives can also be used, for example: (1) Fixed threshold method: set an absolute threshold (e.g. heart rate > 120 bpm or RMSSD < 20 ms) to simply determine whether to alarm when the value exceeds; (2) Sliding window statistical method: calculate the short-term mean and standard deviation of the feature in real time, and judge as abnormal when the current value deviates from the mean by ±k times the standard deviation (e.g. 3 times); (3) Template matching method: use dynamic time warping (DTW) or cosine similarity to compare the current feature curve with the historical typical panic attack pattern, and determine abnormality if the similarity is high; (4) Subjective report trigger: combined with the clinical diagnostic criteria (DSM-5), if the user reports a sudden strong fear through the App and is accompanied by typical symptoms (palpitations, shortness of breath, etc.), it is directly used as one of the alarm trigger conditions; (5) Group model comparison: use a large number of normal population feature distributions to establish a principal component analysis (PCA) model to determine whether the current data is outside the normal range of the population. These schemes can be used as redundancy or verification means for this module, and can be selected according to application requirements. Overall, this recognition module realizes early and accurate recognition and reliable alarm of panic attack state through multi-index fusion and dynamic adjustment, providing a decision basis for the timely start of subsequent sound intervention.

[0063] 3. Sound stimulus control module The sound stimulation control module is the execution unit of the system's closed-loop regulation. After a panic attack is identified, it outputs appropriate sound stimulation to regulate the user's autonomic nervous system state based on the decision results of the identification module and the user's individual profile. This module implements a three-level intervention strategy, managing the switching between different intervention modes through a state machine, gradually guiding the user from the peak of panic to a stable state with progressively increasing sound types and intensities. The module also includes intervention termination judgment logic to determine when to safely end the sound stimulation based on the user's physiological recovery.

[0064] Inputs include alarm trigger and deactivation signals from the panic detection alarm module. = 1 or 0), user thresholds and preference parameters from individual acoustic profiles (threshold sound pressure level EMR, preference degree, initial recommendation strength, safety limit, etc. for each sound type), and real-time physiological feedback from the acquisition / fusion module (anxiety index AI, abnormal intensity score). (HR, HRV, EDA, etc.). In addition, the parameter optimization learning module provides real-time parameter adjustment instructions to guide this module in fine-tuning specific parameters of the sound output (such as sound intensity and rhythm) in a given mode.

[0065] The sound stimulation control module is divided into three continuous modes based on the developmental stages of panic attacks: rapid intervention, rhythmic calming, and maintenance consolidation. The mode switching logic is managed through a finite state machine. Each mode has its specific function: the rapid mode is used for emergency symptom suppression, the rhythmic mode is used for consolidation and stabilization, and the maintenance mode is used for recovery and transition. The entry conditions, stimulation strategies, durations, and switching conditions for each mode are explained below: 3.1 Rapid Intervention Submodule Entry condition: Alarm triggered by the identification module ( = 1), and the overall anomaly intensity score reaches high risk (e.g., (or anxiety index) This mode also automatically enters when there is no alarm but the user manually triggers emergency intervention. At this time, the user is in the early or peak stage of a panic attack and needs immediate and strong intervention.

[0066] Stimulation Action: The module immediately plays selected fast-relaxing sounds from the user's acoustic profile. Prioritizing ASMR sounds that have been verified to rapidly induce a parasympathetic response (such as whispers, rustling, tapping, etc.), the volume is set to the user's effective minimum threshold EMR for that sound type plus approximately 3 dB(A) (slightly higher than the threshold to ensure effectiveness). If this information is lacking in the user profile, the system's default low-frequency, gentle ASMR sound is used as the initial stimulus (frequency 50–500 Hz, sound pressure level approximately 45–55 dB(A)). During sound playback, the module continuously monitors real-time changes in the user's heart rate, skin conductance, and other physiological indicators, and obtains the data calculated by the recognition module. And AI values ​​to assess the effectiveness of the intervention.

[0067] Emergency Intervention Sub-process: If an extremely abnormal and rapidly deteriorating user condition is detected (e.g., HR spikes to ≥120 bpm or AI spikes to ≥90), the module will trigger an emergency intervention sub-process that overlays a heartbeat-synchronized pulse sound. Specifically, it acquires the user's current heart rate (determined by the ECG RR interval) and inserts a short, low-frequency pulse sound (frequency 0.5–4 Hz, pulse width 50–100 ms, duty cycle ≤30%) at the peak of each heartbeat, continuously outputting several heartbeat cycles (usually 3–5 consecutive pulses within a few seconds, with a total duration not exceeding 10 seconds). This rapid pulse, by simulating the heartbeat rhythm and intervening slightly ahead of time, can trigger a vagal reflex within approximately 1 second, causing a sudden drop in heart rate of 1–2 bpm, thereby quickly interrupting the overactivation of the sympathetic nervous system and preventing further symptom deterioration. The emergency sub-process and ASMR sound can be superimposed, but the duration is strictly limited to avoid excessive stimulation.

[0068] Duration: The rapid intervention mode typically lasts no more than 60 seconds. If symptoms do not improve during this period, it may be extended appropriately, but not exceeding 120 seconds (including emergency sub-processes that may be triggered multiple times). The design aims to pull the user back from a high-risk state as quickly as possible; if 1 minute is not effective, it may be extended to 2 minutes.

[0069] Mode switching: When the rapid intervention mode ends, the system determines the next mode path based on the degree of improvement in the user's condition. If, after approximately 60 seconds of rapid intervention, the user's condition has significantly recovered to a low-risk level (e.g., ...), If the risk level is <0.30 and AI(t)<50), and the low-risk state has been stable for at least 20 seconds by the m-of-k criterion, and the signal quality Q(t) meets the standard, then the system determines that the user's autonomic nervous system has been greatly calmed, skips the second stage and directly switches to the maintenance and consolidation mode (believing that there is no need to go through rhythm relaxation).

[0070] If the user's condition has improved but is still in the medium-risk range (e.g., 0.30≤ If the condition is <0.60 or 50≤AI(t)<70), and there has been some improvement in indicators relative to the peak of the attack (e.g., the heart rate has decreased by more than its MDC from the peak, but is still more than 5% higher than the baseline), and the above condition is maintained stably with good Q(t), then the system considers the rapid intervention to be initially effective but still needs to be consolidated, and switches to the rhythm relaxation mode to further stabilize the autonomic nervous function.

[0071] If the user remains in the high-risk zone ( If the abnormality intensity does not decrease significantly (i.e., AI(t) < 0.60 or AI(t) < 70) within the total duration, the system judges that the attack has not been effectively controlled and continues to maintain the fast intervention mode (extend the total duration to a maximum of 120 seconds) and triggers the emergency pulse sub-process again if necessary.

[0072] If the abnormality intensity does not decrease significantly (i.e., AI(t) < 0.60 or AI(t) < 70) within the total duration, the system judges that the attack has not been effectively controlled and continues to maintain the fast intervention mode (extend the total duration to a maximum of 120 seconds) and triggers the emergency pulse sub-process again if necessary.

[0073] 3.2 Rhythm relaxation sub-module Entry condition: When the fast intervention ends and the user's state enters the above-mentioned "transition / medium risk" zone, the system automatically switches to the rhythm relaxation mode. This mode aims to further reduce physiological arousal using specific rhythm sounds, guiding the user from a medium tension level to a completely calm state. If it is detected in the fast mode that it needs to be rolled back (as described below in the maintenance mode state deterioration), re-entering the rhythm relaxation also follows the same action.

[0074] Stimulus action: The module outputs a theta rhythmical tapping sound as the main sound stimulus in this mode. Implementation: A continuous background sound (e.g., soft and deep drum sound, "puff-puff" sound simulating heartbeat, etc.) is played, with a fundamental frequency in the range of 50-500 Hz, and a low-frequency amplitude modulation of 4-8 Hz is applied, making the sound exhibit a slow fluctuating rhythm (equivalent to 240-480 micro-oscillations per minute, corresponding to a rhythm of 4-8 breaths per minute). This design aims to guide the user to gradually slow down their breathing rhythm to 4-8 times per minute and match the cardiovascular resonance frequency (about 0.1 Hz, i.e., 6 times per minute), maximizing the promotion of parasympathetic nerve activity. In addition, the rhythm stimulus in the theta band (4-8 Hz) helps to stabilize the user's brain low-frequency brain activity (enhance alpha / theta waves and suppress high-frequency beta / gamma waves), further relieving anxiety. During the playback process, the system continuously monitors the changes in user HR, EDA, etc., and calculates and AI to assess whether the user is continuously improving.

[0075] Duration: The rhythm relaxation mode usually lasts about 3-5 minutes. This period is used to consolidate the effects of fast intervention and ensure that the user's physiological indicators are fully restored and stable. If the user has completely calmed down within 3 minutes, it can be ended early; if it is still not fully recovered after 5 minutes, it may need to be extended or other measures (see the switching conditions below).

[0076] Mode switching: Rhythm relaxation mode decides to continue, end early or fallback based on user state changes: If during rhythm relaxation, user metrics have entered low risk range (e.g. <0.30 and AI(t)<50) and sustained by m-of-k criterion for at least 20 seconds and Q(t) is met, the system considers the user autonomic state has recovered to a stable level and switches to maintenance consolidation mode to enter the end phase of intervention.

[0077] If during rhythm relaxation, user state deteriorates again to high risk range (e.g. ≥0.60 or AI(t)≥70), it indicates a new panic attack or previous attack not fully controlled. The system immediately falls back to fast intervention mode to suppress symptoms with stronger stimulation (regarded as reprocessing of secondary attack). To avoid mode oscillation, the system is designed to not fallback from rhythm relaxation to fast intervention more than twice in a single attack intervention cycle; if exceeded, manual intervention is prompted.

[0078] If rhythm relaxation mode continues to run for more than 5 minutes, and the user still has a considerable proportion of time (e.g. >80%) staying in the medium risk range (state fails to further improve), the system will also prompt manual intervention or other auxiliary means to avoid the user being "stuck" in this state for a long time. At the same time, the system can try to adjust the stimulation parameters (such as slightly increasing the sound intensity or changing the tone) to seek breakthrough according to the optimization module.

[0079] 3.3 Maintenance consolidation sub-module Entry condition: When rhythm relaxation mode ends and the user's physiological state has entered the low risk stable range, the system switches to maintenance consolidation mode. In addition, if the recovery criteria are met directly after fast mode (e.g. the aforementioned case of skipping rhythm relaxation), it will also directly enter the maintenance mode. This mode aims to consolidate the intervention effect, prevent physiological rebound in the short term, and let the user return to a relaxed state completely.

[0080] Stimulus action: After entering the maintenance mode, the module switches the output sound type. Instead of the previously used ASMR or tapping sound stimulus, it plays soothing background ambient sounds or light music. These sounds can be chosen according to the user's preferences in the acoustic profile, such as: natural sounds (rich in frequency components but overall soft, not triggering tension, typical volume 45-60 dB(A)) like ocean waves, rain, wind chimes, campfire crackling, etc.; or gentle music therapy tracks, guided meditation voice, etc. (slow and steady rhythm, giving a sense of safety and relaxation). The focus of the sound stimulus in the maintenance phase is to provide a calm sound environment that helps the user maintain the established parasympathetic dominance state and gradually reduce residual physiological stress response. Because the user's physiology has been greatly stabilized after the previous two phases, this phase does not require special sound effects, only to prevent external stimuli from being too strong or internal anxiety from rising.

[0081] Duration: The maintenance consolidation mode usually lasts 3-5 minutes. However, the specific time can be adjusted according to the user's actual recovery situation: if the user's indicators have been stable for a long time at 3 minutes, the intervention can be considered to end early; otherwise, if the user is still slightly tense, the intervention can be extended to a maximum of about 10 minutes. If the user feels completely calm, the intervention can also be terminated by the user through the terminal interface (the system allows the intervention to end early after confirming that the user's indicators are normal).

[0082] Mode switching / ending: The maintenance mode is the last phase of the intervention, mainly facing two situations: normal ending or abnormal rollback. If the maintenance mode lasts for a period of time (at least 3 minutes) and the user's physiological indicators remain stable in the low-risk zone, for example Always below 0.25, and AI(t) is always below 45, while heart rate, HRV, EDA, etc. are close to resting baseline levels, and meet the set stable time (such as 3 consecutive minutes meeting the above standards), the system determines that the intervention has been successful and the user has recovered calm. At this time, the intervention ending process is triggered: the module smoothly fades out the sound output (usually within about 2 seconds, the volume gradually decreases to zero), then stops all sound stimuli, and announces the completion of this round of intervention. If the maintenance mode has lasted for 5 minutes but the indicators are in the low-risk range but not completely meet the exit criteria (for example <0.30 and AI(t)<50 is not lower), the system can extend the mode to a maximum of 10 minutes; if it still does not meet the standards after 10 minutes, it will also take a fade-out ending (to avoid over-reliance on stimulation) and prompt the user to take a break or consider other help.

[0083] If the user's state repeatedly rises during the maintenance mode: the indicators re-increase to the transition zone (such as 0.30≤ <0.60 or 50≤AI(t)<70) or the high-risk zone ( ≥ 0.60 or AI(t) ≥ 70), it indicates that instability may occur again. The system will immediately fallback to the corresponding mode according to the degree of increase: moderate degree to rhythm mitigation mode with enhanced intervention, severe degree to fast mode for emergency treatment. Similar to the foregoing, to avoid infinite loop, no more than 2 times of fallback from maintenance mode within an intervention period; if repeated frequently, manual intervention or end of intervention is recommended.

[0084] Intervention termination decision logic: To ensure that the intervention is ended at the right time, the module has a built-in termination decision mechanism to assess in real time whether the user's physiological state has recovered to a stable state that can safely stop the sound stimulation. The termination decision logic starts from the beginning of the intervention, and checks the key indicators every fixed time: Decision criteria: A sliding window statistical method is used to continuously compare the user's current physiological indicators with the individual resting baseline. The specific method is to take a time window of 60 seconds, calculate the average value and standard deviation of the core indicators (heart rate HR, HRV indicator RMSSD, skin electricity EDA) in the window, and compare with the average value and fluctuation of the resting baseline. When the following conditions are met, it is considered that the physiological state has recovered to a stable state: (1) the sliding window average of heart rate and HRV indicators returns to the ±10% range of the resting baseline, and the EDA level is not significantly higher than the resting state; (2) the fluctuation (standard deviation) of these indicators in the window is not more than 1.2 times the fluctuation in the resting state; (3) the above state is continuously maintained for at least 3 window periods (about 3 minutes). Only when all conditions are met, the termination decision module outputs a "recovery stable" signal.

[0085] Normal end process: Once the recovery stable signal is detected, the module triggers the intervention end instruction to the sound control unit, executes the sound fade-out and stops outputting (as described above, fade-out for 2 seconds and stop). Since the end of the intervention usually means that the user's attack has subsided, the system can prompt the user through the interface that the intervention has ended at the same time of fade-out, and record the end timestamp.

[0086] User-initiated ending: When the user clicks “End / Pause” on the interface or initiates ending by long-pressing the hardware button, the system first performs a false-touch prevention confirmation. If the current physiological indicators are still in the abnormal range and have not yet reached the stable determination, a secondary confirmation and “Continue intervention / End immediately / Seek help” options are popped up. After the confirmation, the sound control unit stops with the same 2-second linear fade-out; the system records the end timestamp, end type “user-initiated”, reason code (such as “uncomfortable / time up / invalid / other”), immediate subjective score, and remarks, and marks the event as “user-initiated ending” for subsequent effect analysis and parameter weight adjustment. If the session length does not reach the minimum effective duration, the intervention effectiveness evaluation is processed as “partial data” or with reduced weight; the system gives brief ending guidance (such as maintaining slow breathing for 30-60s) in the ending interface, and sets a cooling time according to the safety strategy before it can be started again. If obvious abnormalities are still detected (such as HR / EDA exceeding the safety threshold) at the time of ending, the system will prompt the risk and suggest switching to the fast relief mode or contacting professional support.

[0087] Timeout forced ending: If the intervention continues to reach the preset maximum duration (e.g., 15 minutes) and still does not meet the stable standard, to prevent discomfort or fatigue reactions caused by long-time sound stimulation, the system will also be forced to enter the ending process: gradually reduce the sound intensity to a safe level (such as below 45 dB(A) for tens of seconds), while popping up a prompt to suggest that the user consider seeking professional medical help or implementing other relaxation methods, and then executing the fade-out stop sound. Forced ending still follows a smooth transition to avoid sudden discomfort to the user.

[0088] Termination determination logic and mode switching mechanism determine the end of the intervention: usually end when the maintenance mode reaches the condition, but if the user reaches stability in advance, it may also end in the rhythm relaxation mode (at this time, it will skip the maintenance mode and directly end the intervention). The decision signal of this module is also provided to the parameter optimization and data archiving module for marking the intervention effect and data processing at the end time. The termination module ensures that the intervention neither interrupts too early nor drags too long, making the whole closed loop more intelligent and reliable.

[0089] Output: The sound stimulation control module directly outputs digital audio signals to the user's headphones or speakers to achieve physical stimulation. This includes the specific sound content played at each stage, as well as its dynamically changing parameters over time (such as sound pressure level, frequency components, rhythm patterns, etc.). Simultaneously, the module outputs the current intervention mode status (fast / rhythmic / maintained) and related parameters (current sound type, sound pressure level, rhythm frequency, etc.) in real time for other modules to access. On one hand, the parameter optimization module adjusts its algorithm based on this information and the user's physiological feedback; on the other hand, the data archiving module records the sound control module's behavior log (including mode switching times, sound types used, parameter adjustment trajectories, etc.).

[0090] Module Interaction: The sound stimulus control module cooperates with other modules in the following ways: when the alarm module outputs... When the value is 1, this module immediately responds and enters the intervention process; after the alarm is cleared or a termination signal is issued, this module executes the termination action. When initiating intervention, the control module queries the user's acoustic profile to obtain individualized parameters (e.g., preferred sound type, threshold sound pressure level (EMR), safety constraints, etc.) to formulate the initial stimulation plan. During intervention, the module continuously receives physiological feedback (e.g., heart rate, EDA, and anxiety index AI) from the acquisition / fusion module. This feedback is either directly used for simple internal adjustments (e.g., emergency pulse triggering conditions) or processed by the optimization module to form adjustment instructions. The parameter optimization module and the control module have a closed-loop cooperative relationship: the optimization module calculates the optimal parameter combination for the next step based on feedback, but the control module checks and executes it according to the current intervention mode and safety constraints (the optimization module cannot directly change the mode, but only provides parameter fine-tuning suggestions within that mode). When executing external adjustment instructions, the control module ensures a smooth transition in output changes (e.g., sound fade-in / fade-out to avoid abrupt changes). Throughout the intervention process, the control module acts as both an "actuator" and a "monitor": it completes the sound output and simultaneously senses changes in the user's physiological response and makes mode-level decisions. After the intervention ends, the control module stops the sound and notifies other modules to close the current round of intervention. Through the above mechanism, the sound stimulation control module realizes the concrete implementation of the intervention strategy, acting as a bridge between the upper and lower layers of the closed-loop system: connecting the recognition decision trigger at the top and the physiological effects generated by the user at the bottom, forming a virtuous cycle with the optimization and self-learning modules.

[0091] 4. Parameter Optimization Learning Module The parameter optimization learning module is the system's adaptive adjustment and continuous improvement unit. On the one hand, it evaluates the effect of sound stimulation in real time during each intervention and optimizes the stimulation parameters for the next step, so that the intervention process is not limited to a fixed plan, but is adjusted according to the individual and the time. On the other hand, after the intervention, it archives and learns from the entire data, updates the individualized model, and improves the intervention effect in the future.

[0092] 4.1 Individual acoustic profile submodule Upon initial use of the system or when the user is in a calm state without seizure, the parameter optimization module first performs an individualized sound threshold determination procedure to establish a user-specific acoustic parameter profile. This profile can be regarded as the calibration data of the system, providing individualized sound stimulation thresholds, preferences, and safety limits, which serve as the basis for subsequent real-time optimization.

[0093] The input of the individual acoustic profile submodule comes from the multi-modal physiological signals of the user in resting state provided by the signal acquisition submodule, as well as the various test sound signal libraries built into the system. The sound library contains different types of sound materials, including multiple ASMR sounds (such as whisper, brush friction, paper rustling, etc.), natural environmental sounds, soft music segments, etc., all of which are standardized in terms of loudness and spectrum for comparability.

[0094] Resting baseline recording: Before the start of the measurement, the user is guided to relax in a quiet environment while wearing the system's designated audio output device (earphones). The system records the user's physiological data in a resting state for at least 3 minutes, calculates the mean and standard deviation of key indicators (such as heart rate HR, HRV's RMSSD, skin conductance level, etc.), and calculates the minimum detectable change (MDC) of each indicator through permutation test or resampling to evaluate the stability of the baseline. If the baseline fluctuates too much, the relaxation collection time is extended until the data is stable. The MDC is usually defined at a 95% confidence level, for example: ; Where SD is the resting standard deviation of the indicator, and ICC is the reliability coefficient calculated by repeated measurements.

[0095] Sound stimulation threshold determination: Select several typical sound source types from the sound library for testing (avoiding too much testing causing fatigue, usually prefer low-frequency, soft sounds that can easily trigger autonomic nervous responses, such as selecting 3 different types of ASMR sounds: whisper, brush sound, and paper sound each for one segment). For each test sound, use the ascending-descending ladder method to determine its effective stimulation threshold: initially play at a lower loudness (such as 35 dB(A)), if no obvious physiological changes are observed, gradually increase the volume by about +3 dB each step, until clear physiological response signs or subjective reports of significant stimulation are observed; then reduce the volume to find the critical sound pressure level between response and non-response. Each sound pressure level is usually played continuously for about 30 seconds (with 1 second fade-in and fade-out, 9 seconds of sustained sound in between, followed by about 20 seconds of silence to observe the aftereffect), and the user's multi-modal physiological indicator changes are recorded synchronously.

[0096] The basis for judging "effective physiological response" is the change of multiple indicators, which is judged by constructing the joint effect E~: including the decrease of heart rate relative to baseline (-ΔHR), the increase of HRV RMSSD (ΔRMSSD), the decrease of skin conductance (-ΔEDA), and the user's subjective relaxation / pleasure score (S). Each change is divided by its minimum detectable change value for normalization, and then weighted Weighted sum: + + + ; Weight The weights are used to balance the contributions of different channels, which can be set empirically (e.g. 25% each) or adjusted adaptively according to the signal-to-noise ratio and reliability of each indicator to ensure that the judgment takes into account individual differences and is robust. The judgment threshold is usually set to (i.e. the joint effect reaches the sum of the MDC levels of each indicator) to determine the occurrence of effective response. When , it is considered that the sound pressure level has caused a significant relaxation response. Then the sound pressure level is adjusted up and down around the threshold several times to find the minimum sound pressure level that just triggers the response (called EMR, Effective Minimum Response level) more precisely. For example, reduce the intensity by 1 level from the first response to confirm whether there is no response, if not, then take the lower level, and repeat several times until the intensity reverses more than 6 times to get a stable value. If the difference between the two measured threshold values exceeds 3 dB, additional tests are automatically added. Finally, the posterior median of multiple measurements is taken as the EMR of the sound type, and its 95% confidence interval is recorded for reference.

[0097] Preference evaluation: After determining the threshold EMR of each sound, to evaluate the user's subjective preference, the system plays each test sound at an intensity of "threshold + 3 dB(A)" for about 10 seconds, and asks the user to subjectively rate or rank the comfort or preference of each sound. The module collects these ratings and calculates the user's preference level for different sound types. This step helps the system choose the user's preferred sound for intervention among multiple options.

[0098] Acoustic profile generation: Based on the above test results, the module establishes an individual acoustic profile for the user. The profile contains: the effective minimum sound pressure level threshold EMR for each type of test sound; the user's preference degree (e.g. like, neutral, dislike) for the sound; the recommended initial intervention intensity (usually EMR + 3 dB as the initial sound pressure level); and the user's safety limit parameters (e.g. the upper limit of the single adjustment amplitude, the maximum sound pressure level of the sound, etc.). For example, the safety parameters can be set: the sound intensity does not change more than ±6 dB every 30 seconds, and the intensity of any sound type does not exceed EMR + 9 dB(A). The profile also records the sensitivity indicators of the user's physiological response during the test process (such as the maximum amplitude of HR decrease), which is used as a reference for the optimization module. The acoustic profile is stored in the system database after establishment and can be used for individualized configuration of subsequent interventions. During actual intervention, the control module will preferentially select the sound type with the highest preference degree and within the safety range in the profile as the initial stimulus, and adjust the intensity according to the threshold and safety parameters in the profile. The profile data also provides a customized search space for the parameter optimization module (for example, the optimization algorithm only adjusts in the sounds the user likes, and does not try to use the sounds the user explicitly dislikes). The profile can be updated gradually as the user uses it (for example, the user's long-term non-use or change in physical condition can be re-measured).

[0099] The individual acoustic profile determination sub-module is usually run once when the user first uses the system, or initiated by the user when the system is idle. The data obtained by the determination is mainly used by the sound stimulus control module and the parameter optimization module to achieve personalized intervention. In real-time monitoring and intervention, the profile determination module itself does not participate in the calculation, but its results are frequently consulted: for example, the control module decides the sound selection and initial volume according to the profile, and the optimization module uses the profile to limit the optimization range and adjust the weight. If the user's physiological condition or preference changes significantly, the system should re-run the determination to update the profile, so that the model parameters remain accurate.

[0100] 4.2 Real-time parameter optimization During each intervention execution, the parameter optimization module runs the optimization algorithm at a fixed period (e.g. every 60 seconds) to make adjustment decisions for the current sound stimulus parameters. Its workflow is a "evaluation - prediction - optimization - execution" cycle: Intervention utility evaluation: At the end of each optimization period, the module calculates an intervention utility score U(t) based on the changes in the user's physiological state in the last period, which is used to quantify the effectiveness of the current sound stimulus. When defining the utility U, the improvement of multiple key indicators is considered, such as heart rate decrease, HRV increase, EDA decrease, etc. A linear weighted model can be used: ; wherein represents the change of heart rate from the start of the previous cycle (a negative value indicates an improvement), represents the change of HRV indicator RMSSD (a positive value indicates an improvement), represents the change of skin conductance level (a negative value indicates an improvement). is the weight coefficient, satisfying The weight w can be dynamically set according to the user's acoustic profile and historical intervention effects (for example, the greater the weight, the more sensitive the indicator is to the user). If > 0, it means that the overall state has improved from the beginning of the cycle; < 0, it means that the effect is not good or even deteriorating. The module provides this utility score to the optimization algorithm for determining the adjustment direction, and records it in the database for post-analysis.

[0101] Candidate parameter generation: at the beginning of each new optimization cycle, the module generates a candidate parameter set Θ for the next cycle according to the current intervention mode and user profile. This step defines the search space of the optimization, so that the adjustment does not deviate from the current situation or break the safety boundary. For example: Fast intervention mode: = {ASMR, pulse sound, slow beat sound, heartbeat-synchronized ultra-low frequency pulse sound}, SPL ∈ [EMR, EMR+6 dB], r ∈ [0.8, 1.2].

[0102] Rhythm relaxation mode: = { rhythm slow beat sound}, SPL ∈ [45, 65] dB(A), r ∈ [0.5, 1.0].

[0103] Maintenance consolidation mode: = {natural sound, meditation voice, soft music}, SPL ∈ [45, 60] dB(A), r ∈ [0.8, 1.2] The module forms several candidate parameter combinations according to the above rules, each contains specific sound type selection, target sound pressure level, and rhythm-related parameters, etc. These candidate parameters all meet the mode requirements and safety constraints, laying the foundation for the next step of utility prediction.

[0104] Bayesian optimization selection: the parameter optimization module core adopts a Bayesian optimization strategy to search for the optimal parameters in the parameter space, that is, it uses a priori Gaussian Process (GP) model to predict the utility of each candidate parameter, and selects the best balance between exploration and utilization through a sampling function. The specific steps are as follows: the module inputs each candidate parameter into the GP model to obtain a predicted utility mean and prediction uncertainty (GP makes inference from past observed parameter-utility data to provide estimated utility and confidence interval for current untried parameters). Then, the acquisition function value of each candidate is calculated to evaluate its optimization value. Commonly used acquisition functions include Upper Confidence Bound (UCB) strategy: ; where is a factor to control the exploration-exploitation balance (large means more exploration of uncertain regions, small means more exploitation of high mean regions). Thompson Sampling can also be used: randomly sample a possible utility function from the posterior distribution of GP, then choose the optimal parameter on this sampled function, which is equivalent to balancing exploration / exploitation in a probabilistic sense. Regardless of the strategy, the module selects the parameter combination that maximizes the acquisition function as the optimal parameter θ* for the next period. θ* contains specific instructions for sound output (such as switching sound type to X, setting volume to Y dB, adjusting rhythm factor to r times, etc.).

[0105] Parameter adjustment and safety control: After obtaining θ*, the module compares the current parameters with the recommended parameters and decides the execution plan considering safety constraints: if the variation amplitude is within the safety range, it is directly adopted; if it exceeds, it is adjusted gradually in steps. Main control rules: Sound pressure level change limit: the volume adjustment amplitude cannot exceed ±6 dB within 30 seconds. If the increment required by θ* is too large, only a part of it (e.g. within 6 dB) is adjusted in this period, and the remaining part continues in the next period to avoid sudden volume changes causing user discomfort.

[0106] Rhythm adjustment limit: the change rate of rhythm frequency or rhythm speed r cannot exceed ±0.05 (dimensionless) per minute to prevent rapid changes that interfere with the user's adapted breathing / heartbeat rhythm.

[0107] Sound type switching smoothness: if θ* suggests switching sound types (e.g. from whisper to rain sound), the control module performs a fade-in and fade-out smooth transition: first fade out the current sound in 1-2 seconds, while the new sound fades in to the target volume, ensuring a natural and unobtrusive switch. If the recommended type does not match the current mode (e.g. recommending pulse sound in maintenance mode), the unreasonable instruction is rejected or delayed, and only type replacement is considered when mode switching.

[0108] After applying the above strategies, the final execution parameters are obtained and record the changes. The parameter optimization module then enters the next cycle of monitoring and evaluation phase, repeating the above cycle. It is worth mentioning that if the utility U is negative for consecutive multiple cycles (e.g. U(t) < 0 for 3 consecutive cycles), it indicates that the current stimulation scheme may be ineffective or even counterproductive. At this time, the optimization module triggers an early warning mechanism: instructing to reduce the stimulation intensity (e.g. slightly reducing the volume) or suggesting to terminate the intervention / artificial intervention to avoid overstimulation. The module can also make forward-looking adjustments based on the prediction model: for example, if a rebound is detected, the module will slightly reduce the volume by 2 dB or slow down the rhythm by 0.05 in advance to prevent the physiological indicators from deteriorating again.

[0109] Output results: In each optimization cycle, the parameter optimization module outputs new sound parameter configurations (type, sound pressure level, rhythm, etc.) for the control module to execute. This is equivalent to setting a "recipe" for the next period of intervention. At the same time, the utility score U calculated in this cycle and other monitoring data (such as the change in each indicator) are output for recording. For significant optimization decisions (such as changing the sound type or making a large adjustment to the volume), the module can note the decision reason in the log (e.g. "HR did not decrease in the last cycle, try switching sound type"). These outputs guide immediate intervention adjustments and, through the data archiving module, enter the database to provide training samples for subsequent model updates.

[0110] 4.3 Data archiving and model updating After each intervention process is completed, the parameter optimization and self-learning module summarizes and stores the data and effects of this intervention, and uses these data increments to update the individual model. This process ensures that the system translates experience into progress and performs more intelligently in the future. The main steps include: Intervention data archiving: After the intervention end signal is triggered, the system organizes and securely stores all relevant data generated during the intervention process. This includes: physiological signal curves (changes in HR, HRV, EDA, etc. over time), sound stimulation parameter change trajectories (when to switch sounds, how to adjust the volume, etc.), utility score U(t) for each cycle, alarm records and abnormal score trends of the panic recognition module, user subjective feedback before and after the intervention (such as how much the user self-evaluates the panic intensity has decreased), etc. The module first stores these data locally in encrypted form to ensure user privacy is not compromised. Subsequently, if the user authorizes and the network is available, the module will synchronize the data to the cloud user database through an encrypted channel for long-term storage and in-depth analysis. Each user's data is stored separately in the cloud using a unique ID.

[0111] Model updating and self-learning: After data archiving is complete (which can be real-time or timed batch processing), the system updates various models and parameters using the newly accumulated data: Gaussian Process Model Update: The parameter combination tried in this intervention and its utility result $(\theta, U)$ are added as new samples to the training set of the GP model, updating the model's posterior distribution. This improves the GP's understanding of the user's preference parameters, making the next Bayesian optimization decision more accurate and reliable. In particular, if a new parameter combination is found to be more effective than ever before, the GP will predict higher utility near it, guiding more exploration or exploitation in the future.

[0112] History Optimal Parameter Repository Update: The module extracts the parameter configuration with the highest utility during this intervention (or several high-utility configurations) from the new data and stores it in the user's "optimal parameter repository." Next time the user triggers a panic intervention, the system will preferentially call the highest-utility solution in the historical record as the starting parameter (warm start) rather than conservatively starting from the recommended value in the profile. This can reduce the parameter exploration time. For example, if the user has a seizure again within 24 hours, the system can start the intervention directly with the sound type and volume that had the best effect last time, which can speed up the effect and improve the immediate effectiveness and long-term stability of the intervention.

[0113] Panic Recognition Model Update: Use the normal and abnormal data segments newly acquired during this episode to fine-tune the model of the panic recognition module. For example, merge the calm data after the intervention into the normal data set, update the One-Class SVM and Isolation Forest model boundaries (expand the normal range); at the same time, according to the peak characteristics of this episode, check whether the setting of the fast threshold path is reasonable, and adjust the feature threshold value if necessary (refer to the formula mentioned earlier); if there is a missing sensor signal in this episode, train / enable the corresponding dimensionality reduction recognition model to ensure that the system can still work in similar situations next time.

[0114] Individual Acoustic Profile Update: If new sound types or different intensities that are not included in the profile are tried during the intervention and good results are achieved, these discoveries can be fed back to update the profile. For example, increase the preference score of a certain sound type, or correct the user's actual response threshold (if it is found that the user is more sensitive to a certain sound than when measured, then lower the corresponding EMR value). The safety parameters of the profile can also be gradually optimized based on history, for example, if multiple interventions have confirmed that +5 dB is effective, there is no need to set the upper limit to +9 dB.

[0115] Group-level self-learning: Aggregate multi-user data in the cloud for group analysis to assist system optimization. For example, unsupervised clustering can be used to divide users into groups with similar response patterns to different sound types. Extract common features for each group (e.g., some people are generally sensitive to rain sounds), and optimize initial parameter recommendations for new users based on these features. Alternatively, statistical analysis of a large amount of test data can be used to calculate the average threshold for each type of ASMR sound for the general population, which can be used as a default value for new users who have not been fully tested. These group rules do not directly change individual models, but they are added to the system's default configuration library, improving the overall intelligence of the system.

[0116] Version management and rollback: The module uses version control for model updates, preserving the model parameters before updates. When a new model performs poorly or abnormally in subsequent use, it can be quickly rolled back to the previous version to ensure safety. In addition, strict access permissions are set for archived sensitive physiological data, allowing only algorithm modules to call, preventing unauthorized personnel from viewing, and meeting medical data privacy requirements.

[0117] Output results: After model updates, the system does not produce direct visible outputs, but the internal model parameters, thresholds, and archives have been completely renewed. The next time the system runs, it will automatically load these updated models and parameters, providing better performance. For example, the recognition module uses updated thresholds and anomaly detection models (with reduced false positive rates), and the control module and optimization module refer to updated acoustic archives and GP models (more tailored to current user needs), resulting in continuous optimization of the entire system through self-learning. It is important to note that the parameter optimization and self-learning module only activates after the recognition module confirms that the user is in an episode and triggers intervention; otherwise, it remains in standby mode to conserve resources and avoid overfitting to normal state data. Through induction and summary after each intervention, the system achieves a closed-loop self-improvement, making interventions more effective and intelligent over time.

[0118] The above-described embodiments detail the technical solutions and benefits of the present application. It should be understood that the above-described embodiments are only specific embodiments of the present application and are not intended to limit the present application. Any modifications, supplements, and equivalent replacements made within the scope of the principles of the present application should be included within the scope of protection of the present application.

Claims

1. A sound stimulus adaptive control system based on abnormal state recognition, characterized in that, It includes a signal acquisition and calculation module, a panic recognition and alarm module, an audio stimulus control module, and a parameter optimization and learning module; The signal acquisition and calculation module acquires various physiological and motion sensing data of the user in real time through wearable sensors, and simultaneously fuses the acquired physiological data to calculate the anxiety index. The panic detection and alarm module first corrects motion state based on motion sensor data, then runs a fast thresholding pathway and an unsupervised model pathway in parallel, and outputs a panic attack alarm signal. and abnormal intensity score Among them, panic attack alarm signals It is a Boolean quantity. A value of 1 indicates that the user is currently experiencing a panic attack. = 0 indicates no seizure alarm or the alarm has been cleared; Abnormal intensity score The risk levels are further categorized into low risk, high risk, and high risk. The sound stimulation control module selects a sound stimulation mode based on the output of the panic recognition alarm module and the anxiety index, and outputs corresponding sound stimulation according to the user's individual profile under the selected sound stimulation mode. The parameter optimization learning module evaluates the effect of sound stimulation in real time during each stimulation intervention by the sound stimulation control module and optimizes the next stimulation parameters; at the same time, it archives and learns the entire data after the intervention ends.

2. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, The physiological signals acquired in the signal acquisition and calculation module include: electrocardiogram (ECG) signal, captured heart rate (HR), heart rate variability (HRV), respiratory waveform, and skin conductance (EDA). The collected motion sensing data includes: triaxial accelerometer data and gyroscope data.

3. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, After acquiring various physiological and motion sensor data, the signal acquisition and calculation module needs to calculate the signal quality index of each channel signal. The formula is as follows: ； In the formula, For signal-to-noise ratio, For data missing rate, The degree of the wake; , , These are the weighting coefficients; The range of values ​​is ,when Once the signal quality is deemed satisfactory, the system's panic detection and alarm module will activate. This is a preset signal quality threshold.

4. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, The specific process for calculating the anxiety index in the signal acquisition and calculation module is as follows: Physiological signals collected by different sensors are preprocessed to obtain a synchronization signal sequence. Each second in this synchronization signal sequence contains a set of synchronized physiological signal values ​​for feature extraction. Using a 1-second update step, a sliding time window is selected to calculate various physiological characteristics on the synchronization signal sequence; Calculate different weights for each physiological characteristic. Standardized bias of all physiological characteristics According to the corresponding weight Summing yields the overall deviation score. ; The composite deviation score is mapped to an anxiety index in the range of 0–100 using a sigmoid function. During mapping, the midpoint θ and slope k of the Sigmoid curve are adjusted according to the individual user situation so that 50 points correspond to the user's resting state. The instantaneous anxiety index calculated every second is smoothed using an exponential moving average, resulting in the final output time. Corresponding smoothed anxiety index for: ； In the formula, This is the smoothing coefficient.

5. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, The aforementioned panic detection alarm module performs motion state correction based on motion sensor data, specifically including: Calculate the user's average acceleration per unit time based on motion sensor data. and step frequency ; if Exceeding the preset threshold and lasting for more than 5 seconds, or step frequency If the user reaches the brisk walking / running level, the user's exercise status flag is set to TRUE; otherwise, the exercise status flag is set to FALSE. When the motion state flag is set to TRUE, the anomaly detection conditions for the fast thresholding path and the unsupervised model path are corrected; when the internal motion state flag is set to FALSE, no correction is made.

6. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, The fast thresholding pathway works as follows: Direct threshold judgment rules are set for physiological features extracted from physiological data, for each physiological feature. Set an abnormal threshold ; Whenever there is a deviation in physiological characteristics Exceeding the abnormal threshold If the direction of change is consistent with the pattern of a panic attack, then this physiological characteristic is marked as a suspicious abnormality at the current moment. ; Multiple characteristic deviations After normalizing each MDC, the weighted sums are used to obtain the joint effect size. If the combined effect size Exceeding the set threshold for combined effect size Then the suspicious anomaly will be considered. Perform a count to obtain an anomaly count. ; Based on joint effect size With joint effect size threshold The relationship between the numbers and the anomaly count The value, the output boolean exception flag. .

7. The sound stimulus adaptive control system based on abnormal state recognition according to claim 6, characterized in that, The working process of the unsupervised model pathway is as follows: Anomaly detection of physiological feature patterns is performed using pre-trained single-class support vector machines and isolated forest models, respectively; the normalized anomaly probabilities output are respectively... and .

8. The sound stimulus adaptive control system based on abnormal state recognition according to claim 7, characterized in that, The panic detection and alarm module outputs a panic attack alarm signal. and abnormal intensity score The specific process is as follows: The comprehensive anomaly strength score is calculated based on the outputs of the fast thresholding pathway and the unsupervised model pathway. The formula is: ； In the formula, , , Let be the weighting coefficient, satisfying ; The trigger condition is set using m-of-k logic, within a decision window of the most recent k seconds. If at least m moments meet the condition... This will trigger an alarm and output a panic attack alarm signal. ; This is the preset alarm threshold.

9. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, The sound stimulation control module includes a rapid intervention mode, a rhythm slowing mode, and a maintenance and consolidation mode. The rapid intervention mode is activated when the user is in a high-risk state, at which point the panic detection alarm module outputs a panic attack alarm signal. = 1 and overall anomaly intensity score Or anxiety index This mode plays fast-paced, relaxing sounds; the duration will not exceed 60 seconds. The condition for entering rhythm-easing mode is: the user's status is in the medium-risk range, at which point... or In this mode, theta rhythm beats are played for 3–5 minutes. The entry condition for maintaining consolidation mode is: the user's status is at a low-risk level. < 0.30 and In this mode, soothing background sounds or soft music play for 3-5 minutes. During the playback of each sound stimulus pattern, the system continuously monitors the real-time changes in the user's physiological indicators and obtains a comprehensive abnormality intensity score. and anxiety index To assess the effectiveness of the intervention, after each sound stimulation pattern ends, the next phase pattern is selected based on the intervention effect, or the sound stimulation is discontinued.

10. The sound stimulus adaptive control system based on abnormal state recognition according to claim 1, characterized in that, A user's personalized profile includes individualized sound stimulation thresholds, sound preferences, and safety restrictions.

Images