A progressive sound therapy scene generation method and system for sleep disorder intervention

Abstract

The application relates to the technical field of sleep monitoring, in particular to a progressive sound therapy scene generation method and system for sleep disorder intervention. The method comprises the following steps: acquiring multi-modal sleep monitoring data in a user's sleep process and performing preprocessing, extracting sleep stability feature parameters and micro-awakening precursor feature parameters; performing weighted fusion analysis based on the same, determining a risk value of a micro-awakening event occurring in a future prediction window; determining a risk level based on the same and matching corresponding scene control parameters; judging whether to switch the risk level based on an entry threshold and an exit threshold; after confirming the entry of the risk level, generating a target sound scene by progressively adjusting a current sound scene; and updating a risk determination threshold, feature weights and a scene control parameter group according to an actual sleep state after the prediction window ends. The application continuously improves the accuracy of prediction and intervention through individualized learning, thereby effectively improving the continuity and stability of overnight sleep.

Description

Technical Field

[0001] This application relates to the field of sleep monitoring technology, and in particular to a method and system for generating progressive sound therapy scenarios for sleep disorder intervention. Background Technology

[0002] Sleep is a vital physiological process for the human body, playing a crucial role in health maintenance, memory consolidation, and mood regulation. With the accelerating pace of life and increasing stress, sleep disorders are becoming increasingly prevalent, posing a significant threat to public health. In the field of sleep intervention technology, sound therapy, as a non-pharmacological intervention, improves sleep quality by playing specific sounds and has been widely applied in sleep aids and sleep management services. Existing technologies typically employ fixed audio playback strategies, including white noise, natural sounds, and soothing music; some systems recommend audio based on user history preferences; others switch sounds based on a single sleep state identification result; furthermore, there are solutions on the market that separate sleep monitoring from sleep aid playback, where monitoring data is primarily used for post-sleep quality scoring and not involved in the real-time intervention process.

[0003] However, in existing technologies, sleep intervention systems typically adjust the sound only after the user has shown significant behavioral changes, and the assessment of sleep status often relies on a single physiological indicator. The adjustment of sound therapy parameters is relatively simple and direct, lacking a continuous tracking and dynamic response mechanism for the evolution of sleep status. Summary of the Invention

[0004] This application provides a progressive sound therapy scene generation method and system for sleep disorder intervention to solve the above problems.

[0005] In a first aspect, this application provides a method for generating progressive sound therapy scenarios for sleep disorder intervention, the method comprising: S1. Acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to a preset sliding time window; S2. Based on a preset short-term analysis window and a preset trend analysis window, a weighted fusion analysis is performed on the sleep stability characteristic parameters and the micro-awakening precursor characteristic parameters to determine the micro-awakening risk value of micro-awakening events occurring within a preset prediction window in the future. S3. Determine the corresponding risk level based on the micro-awakening risk value, and determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level; S4. Based on the micro-awakening risk value and the corresponding risk level entry and exit thresholds, determine whether to enter or exit the risk level; S5. After confirming that the risk level has been entered, the current sound scene is progressively adjusted according to the target sound scene control parameters to generate the target sound scene. S6. After the prediction window ends, determine whether a micro-awakening event has occurred based on the user's actual sleep state, and update the risk judgment threshold, feature weights and / or scenario control parameter group used for subsequent micro-awakening risk value assessment based on the determination result.

[0006] This application acquires multimodal sleep monitoring data and extracts sleep stability characteristic parameters and micro-awakening precursor characteristic parameters. It obtains micro-awakening risk values ​​through weighted fusion analysis using short-term analysis windows and trend analysis windows. Then, it matches target sound scene control parameters according to the risk level, and achieves robust state switching by combining an entry / exit dual-threshold mechanism. After confirming entry, it performs progressive sound scene adjustments, and finally completes parameter closed-loop updates based on posterior verification after the prediction window. This synergistic effect enables the system to initiate flexible, graded, and adaptive sound therapy interventions before micro-awakening occurs. This avoids secondary disturbances caused by sudden sound changes and continuously improves the accuracy of prediction and intervention through individualized learning, thereby effectively improving the continuity and stability of sleep throughout the night.

[0007] Optionally, the multimodal sleep monitoring data includes at least two of the following: respiratory signals, heart rate or heart rate variability signals, body movement signals, snoring or breathing sound signals, and blood oxygenation signals. The preprocessing in S2 includes timestamp alignment, noise reduction, outlier removal, and data quality scoring. The data quality score is determined based on the signal integrity, continuity, signal-to-noise level, and temporal validity of various types of multimodal sleep monitoring data; For multimodal sleep monitoring data with a data quality score below a preset threshold, they are removed, or the feature parameters extracted from them are downweighted.

[0008] This application achieves structured control over the reliability of input signals by limiting the combination of multimodal data types, constructing a four-dimensional quality evaluation system, and implementing differentiated data processing strategies. On this basis, timestamp alignment ensures the temporal synergy of cross-modal features, denoising and anomaly removal improve the purity of single-channel signals, and the quality-driven removal or weighting mechanism suppresses the interference of poor data on micro-awakening risk prediction at the system level, significantly enhancing the robustness and generalization ability of the method in real sleep environments.

[0009] Optionally, step S2 involves extracting the sleep stability feature parameters, including: Based on the respiratory signals, the respiratory cycle variability and respiratory amplitude fluctuation rate are extracted; Based on the heart rate or heart rate variability signal, extract the amplitude of heart rate changes and the degree of heart rate fluctuation; Based on the body movement signals, the body movement frequency and body movement energy are extracted; Extract the main frequency drift and spectral structure change based on the snoring or breathing sound signal; And at least two of the above-mentioned characteristic parameters are used as the sleep stability characteristic parameters.

[0010] This application characterizes respiratory rhythm stability through respiratory cycle variability and respiratory amplitude fluctuation, reflects the state of autonomic nervous system regulation through heart rate change amplitude and heart rate fluctuation, describes limb activity activity through body movement frequency and body movement energy, and reveals airway acoustic stability through dominant frequency drift and spectral structure change. Each feature independently characterizes sleep stability in its respective physiological dimension, and is dynamically filtered and combined based on data quality scores, thereby constructing a multimodal sleep stability feature parameter set covering the four major systems of breathing, circulation, movement, and airway, providing a high-resolution, low-redundancy, and highly interpretable input basis for subsequent micro-arousal risk prediction.

[0011] Optionally, the feature parameters of the micro-awakening precursor are extracted in S2, including: Calculate the difference values ​​of each of the sleep stability characteristic parameters between adjacent sliding time windows; Calculate the slope of change of each of the sleep stability characteristic parameters between multiple consecutive sliding time windows; Determine the degree of synchronous abnormality when different types of sleep stability characteristic parameters simultaneously exceed the corresponding abnormality threshold within the same sliding time window; Based on the difference value, the slope of change, and the degree of synchronization anomaly, the micro-awakening precursor characteristic parameters are generated.

[0012] This application captures sudden micro-awakening disturbances through differential values, identifies slow deterioration trends through changing slopes, and reflects the coordinated instability of multiple systems through the degree of synchronization anomalies. These three aspects complement each other in terms of temporal granularity, change patterns, and physiological dimensions. Based on this, the micro-awakening precursor characteristic parameters can more comprehensively and robustly characterize the critical state of sleep stability, thereby supporting the precise triggering of subsequent risk grading and progressive sound therapy interventions.

[0013] Optionally, step S3 specifically includes: The micro-awakening precursor characteristic parameters are aggregated according to the short-term analysis window to obtain short-term fluctuation characteristic parameters; The sleep stability characteristic parameters are aggregated according to the trend analysis window to obtain the trend change characteristic parameters. The weighting coefficients are determined based on the data quality scores corresponding to each feature parameter; The short-term fluctuation characteristic parameters and the trend change characteristic parameters are weighted and fused based on the weighting coefficients. The micro-awakening risk value is determined based on the weighted fusion result and the preset risk judgment rules.

[0014] This application aggregates micro-arousal precursor characteristic parameters through a short-term analysis window to obtain short-term fluctuation characteristic parameters reflecting the intensity of acute disturbances; it aggregates sleep stability characteristic parameters through a trend analysis window to obtain trend change characteristic parameters characterizing the evolution of chronic instability; it determines differentiated weighting coefficients based on the data quality scores corresponding to each characteristic parameter to achieve credibility-oriented risk fusion; and it outputs risk values ​​based on the weighted fusion results and preset risk judgment rules. This technical approach enables the system to quickly respond to sudden precursor signals and identify slowly progressing sleep degradation trends, while suppressing misjudgment interference caused by low-quality signals. This ensures predictive sensitivity while improving overall robustness, effectively supporting the scientific decision-making for subsequent graded sound therapy interventions.

[0015] Optionally, in step S4, the risk level is divided into at least three levels, including low risk, medium risk and high risk; Pre-set scenario control parameter groups for different risk levels; Each of the scenario control parameter groups includes at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target hold duration.

[0016] This application divides risk levels into three levels: low risk, medium risk, and high risk. For each level, a set of scenario control parameters is preset, including at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target duration. This achieves a structured mapping between risk assessment results and sound therapy intervention actions. With the help of this mapping mechanism, the system can enhance masking stability as needed when the risk of micro-awakening increases and delay exit according to hysteresis rules when the risk decreases. This improves the continuity of sleep throughout the night and the physiological adaptability of sound therapy intervention without increasing the burden on additional sensors.

[0017] Optionally, in step S5, when the micro-awakening risk value is higher than the entry threshold of the corresponding risk level for at least two consecutive sliding time windows, it is confirmed that the risk level has been entered. When the micro-awakening risk value is lower than the exit threshold of the corresponding risk level for at least two consecutive sliding time windows, it is confirmed that the risk level will be exited. The entry threshold is higher than the exit threshold to suppress frequent switching of the sound scene caused by short-term fluctuations in the micro-awakening risk value.

[0018] This application achieves robust identification of micro-awakening risk state transitions by setting entry and exit thresholds to form a hysteresis interval and combining it with a dual confirmation mechanism of continuous sliding time windows. On this basis, the system only drives the sound scene control parameters to update after confirming the entry or exit risk level, thereby effectively suppressing frequent sound scene switching caused by short-term fluctuations in physiological signals, reducing the possibility of the sound therapy system itself becoming a sleep disturbance source, and improving the continuity of sleep throughout the night and the reliability of intervention.

[0019] Optionally, generating the target sound scene in step S6 includes: The corresponding scenario control parameter group is invoked based on the current risk level; Based on the current output state of the sound scene, the parameter combination with the smallest difference from the current sound scene is determined from the scene control parameter group as the target sound scene control parameter; Retrieve the scene template corresponding to the current risk level from the preset sound scene library; The scene template is then parametrically adjusted based on the target sound scene control parameters to generate the target sound scene. The preset sound scene library includes one or more of the following: white noise scene, pink noise scene, brown noise scene, natural sound scene, low dynamic range ambient sound scene, and composite masking scene.

[0020] This application achieves efficient generation and natural transition of target sound scenes by binding risk levels with scene control parameter groups, optimizing parameter combinations based on the current sound scene state, calling adaptive templates, and performing parameterized adjustments. With the help of parameter space distance constraints and template scalability architecture, it ensures both the hierarchical controllability of intervention strategies and maintains the individual adaptability and temporal continuity of sound therapy output. On this basis, combined with the entry / exit threshold hysteresis mechanism, it effectively suppresses frequent switching of sound scenes caused by short-term fluctuations in micro-arousal risk values, thereby avoiding secondary disturbances and improving the stability of sleep throughout the night.

[0021] Optionally, step S6 involves progressively adjusting the current sound scene, including: Within the target transition time, the output volume is gradually adjusted according to the preset gradation rate; The energy proportion of different frequency bands is gradually adjusted according to the preset ratio; The rhythm density is gradually adjusted according to the preset step size; Perform a cross-fade-in / fade-out transition between the current sound scene and the target sound scene; In step S7, it is determined whether a micro-arousal event has occurred based on whether the user exhibits at least one of the following: increased body movement, sudden changes in respiratory rhythm, abnormal increase in heart rate, abnormal changes in snoring spectrum, or decreased sleep stability. When a micro-awakening event is determined to occur within the prediction window, the weight of the abnormal feature corresponding to the current micro-awakening event is increased, and / or the entry threshold for the corresponding risk level is increased. When it is determined that no micro-awakening event has occurred within the prediction window, the current scene control parameter group is kept unchanged, or the entry threshold of the corresponding risk level is lowered, so as to form subsequent micro-awakening risk value assessment parameters and sound scene control parameters for individual users.

[0022] This application achieves seamless sound scene transitions by constraining volume gradual change rhythm with target transition duration, adjusting spectral energy distribution with preset ratio, regulating rhythm density with preset step size, and using cross-fade-in and fade-out. It also combines five types of indicators—body movement, breathing, heart rate, snoring, and stability—for micro-awakening posterior judgment. Based on this, the weights of abnormal features and risk level entry thresholds are updated differently according to the judgment results. The synergistic effect of these technical features is to ensure the continuity of the auditory pathway through multi-dimensional progressive control, ensure the reliability of the judgment through multi-source posterior verification, and drive the system to converge toward the user's optimal intervention strategy through a feature-threshold dual-track update mechanism. Ultimately, it achieves the goal of sleep disorder intervention with pre-prediction, smooth intervention, closed-loop feedback, and increasing accuracy with use.

[0023] Secondly, this application provides a progressive sound therapy scene generation system for sleep disorder intervention, the system comprising: The preprocessing module is used to acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to a preset sliding time window. The fusion analysis module is used to perform weighted fusion analysis on the sleep stability characteristic parameters and the micro-arousal precursor characteristic parameters based on a preset short-term analysis window and a preset trend analysis window, so as to determine the micro-arousal risk value of micro-arousal events occurring within a preset prediction window in the future; The control and analysis module is used to determine the corresponding risk level based on the micro-awakening risk value, and to determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level. The risk analysis module is used to determine whether to enter or exit the risk level based on the micro-awakening risk value and the corresponding entry and exit thresholds of the risk level. The scene analysis module is used to progressively adjust the current sound scene according to the target sound scene control parameters after confirming that the risk level has been entered, so as to generate the target sound scene. The parameter iteration module is used to determine whether a micro-awakening event has occurred based on the user's actual sleep state after the prediction window ends, and to update the risk judgment threshold, feature weights and / or scenario control parameter group used for subsequent micro-awakening risk value assessment based on the determination result. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart illustrating a progressive sound therapy scene generation method for sleep disorder intervention provided in one embodiment of this application; Figure 2 This is a structural diagram of a progressive sound therapy scene generation system for sleep disorder intervention provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0028] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0029] Example 1: See Figure 1 As shown in the figure, a progressive sound therapy scene generation method for sleep disorder intervention provided by an embodiment of this application is included in the following steps: Step S1: Acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to the preset sliding time window; Multimodal sleep monitoring data refers to a collection of two or more heterogeneous signal data reflecting a user's physiological and behavioral states, collected synchronously during sleep. It can be at least two of the following: respiratory signals, heart rate or heart rate variability signals, body movement signals, snoring or respiratory sounds, and blood oxygenation signals. Preprocessing includes timestamp alignment, noise reduction, outlier removal, and data quality scoring. Data quality scoring is determined based on the signal integrity, continuity, signal-to-noise ratio, and temporal validity of various types of multimodal sleep monitoring data. Multimodal sleep monitoring data with a data quality score below a preset threshold are removed, or the feature parameters extracted from them are downweighted. A preset sliding time window is used to segment continuous time-series data, with a length of 5 seconds, 8 seconds, or 10 seconds, and a sliding step size. The duration is 2 seconds, 3 seconds, or 5 seconds; the sleep stability feature parameter is used to characterize the stability of the user's current sleep state. It is a quantitative indicator calculated within a single sliding time window, reflecting the stability of dimensions such as respiratory rhythm, heart rate fluctuation, body movement intensity, snoring structure, or blood oxygen level; the micro-awakening precursor feature parameter is used to characterize the early physiological / behavioral abnormality tendency of the user about to have a micro-awakening. It is a quantitative indicator calculated between adjacent sliding time windows or within multiple consecutive sliding time windows, reflecting the changing trend, abrupt change amplitude, and coupling instability degree of the aforementioned stability feature parameter; this application extracts two types of feature parameters at the same time granularity after uniform preprocessing of the multimodal raw signal, providing a semantically consistent, temporally aligned, and quality-controllable input basis for subsequent weighted fusion analysis.

[0030] This application may acquire multimodal sleep monitoring data by means of timestamp alignment and signal-to-noise ratio evaluation; this application may also extract sleep stability feature parameters and micro-arousal precursor feature parameters by means of sliding window segmentation and quality score weighting; further, this application may generate micro-arousal precursor feature parameters by means of outlier removal and differential slope calculation; this application obtains a dual-track feature set that can be used to characterize the evolution trend of sleep state and disturbance precursors based on any of the above methods.

[0031] For example, this application could involve the system synchronously collecting body movement signals and snoring signals via a mattress pressure sensor array and a pillow-side microphone after the user falls asleep; unifying the sampling rate of the two signals (both resampled to 50Hz), aligning the timestamps (achieving cross-sensor timing synchronization with millisecond-level precision), and calculating their signal integrity (missing point ratio < 5%), continuity (no breakpoints > 200ms), signal-to-noise ratio (SNR > 15dB), and timing validity (incrementing timestamps without jumps); retaining full use of signals with a quality score ≥ 0.7, and using signals with a score between 0.4... The feature parameters extracted from signals with scores between 0.7 and 0.7 are multiplied by a weight of 0.6, while signals with scores <0.4 are directly discarded. Then, with an 8-second sliding time window and a 3-second sliding step, body movement frequency and energy are extracted from body movement signals, and the dominant frequency drift and spectral structure change are extracted from snoring signals to form sleep stability feature parameters. The difference in body movement energy between adjacent windows, the slope of the dominant frequency drift within 5 consecutive windows, and the degree of synchronization abnormality when body movement frequency and dominant frequency drift exceed their respective abnormal thresholds in the same window are calculated to generate micro-awakening precursor feature parameters.

[0032] Step S2: Based on the preset short-term analysis window and the preset trend analysis window, perform weighted fusion analysis on the sleep stability characteristic parameters and the micro-arousal precursor characteristic parameters to determine the micro-arousal risk value of micro-arousal events occurring within the preset prediction window in the future; The preset short-term analysis window can refer to the time scale used to aggregate micro-awakening precursor characteristic parameters, with a length of 2 seconds, 5 seconds, or 10 seconds, used to capture rapidly changing precursor signal mutations; the preset trend analysis window can refer to the time scale used to aggregate sleep stability characteristic parameters, with a length of 15 seconds, 30 seconds, or 60 seconds, used to characterize the slow degradation trend of sleep state; the weighted fusion analysis can refer to determining the weighting coefficients based on the data quality scores corresponding to each characteristic parameter, and performing linear or nonlinear weighted combination of short-term fluctuation characteristic parameters and trend change characteristic parameters; micro The arousal risk value can be a quantified value ranging from [0,1], used to characterize the probability or confidence level of a user experiencing a micro-awakening event within a preset prediction window in the future; the preset prediction window can be a time interval extending into the future from the current moment, with a length of 10 seconds, 20 seconds, 30 seconds, or 60 seconds, used to achieve pre-triggered intervention; this application focuses on precursory mutations through a short-term analysis window and pays attention to state degradation through a trend analysis window, and the two work together to model the risk value so that it not only reflects instantaneous disturbances but also reflects cumulative instability effects, thereby improving the robustness and foresight of the prediction.

[0033] This application may determine the micro-awakening risk value by means of aggregating precursor features within a short-term analysis window and stability features within a trend analysis window; this application may also determine the micro-awakening risk value by means of dynamic weighted fusion driven by data quality scores; furthermore, this application may determine the micro-awakening risk value by means of joint discrimination by rule engine and lightweight time series model; this application obtains a risk quantification output that can characterize the probability of micro-awakening occurring in future periods based on any of the above methods.

[0034] For example, this application may use a 5-second short-term analysis window to aggregate the generated micro-arousal precursor feature parameters by moving average to obtain short-term fluctuation feature parameters; use a 30-second trend analysis window to aggregate the extracted sleep stability feature parameters by moving median to obtain trend change feature parameters; assign weighting coefficients according to the quality scores of each original signal (e.g., respiratory signal score 0.85 → weight 0.35, body movement signal score 0.92 → weight 0.42, snoring signal score 0.76 → weight 0.23), and perform a weighted summation of the short-term fluctuation feature parameters and the trend change feature parameters; then input the weighted result into a preset mapping function (e.g., a sigmoid monotonically increasing function), normalize it to the [0,1] interval, and output the micro-arousal risk value R = 0.67 within the next 30 seconds.

[0035] Step S3: Determine the corresponding risk level based on the micro-awakening risk value, and determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level; Here, the risk level can refer to the category label obtained by discretizing the micro-awakening risk value, and the classification is based on the preset risk judgment rule; the preset scene control parameter group can refer to a set of sound therapy parameters pre-bound to a specific risk level, the content of which includes at least two of the target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target holding duration; the target sound scene control parameter can refer to the specific parameter values ​​selected from the parameter group of the corresponding risk level to guide the adjustment of the current sound scene; this application realizes the semantic conversion from uncertainty assessment to executable intervention instructions by mapping continuous risk values ​​to discrete levels, and then mapping levels to structured parameter groups, ensuring that the sound therapy strategy has clear hierarchical response logic and configurability.

[0036] This application may determine the risk level by means of risk value threshold classification and level mapping table lookup; this application may also determine the target sound scene control parameters by means of risk value interval classification and parameter group index matching; further, this application may determine the target sound scene control parameters by means of joint decision-making based on risk level and current sound scene output state; this application obtains a set of control parameters that matches the current risk situation and can directly drive the sound therapy module to execute based on any of the above methods.

[0037] For example, this application can set a five-level classification rule for risk value R∈[0,1]: R<0.20 is low risk, 0.20≤R<0.40 is low-to-medium risk, 0.40≤R<0.60 is medium risk, 0.60≤R<0.80 is medium-to-high risk, and R≥0.80 is high risk; when the above output R=0.67, it is determined to be medium-to-high risk; query the preset scene control parameter group corresponding to medium-to-high risk, which includes the target volume range [45dB, The target sound scene control parameters are: target volume 50dB, target frequency band energy distribution (low frequency: mid frequency: high frequency = 45%: 35%: 20%), target rhythm density [0.3Hz, 0.5Hz], target transition duration [30s, 45s], and target hold duration [120s, 180s].

[0038] Step S4: Based on the micro-awareness risk value and the corresponding entry and exit thresholds of the risk level, determine whether to enter or exit the risk level; The entry threshold can be the minimum risk value threshold required to trigger a certain risk level; the exit threshold can be the minimum risk value drop threshold required to remove a certain risk level; the entry threshold is higher than the exit threshold to suppress frequent switching of the sound scene caused by short-term fluctuations in the micro-awakening risk value; this application avoids misjudgment of risk level and repeated oscillation of the sound scene due to noise interference or instantaneous jitter of physiological signals by setting a dual-threshold mechanism with hysteresis characteristics, thus ensuring the stability of the intervention strategy and the continuity of the user's sleep experience.

[0039] This application may, for example, confirm the entry risk level by means of the risk value continuously exceeding the entry threshold and meeting the continuous duration condition; this application may also confirm the exit risk level by means of the risk value continuously falling below the exit threshold and meeting the continuous duration condition; furthermore, this application may also determine whether to enter or exit the risk level by means of jointly judging the direction of the risk value crossing the threshold and the historical level status; this application obtains a risk level state switching decision with anti-disturbance capability based on any of the above methods.

[0040] For example, this application may set the entry threshold for medium-to-high risk level to 0.65 and the exit threshold to 0.55; when the risk value output by the above three consecutive sliding time windows (a total of 15 seconds) is ≥0.65, it is confirmed to enter the medium-to-high risk level; after entering, if the risk value output by the five consecutive sliding time windows (a total of 25 seconds) is ≤0.55, it is confirmed to exit the medium-to-high risk level; because the entry threshold (0.65) is higher than the exit threshold (0.55), this hysteresis design effectively prevents repeated switching of sound scenes caused by small oscillations of the risk value around 0.60.

[0041] Step S5: After confirming entry into this risk level, the current sound scene is progressively adjusted according to the target sound scene control parameters to generate the target sound scene; The gradual adjustment can refer to the process of gradually changing various parameters of the sound scene in a non-step, non-abrupt manner within the target transition period; its technical purpose is to avoid inducing new micro-awakening due to sudden increases in volume, changes in frequency band, or abrupt changes in rhythm; the target sound scene can refer to the sound therapy output form that finally appears after gradual adjustment and meets the intervention needs of the current risk level; this application makes the sound therapy intervention itself a supporting element for sleep maintenance rather than a source of interference by explicitly constraining the parameter adjustment process as a gradual behavior that is continuous in time, limited in amplitude, and controllable in rhythm.

[0042] This application may achieve progressive adjustment by means of linear interpolation of volume within the target transition duration; this application may also achieve progressive adjustment by means of stepwise migration of frequency band energy proportion; further, this application may achieve progressive adjustment by means of coupling cross-fade-in / fade-out transition with gradual change of rhythm density; this application obtains a smooth, flexible and user-friendly target sound scene output based on any of the above methods.

[0043] For example, this application can be a current sound scene that is a natural sound scene (volume 42dB, frequency band distribution 50%:30%:20%, rhythm density 0.1Hz); the target sound scene control parameters are volume 50dB, frequency band distribution 45%:35%:20%, rhythm density 0.4Hz, and transition duration 35s; the system linearly increases the volume at a rate of 0.23dB per second within 35 seconds; the low-frequency energy ratio is uniformly reduced from 50% to 45%, the mid-frequency ratio is increased from 30% to 35%, and the high-frequency ratio is maintained at 20%; the rhythm density is gradually increased from 0.1Hz to 0.4Hz in steps of 0.06Hz every 5 seconds; at the same time, a 35-second cross-fade-in and fade-out is performed between the old and new scenes, so that the natural sound component decays over time and the powder noise component increases over time, ultimately generating a low-dynamic composite masking scene dominated by powder noise and with appropriate rhythm support.

[0044] Step S6: After the prediction window ends, determine whether a micro-awakening event has occurred based on the user's actual sleep state, and update the risk judgment threshold, feature weights and / or scenario control parameter groups used in the subsequent micro-awakening risk value assessment based on the determination result; The actual sleep state can refer to the user's real physiological and behavioral response confirmed by posterior observation at the end of the prediction window; micro-arousals can refer to short-term arousal tendencies that occur within the prediction window and meet clinical definitions, and their determination criteria include at least one of the following: increased body movement, sudden changes in respiratory rhythm, abnormal increase in heart rate, abnormal changes in snoring spectrum, or decreased sleep stability; updates can refer to adaptively modifying the risk judgment threshold, feature weights, and / or scenario control parameter groups based on the consistency between the prediction results and the actual results, so as to achieve individualized long-term optimization; this application constructs a complete closed loop of prediction-intervention-feedback-correction by introducing a posterior verification link, enabling the system to continuously improve prediction accuracy and intervention adaptability based on the real intervention effect.

[0045] This application may determine whether a micro-arousal event has occurred by means of a sudden increase in body energy and a simultaneous exceedance of respiratory cycle variability; this application may also determine whether a micro-arousal event has occurred by means of a combined judgment of the short-term increase in heart rate and the drift of the dominant snoring frequency; further, this application may also determine whether a micro-arousal event has occurred by means of a multimodal coupling instability index exceeding a dynamic threshold; this application obtains an objective posterior evaluation basis for the intervention effect based on any of the above methods.

[0046] For example, this application may have a prediction window of 30 seconds, with the above output R=0.67 and triggering a medium-to-high risk intervention; after the 30 seconds, the system detects that the user's body energy has increased by 210% compared to the previous 5 seconds, the respiratory cycle variability has increased by 180%, and the snoring frequency drift has reached 23Hz, with all three indicators exceeding the limit simultaneously, and is judged to have occurred as a micro-awakening event; accordingly, the system increases the weights of body energy and respiratory cycle variability in the risk prediction model by 15% and 12% respectively, and raises the entry threshold for the medium-to-high risk level from 0.65 to 0.68, while lowering the target volume limit in the corresponding scenario control parameter group from 52dB to 50dB to enhance the conservatism of the intervention.

[0047] Example 2: In one possible implementation, this application also provides multimodal sleep monitoring data including at least two of the following: respiratory signals, heart rate or heart rate variability signals, body movement signals, snoring or respiratory sound signals, and blood oxygenation signals; the preprocessing in S2 includes timestamp alignment, noise reduction, outlier removal, and data quality scoring; the data quality score is determined based on the signal integrity, continuity, signal-to-noise ratio, and temporal validity of various types of multimodal sleep monitoring data; for multimodal sleep monitoring data with a data quality score lower than a preset threshold, it is removed, or the feature parameters extracted from it are weighted down, including: Step 1: Multimodal sleep monitoring data includes at least two of the following: respiratory signals, heart rate or heart rate variability signals, body movement signals, snoring or respiratory sound signals, and blood oxygenation signals; Among them, respiratory signals can refer to the temporal physiological signals reflecting chest and abdominal undulations or airflow changes collected by pressure sensors, piezoelectric films, millimeter-wave radar, or nasal thermistors; heart rate or heart rate variability signals can refer to the heart rhythm and its adjacent RR interval fluctuation sequence obtained by photoplethysmography (PPG), electrocardiogram (ECG), or radar vital sign detection; body movement signals can refer to the temporal behavioral signals reflecting trunk micro-movements, turning tendencies, or local limb displacements obtained by accelerometers, gyroscopes, pressure sensor arrays, or millimeter-wave radar; snoring or breathing sound signals can refer to the audio signals generated by upper respiratory tract airflow vibrations collected by microphone arrays, occipital electret microphones, or bedside directional pickups; and blood oxygen signals can refer to the dynamic change sequence of arterial blood oxygen saturation measured by finger clip or earlobe PPG sensors.

[0048] In this embodiment, not all of the above five types of signals are collected. Only a combination of at least two of them is required to constitute an effective multimodal input. For example, it can be a combination of respiratory signals and body movement signals, or a ternary combination of heart rate variability signals, snoring signals, and blood oxygenation signals. Different combinations correspond to different precursor sensitive paths, thereby improving the coverage of heterogeneous micro-arousal events.

[0049] Step 2: Preprocessing in S2 includes timestamp alignment, noise reduction, outlier removal, and data quality scoring; Timestamp alignment refers to mapping multi-source signals from different sampling frequencies, start times, and hardware channels to the same reference clock system, and performing point-by-point interpolation or resampling with sub-second precision to make the signals comparable at the same physical time point. This operation provides a time consistency basis for subsequent cross-modal feature fusion. Denoising can refer to using common signal processing techniques such as bandpass filtering, wavelet thresholding, empirical mode decomposition (EMD), or adaptive noise cancellation to remove environmental electromagnetic interference, mechanical vibration artifacts, breathing harmonic crosstalk, power supply frequency noise, and sensor inherent drift. Component shifting; outlier removal can refer to identifying and removing outlier sampling points or abnormal segments caused by sensor detachment, poor contact, signal saturation, or sudden strong interference based on sliding window statistics (such as Z-score, IQR), model residual analysis, or contextual consistency tests; data quality score is a quantitative reliability index calculated independently for each type of multimodal sleep monitoring data, with a value range of 0 to 1, used to characterize whether the signal meets the reliability requirements for subsequent feature extraction within the current analysis period; this score does not participate in risk prediction modeling itself, but only serves as the basis for weight adjustment in the preprocessing stage.

[0050] Step 3: Data quality scoring is determined based on the signal integrity, continuity, signal-to-noise ratio, and temporal validity of various multimodal sleep monitoring data; Signal integrity refers to the ratio of the actual number of valid sampling points to the theoretically required number of sampling points within a preset sliding time window. For example, if a respiratory signal loses 20% of its data due to a brief sensor disconnection, its integrity score will decrease accordingly. Continuity refers to whether there is a non-human-caused interruption in the signal over a time dimension that is longer than a preset threshold of 500ms. This indicator is used to identify sensor momentary failures or packet loss. Signal-to-noise ratio (SNR) refers to the estimated ratio of the effective physiological component energy to the background noise energy within the target frequency band. For example, for snoring signals, the SNR in the 100Hz–1kHz frequency band is evaluated; for heart rate signals, the SNR in the 0.5Hz–4Hz frequency band is evaluated. Timing validity refers to whether the timestamp sequence of the signal conforms to basic timing logic such as monotonically increasing, stable step size, and no repetition or backtracking. If timestamp jumps, rollbacks, or out-of-order occurrences occur, the timing validity is deemed compromised.

[0051] In this embodiment, the above four indicators are calculated separately, and then normalized and weighted to obtain the final data quality score. The weight of each indicator can be preset according to the device type, wearing method and clinical validation results. For example, integrity and signal-to-noise level have higher weights, while continuity and temporal effectiveness have lower weights. This scoring mechanism does not rely on specific model training and has plug-and-play characteristics.

[0052] Step 4: For multimodal sleep monitoring data with a data quality score lower than the preset threshold, remove them or reduce the weight of the feature parameters extracted from them; The preset threshold is either a fixed value of 0.6 or an adaptive threshold that is dynamically adjusted based on the user's historical data. This threshold is used to distinguish between usable and unreliable data sources. Elimination completely ignores the signal in the subsequent feature extraction process and does not participate in any feature calculation or fusion. This is suitable for cases with severely degraded quality, such as continuous loss of >30% of data. Weight reduction is performed during the feature fusion stage by multiplying the feature parameters contributed by the signal by an attenuation coefficient less than 1, such as 0.3, 0.5, or 0.7, so that its impact on the final micro-arousal risk value is reduced but not zero. This is suitable for cases with mild noise or local distortion but still retaining some physiological information.

[0053] This application, for example, determines whether to trigger elimination based on a joint assessment of signal integrity and signal-to-noise ratio; it can also determine whether to perform weight reduction based on a joint assessment of continuity and temporal validity; further, it can dynamically select elimination or weight reduction strategies based on the comprehensive scoring results of the four indicators. Based on any of the above methods, this application achieves robust screening and flexible weighting of multimodal input data, ensuring that subsequent micro-awakening risk assessment is not dominated by low-quality signals.

[0054] Example 3: In another embodiment, this application also provides step S2 for extracting sleep stability feature parameters, including: Step 1: Extract respiratory cycle variability and respiratory amplitude fluctuation rate based on respiratory signals; Among them, respiratory cycle variability can be the ratio of the standard deviation to the mean of the length of adjacent respiratory cycles within a preset sliding time window, used to characterize the regularity of respiratory rhythm; respiratory amplitude fluctuation rate can be the ratio of the mean of the absolute value of the first difference of the peak amplitude of the respiratory waveform to the mean of the original amplitude within the same sliding time window, used to characterize the dynamic trend of respiratory intensity.

[0055] This application can, for example, determine the respiratory cycle variability and respiratory amplitude fluctuation rate by combining zero-crossing detection and envelope extraction of the respiratory waveform; it can also extract the instantaneous frequency and instantaneous amplitude of the respiratory signal after bandpass filtering, and calculate the above two indicators based on their time-series statistical distribution; furthermore, this application can obtain the analytic signal based on the Hilbert transform of the respiratory signal, and then derive the instantaneous cycle and instantaneous amplitude sequence to complete the calculation. This application obtains a quantitative characterization of respiratory rhythm stability based on any of the above methods, providing comparable and modelable physiological dimension inputs for subsequent fusion analysis.

[0056] Step 2: Extract the amplitude of heart rate changes and the degree of heart rate fluctuations based on heart rate or heart rate variability signals; Among them, the amplitude of heart rate change can refer to the maximum deviation of the instantaneous heart rate value from the baseline heart rate value at the beginning of the preset sliding time window, which is used to reflect the trend of heart rate jump or drop on a short time scale; the degree of heart rate fluctuation can refer to the standard deviation (SDNN) of the RR interval sequence or the mean of the absolute values ​​of adjacent RR differences (RMSSD) within the same sliding time window, which is used to characterize the dynamic fluctuation level of autonomic nervous system regulatory activity.

[0057] This application, for example, can calculate the amplitude of heart rate changes and the degree of heart rate fluctuation based on the peak value detection and RR interval sequence reconstruction of the photoplethysmography (PPG) signal; it can also directly derive the above indicators based on the RR interval obtained from the ECG R wave detection; furthermore, it can indirectly deduce the degree of heart rate fluctuation based on the slope of the change in the ratio of low-frequency power (LF) to high-frequency power (HF) in HRV frequency domain analysis. This application obtains a quantitative characterization of the autonomic nervous system excitation state based on any of the above methods, supporting the identification of sympathetic activation precursors in micro-arousal risk assessment.

[0058] Step 3: Extract body movement frequency and body movement energy based on body movement signals; Among them, the frequency of body movement can refer to the number of events in which the amplitude of the acceleration signal exceeds the preset motion threshold (0.15 g) within a preset sliding time window, which is used to characterize the intensity of limb activity; the energy of body movement can refer to the time integral value of the sum of squares of the three-dimensional acceleration signal within the same sliding time window, which is used to characterize the overall intensity and duration of body movement behavior.

[0059] This application, for example, can detect body movement events and count their frequencies using an adaptive thresholding method after high-pass filtering (0.5 Hz) of the output signal from a triaxial accelerometer, while simultaneously performing a sliding window integration on the sum of squares of the original signal to obtain body movement energy. Alternatively, this application can accumulate energy density within the 0.5–3 Hz frequency band based on the short-time Fourier transform energy spectrum of the body movement signal to characterize body movement energy. Furthermore, this application can combine the spatial distribution change rate of the mattress pressure sensor array to count and weight local load abrupt events to improve the robustness of body movement recognition. Based on any of the above methods, this application obtains a quantitative characterization of limb activity states during sleep to identify early body movement enhancement trends driven by micro-arousals.

[0060] Step 4: Extract the main frequency drift and spectral structure change based on snoring or breathing sound signals; Among them, the main frequency drift can refer to the offset of the main peak frequency of the power spectrum of snoring or breathing audio signal relative to the main frequency of the reference window within a preset sliding time window, in Hz, which is used to reflect the slow change in the degree of upper airway opening; the spectrum structure change can refer to the mean Euclidean distance of the dynamic difference parameter of Mel frequency cepstral coefficients (MFCC) within the same sliding time window, which is used to characterize the time-varying characteristics of snoring harmonic structure, formant distribution and noise component ratio.

[0061] This application, for example, can extract the power spectrum of an audio signal acquired by a microphone after pre-emphasis, framing, and windowing, locate the main peak and calculate its offset, and simultaneously extract the MFCC and its first-order difference, calculating the distance between dynamic features of adjacent windows to obtain the spectral structure change. Alternatively, this application can construct a spectral structure change index based on the short-time spectral entropy and the rate of change of the spectral centroid of the audio signal. Furthermore, this application can use wavelet packet decomposition to extract the multi-band energy ratio and characterize the spectral structure stability using its time series variance. Based on any of the above methods, this application obtains a quantitative characterization of the consistency between airway mechanical state and acoustic output, providing acoustic evidence for identifying pre-micro-awakening airway instability.

[0062] Step 5: Use at least two types of feature parameters as sleep stability feature parameters.

[0063] Among them, at least two types of feature parameters can refer to any combination of two or more feature parameters extracted from the above, including but not limited to: the combination of respiratory cycle variability and heart rate change amplitude; the combination of body kinetic energy and dominant frequency drift; the combination of respiratory amplitude fluctuation rate, heart rate fluctuation degree, and spectral structure change; or the complete set of all six types of feature parameters. This combination method does not rely on a fixed pairing relationship, but is dynamically selected based on the data quality score of each signal—when the quality score of a certain type of signal is lower than a preset threshold (0.6), its corresponding feature parameters are automatically excluded from the combination; only the features extracted from the signals with qualified quality scores are retained to form the currently available subset of sleep stability feature parameters.

[0064] This application, for example, can determine the current sleep stability feature parameters based on a respiratory signal quality score of 0.85, a heart rate signal quality score of 0.72, a body movement signal quality score of 0.91, and a snoring signal quality score of 0.53, using only six features: respiratory cycle variability, respiratory amplitude fluctuation rate, heart rate change amplitude, heart rate fluctuation degree, body movement frequency, and body movement energy. Alternatively, this application can add the dominant frequency drift and spectral structure change to the feature parameter set after the snoring signal quality score rises back to 0.78. Furthermore, this application can personalize the feature types retained based on historical validity feedback during multi-night learning, considering different user preferences. This application obtains a set of sleep stability feature parameters with data adaptive capabilities based on any of the above methods, ensuring the reliability and individual adaptability of feature input.

[0065] Example 4: In one possible implementation, this application also provides for extracting micro-awakening precursor feature parameters in S2, including: Step 1: Calculate the difference values ​​of each sleep stability characteristic parameter between adjacent sliding time windows; The difference value can refer to the numerical difference between the current sliding time window and the previous sliding time window for the same type of sleep stability feature parameters. The difference value is a quantitative indicator that reflects the degree of abrupt change or transition of the feature within adjacent time segments; In this embodiment, the difference value is used to characterize the instantaneous disturbance intensity of any sleep stability characteristic parameter such as respiratory cycle variability, heart rate change amplitude, body movement frequency, and snoring frequency drift on a short time scale, serving as the basis for identifying early sudden precursors of micro-awakening. This application, for example, determines whether a sudden disturbance of respiratory rhythm occurs based on the numerical difference in respiratory cycle variability between the current sliding time window and the previous sliding time window; this application, for example, determines whether a short-term increase in autonomic nerve excitation occurs based on the numerical difference in the amplitude of heart rate change between the current sliding time window and the previous sliding time window; furthermore, this application identifies whether there is a sudden accumulation of micro-body energy based on the numerical difference in body kinetic energy between the current sliding time window and the previous sliding time window; this application obtains a quantitative characterization of the precursor to sudden micro-awakening based on any of the above methods.

[0066] Step 2: Calculate the slope of change of each sleep stability characteristic parameter between multiple consecutive sliding time windows; The slope of change can refer to the linear regression slope obtained by fitting the time series of the same type of sleep stability characteristic parameters in N consecutive sliding time window sequences. The slope of the change is a quantitative indicator that reflects the strength and direction of the continuous evolution trend of this feature on a medium time scale; In this embodiment, the slope of change is used to characterize the monotonically increasing or decreasing trend of any sleep stability characteristic parameter, such as the fluctuation rate of respiratory amplitude, the degree of heart rate fluctuation, the frequency of body movements, and the amount of change in the spectral structure of snoring, within multiple consecutive time windows, serving as the basis for identifying the precursors of slow-progressing micro-awakening. This application, for example, fits a linear model to the time series of respiratory amplitude fluctuations within five consecutive sliding time windows, obtains its slope value, and determines whether respiratory instability shows a continuously aggravating trend; this application, for example, fits a linear model to the time series of heart rate fluctuations within five consecutive sliding time windows, obtains its slope value, and identifies whether heart rate variability shows a gradual increase; further, this application fits a linear model to the time series of changes in snoring spectral structure within five consecutive sliding time windows, obtains its slope value, and assesses whether the upper airway vibration pattern has undergone a continuous shift; this application obtains a quantitative characterization of the slow-progressing precursors of micro-awakening based on any of the above methods.

[0067] Step 3: Determine the degree of synchronous abnormality when different types of sleep stability characteristic parameters simultaneously exceed the corresponding abnormal threshold within the same sliding time window; Among them, the degree of synchronization anomaly can refer to the coupling strength index formed by the combination of multiple different types of sleep stability characteristic parameters that independently exceed their corresponding preset anomaly thresholds within the same sliding time window. Synchronization anomaly is a quantitative indicator that reflects the degree of coordinated instability of multimodal signals such as respiration, heart rate, body movement, and snoring in the time dimension. In this embodiment, the degree of synchronization anomaly is used to identify multi-system synchronous activation phenomena, such as respiratory rhythm disorder, increased heart rate, enhanced body movement and snoring spectrum drift occurring in the same time window. Such cross-modal coupling anomalies are highly specific precursors to micro-awakening. For example, this application calculates the synchronization count and normalizes it to a value between 0 and 1 based on the state where the respiratory cycle variability, heart rate change amplitude, and body movement frequency all exceed their respective abnormal thresholds within the same sliding time window; this application also calculates the co-occurrence probability of the three synchronization exceeding limits within the same sliding time window using the Jaccard similarity coefficient based on the synchronization exceeding limits of the respiratory amplitude fluctuation rate, snoring frequency drift, and body movement energy; furthermore, this application obtains the synchronization anomaly coverage rate by dividing the number of feature types exceeding the abnormal threshold among all five types of stability features (respiration, heart rate, body movement, snoring, and blood oxygenation) within the same sliding time window by the total number of feature types; this application obtains a quantitative characterization of the multimodal coupling instability state based on any of the above methods.

[0068] Step 4: Based on the difference value, the slope of change, and the degree of synchronization anomaly, generate micro-awakening precursor characteristic parameters; Among them, the micro-awakening precursor feature parameters are a composite feature vector composed of three types of sub-features: difference value, change slope and synchronization anomaly degree, through weighted combination; The micro-awakening precursor feature parameters are a set of multi-dimensional dynamic features used to characterize the probability of a user experiencing micro-awakening within a future prediction window; In this embodiment, the micro-awakening precursor characteristic parameters are integrated with short-term mutation, mid-term trend and cross-modal coupling information to form a more complete model of the micro-awakening mechanism and avoid misjudgment from a single dimension. This application, for example, assigns weights of 0.3, 0.4, and 0.3 to the difference value, the slope of change, and the degree of synchronization anomaly, respectively, and then performs a linear weighted summation to generate a one-dimensional risk sub-feature; this application, for example, concatenates the three types of sub-features into a three-dimensional vector, inputs it into a lightweight classifier, and outputs the precursor confidence; furthermore, this application multiplies the difference value by the slope of change and then adds it to the degree of synchronization anomaly to construct a nonlinear coupled response term; this application obtains a robust precursor feature representation for micro-awakening prediction tasks based on any of the above methods.

[0069] Example 5: In another optional embodiment, this application also provides step S3, which specifically includes: Step 1: Aggregate the characteristic parameters of micro-awakening precursors according to the short-term analysis window to obtain short-term fluctuation characteristic parameters; Among them, the short-term analysis window is 2 seconds, 5 seconds or 10 seconds, which is a time window used to capture the dynamics of short-term physiological disturbances before micro-arousal occurs; The micro-awakening precursor characteristic parameters include the difference values ​​of each sleep stability characteristic parameter between adjacent sliding time windows, the change slope of each sleep stability characteristic parameter between multiple consecutive sliding time windows, and the degree of synchronous abnormality when different types of sleep stability characteristic parameters simultaneously exceed the corresponding abnormal threshold within the same sliding time window. These are quantitative indicators used to reflect the intensity, rate of change, and degree of multimodal coupling instability of physiological signal mutations near the current moment. The short-term fluctuation characteristic parameters are comprehensive representations obtained by statistically aggregating the characteristic parameters of micro-awakening precursors within a short-term analysis window. They include the mean, maximum value, standard deviation, peak amplitude, percentage of duration of abrupt changes, or weighted cumulative value. The short-term fluctuation characteristic parameter is used to characterize the possibility of micro-arousal caused by acute disturbance within the future prediction window. The higher the value, the more unstable the recent physiological state and the more significant the sudden disturbance. This application obtains short-term fluctuation characteristic parameters by extracting the sliding maximum value of the difference value, detecting the peak value of the change slope, and performing Boolean logic AND operation on the degree of synchronization anomaly. This application obtains short-term fluctuation characteristic parameters by performing local energy integration on the difference value sequence, accumulating and summing the change slope, and weighting and counting the degree of synchronization anomaly. This application normalizes the results of the above-mentioned multiple aggregation methods and then takes a weighted average to obtain short-term fluctuation characteristic parameters.

[0070] This application obtains short-term fluctuation characteristic parameters for characterizing the intensity of short-term disturbances based on any of the above methods.

[0071] Step 2: Aggregate the sleep stability characteristic parameters according to the trend analysis window to obtain the trend change characteristic parameters; The trend analysis window, which is 15 seconds, 30 seconds, or 60 seconds, is a time window used to assess the evolution trend of overall sleep stability. Sleep stability characteristic parameters include respiratory cycle variability and respiratory amplitude fluctuation rate extracted based on respiratory signals, heart rate change amplitude and heart rate fluctuation degree extracted based on heart rate or heart rate variability signals, body movement frequency and body movement energy extracted based on body movement signals, and main frequency drift and spectral structure change extracted based on snoring or breathing sound signals. These are used to reflect quantitative indicators of basic sleep quality dimensions such as autonomic nervous system regulation ability, respiratory rhythm maintenance ability, body rest level and upper airway stability over a longer time scale. The trend change characteristic parameters are comprehensive representations obtained by modeling the long-term trend of sleep stability characteristic parameters within the trend analysis window. They include linear fitting slope, moving average deviation, trend stability index, trace of multidimensional feature covariance matrix, or state drift estimated based on Kalman filter. Trend change characteristic parameters are used to characterize the trend of decline or enhancement of sleep stability over time, and the direction and magnitude of their numerical changes together indicate the development trend of chronic instability risk. This application obtains trend change characteristic parameters by performing a 30-second sliding linear regression on the variability of the respiratory cycle, calculating the moving standard deviation of the heart rate fluctuation, and extracting the trend after low-pass filtering of the body movement frequency. This application obtains trend change characteristic parameters by performing cumulative offset statistics on the snoring frequency drift, performing exponential weighted moving average on the respiratory amplitude fluctuation rate, and extracting the trend of the first principal component after principal component projection on multiple stability features. This application obtains trend change characteristic parameters by performing interval normalization and weighted fusion on the results of the above-mentioned multiple trend modeling methods.

[0072] This application obtains trend change characteristic parameters for characterizing the long-term stability evolution trend based on any of the above methods.

[0073] Step 3: Determine the weighting coefficients based on the data quality scores corresponding to each feature parameter; Among them, the data quality score is a real number between 0 and 1, which is a quantitative reliability index determined based on the signal integrity, continuity, signal-to-noise level and temporal validity of various multimodal sleep monitoring data; The weighting coefficient is a value that is positively correlated with the data quality score. It is the data quality score itself, its square, or a normalized value after being mapped by the Sigmoid function. The weighting coefficient is used to adjust the contribution weight of feature parameters from different sources in the fusion process, so that high-confidence signals dominate risk judgment and low-confidence signals have limited influence, thereby improving the robustness of the system under occasional sensor anomalies or environmental interference. This application sets the weighting coefficient directly based on the data quality score, i.e., weighting coefficient = data quality score; This application sets the weighting coefficient based on the square of the data quality score, that is, weighting coefficient = (data quality score)²; This application inputs the data quality score into the Sigmoid function and takes the output value as the weighting coefficient, i.e., weighting coefficient = 1 / (1 + e) -k(score-threshold) ), where k is a constant for adjusting steepness, and threshold is a preset quality threshold.

[0074] This application obtains weighted coefficients that are adapted to the confidence levels of each feature parameter based on any of the methods described above.

[0075] Step 4: Perform weighted fusion of short-term fluctuation characteristic parameters and trend change characteristic parameters based on weighting coefficients; Among them, weighted fusion is the operation of multiplying the short-term fluctuation characteristic parameter and the trend change characteristic parameter by their respective weighting coefficients and then adding or weighting them. The weighted fusion results are used to comprehensively reflect the synergistic effect of short-term disturbance intensity and long-term stability trend, forming an intermediate risk characterization that balances sensitivity and robustness. This application obtains the weighted fusion result based on the product of short-term fluctuation characteristic parameters and their weighting coefficients, and the sum of the products of trend change characteristic parameters and their weighting coefficients; This application obtains a weighted fusion result based on the weighted average of the product of short-term fluctuation characteristic parameters and their weighting coefficients, and the product of trend change characteristic parameters and their weighting coefficients. This application performs nonlinear mapping (such as logarithmic compression or piecewise linear transformation) on short-term fluctuation characteristic parameters and trend change characteristic parameters respectively, and then weights and fuses them to alleviate extreme value interference.

[0076] This application obtains a weighted fusion result that integrates short-term disturbances and long-term trends based on any of the above methods.

[0077] Step 5: Determine the micro-awakening risk value based on the weighted fusion results and the preset risk assessment rules.

[0078] Among them, the preset risk judgment rule is to map the weighted fusion result into a function relationship of continuous risk values ​​in the interval [0,1], which can be linear mapping, piecewise linear mapping, Sigmoid mapping or table lookup mapping; The micro-awakening risk value is used to quantify the probability or tendency of a user to experience a micro-awakening event within a preset prediction window in the future. The preset prediction window is 10 seconds, 20 seconds, 30 seconds or 60 seconds. The higher the value, the greater the prediction risk. This application obtains the micro-arousal risk value based on the weighted fusion result after linear normalization, namely R = min(max((F F min ) / (F max F min ), 0), 1), where F is the weighted fusion result, F min With F max This represents the minimum and maximum fusion values ​​obtained from historical statistics. This application obtains the micro-arousal risk value, R = 1 / (1 + e^(-1 / 2)), by inputting the weighted fusion result into the Sigmoid function. -a(f-b) ), where a and b are empirical adjustment parameters; This application inputs the weighted fusion result into a preset lookup table and obtains the corresponding micro-awakening risk value through interpolation.

[0079] This application obtains the final output micro-awakening risk value based on any of the above methods.

[0080] Example 6: In another embodiment, this application also provides for dividing the risk level into at least three levels, including low risk, medium risk, and high risk; pre-setting scene control parameter groups for different risk levels; each scene control parameter group includes at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target holding duration: Step 1: Divide the risk level into at least three levels, including low risk, medium risk and high risk; Low risk can refer to a micro-arousal risk value within a preset first threshold range, which corresponds to a relatively stable sleep state for the user and no significant synergistic anomalies in the aura signals; medium risk can refer to a micro-arousal risk value within a preset second threshold range, which corresponds to at least two types of aura feature parameters simultaneously exceeding the abnormal threshold, or short-term fluctuation feature parameters and trend change feature parameters showing an inverse deviation; high risk can refer to a micro-arousal risk value within a preset third threshold range, which corresponds to a situation where multimodal aura feature parameters continuously increase within a continuous sliding time window, and the degree of synergistic anomaly exceeds a preset coupling threshold.

[0081] Low risk, medium risk, and high risk constitute a progressive risk response hierarchy. The purpose of this division is to enable the risk assessment results to carry the semantic expression of differentiated intervention intensity, rather than just making a binary judgment. This division provides a structured anchor point for the subsequent configuration of sound scene parameters, ensuring that the intervention granularity matches the risk triggering of different severity levels.

[0082] Step 2: Pre-set scenario control parameter groups for different risk levels; The scene control parameter set can refer to a set of constraints used to define the generation of the target sound scene. Its composition does not depend on the specific sound source content, but focuses on the behavioral attributes of the acoustic output.

[0083] This application may determine the corresponding parameter group based on a risk level and parameter group mapping table; it may also call a parameter group template pre-stored in local memory based on the current risk level; further, it may load a personalized parameter group that matches the current risk level based on the historical best parameter group index recorded in the user's individual profile.

[0084] This application obtains an executable basis for sound scene control that is tied to the current risk level based on any of the above methods.

[0085] Step 3: Each scenario control parameter group must include at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target hold duration; The target volume range refers to the numerical range between the minimum and maximum volume that the sound therapy output signal is allowed to reach in the target sound scene. Its function is to limit the upper limit of the intervention intensity and avoid the auditory startle reflex caused by a sudden increase in volume. In this embodiment, the target volume range corresponding to low risk is [35 dB, 42 dB], the target volume range corresponding to medium risk is [40 dB, 48 dB], and the target volume range corresponding to high risk is [42 dB, 50 dB]. The target frequency band energy distribution can refer to the set ratio of the energy proportions of the low-frequency band (≤250 Hz), mid-frequency band (250 Hz–2kHz), and high-frequency band (>2 kHz) in the target sound scene. Its function is to affect the auditory masking effectiveness and neural arousal level by adjusting the center of gravity of the spectrum. In this embodiment, the target frequency band energy distribution corresponding to low risk is [45%, 40%, 15%], the target frequency band energy distribution corresponding to medium risk is [50%, 35%, 15%], and the target frequency band energy distribution corresponding to high risk is [60%, 30%, 10%]. The target rhythm density can refer to the frequency of sound events with a perceptible periodic structure in a sound scene per unit time. Its function is to regulate the intensity of the influence of auditory rhythm on the autonomic nervous system. In this embodiment, the target rhythm density corresponding to low risk is 0.2–0.5 Hz, the target rhythm density corresponding to medium risk is 0.1–0.3 Hz, and the target rhythm density corresponding to high risk is 0–0.1 Hz. The target transition duration can refer to the time span required to switch from the current sound scene to the target sound scene. Its function is to control the rate of change of acoustic parameters and prevent transient disturbances. In this embodiment, the target transition duration corresponding to low risk is 30-60 seconds, the target transition duration corresponding to medium risk is 20-40 seconds, and the target transition duration corresponding to high risk is 10-30 seconds. The target retention duration can refer to the shortest duration during which the target sound scene remains unchanged after the switch is completed. Its function is to suppress invalid repeated switching caused by a short-term drop in the micro-awakening risk value. In this embodiment, the target retention duration corresponding to low risk is 60–120 seconds, the target retention duration corresponding to medium risk is 120–240 seconds, and the target retention duration corresponding to high risk is 240–480 seconds.

[0086] The combination of values ​​of the above five types of parameters at each risk level constitutes the complete definition of the parameter group; each parameter group must contain at least two of them to ensure effective constraints on the output behavior of the sound scene; when a parameter group contains three or more parameters, the parameters work together to serve the intervention target at the same risk level.

[0087] Example 7: In an optional implementation, this application also provides a mechanism for determining whether to enter or exit a risk level in step S5, including: Step 1: When the micro-awakening risk value is higher than the entry threshold of the corresponding risk level for at least two consecutive sliding time windows, confirm entry into that risk level; The entry threshold can be the lower limit of the micro-awakening risk value set to trigger a certain risk level; the sliding time window can be a time segment used to continuously calculate the micro-awakening risk value, the length of which is consistent with the preset sliding time window, such as 5 seconds, 8 seconds or 10 seconds; at least two consecutive sliding time windows can be two or more sliding time windows that are adjacent in time and without interval, and the corresponding micro-awakening risk values ​​all meet the condition of being higher than the entry threshold corresponding to the risk level.

[0088] The entry threshold is functionally used as an initial criterion to characterize the substantial deterioration trend of a user's sleep state; its role is to avoid misjudging the abnormally high risk value caused by transient physiological disturbances (such as brief body movements, occasional deviations in respiratory rhythm, or environmental noise interference) within a single time window as a real micro-awakening precursor, thereby preventing premature and falsely triggered adjustments to the sound scene.

[0089] This application can, for example, determine whether the risk value sequences of the current sliding time window and the previous sliding time window are both higher than the entry threshold for the corresponding risk level based on both. Alternatively, it can determine whether two consecutive time windows satisfy the condition of being higher than the entry threshold based on the risk value sequences of the current sliding time window, the previous sliding time window, and the two previous sliding time windows. Furthermore, this application can also be based on a sliding queue cache structure to maintain the risk values ​​of the most recent N sliding time windows in real time, and at any given time check whether there exists a continuous subsequence of length 2 whose entire value is higher than the entry threshold. This application obtains stable confirmation of the risk level entry state based on any of the above methods, ensuring that the initiation of sound scene intervention has a time-based continuity basis.

[0090] Step 2: When the micro-awakening risk value is lower than the exit threshold of the corresponding risk level for at least two consecutive sliding time windows, confirm exit from that risk level; The exit threshold can refer to the upper limit of the micro-awakening risk value set to remove a certain risk level; its value is strictly lower than the entry threshold corresponding to the same risk level; the definition of at least two consecutive sliding time windows is the same as above.

[0091] The exit threshold is functionally used as a termination criterion to indicate that the user's sleep state has substantially returned to stability. Its purpose is to avoid repeated start-stop of the sound scene due to small oscillations of the risk value near the threshold (such as being affected by occasional changes in snoring or a short-term drop in heart rate), thus ensuring the continuity of the intervention process and the stability of sleep maintenance.

[0092] This application can, for example, determine whether the risk value sequences of the current sliding time window and the previous sliding time window are both below the exit threshold for the corresponding risk level; it can also store the most recent M risk values ​​in a circular buffer and scan sequentially to see if two consecutive values ​​are below the exit threshold; further, this application can introduce a weighted moving average mechanism to further verify that, in addition to satisfying the condition that two consecutive original values ​​are below the exit threshold, their neighborhood weighted average is also below the exit threshold, thereby enhancing noise resistance. This application obtains a robust determination of the risk level exit state based on any of the above methods, ensuring that the exit from sound scene intervention has sufficient evidence of recovery.

[0093] Step 3: The entry threshold is higher than the exit threshold to suppress frequent switching of sound scenes caused by short-term fluctuations in the micro-awakening risk value; Among them, the entry threshold being higher than the exit threshold constitutes a hysteresis interval, and the difference reflects the system's tolerance for changes in risk status. Mathematically, this hysteresis interval is represented by two non-overlapping threshold intervals under the same risk level: [exit threshold, entry threshold) is an intermediate transition zone. Within this interval, the risk value neither triggers entry nor exit, and the system maintains the current risk level and the corresponding sound scene control strategy unchanged.

[0094] This hysteresis interval functionally simulates the hysteresis characteristics in a control system. Its technical role is to decouple the transient disturbances and steady-state trends of risk values, enabling the sound scene control logic to have state memory capabilities. Its contribution to the overall process is twofold: on the one hand, it prevents the high-frequency jitter of micro-awakening risk values ​​near the threshold from causing repeated switching of the sound scene; on the other hand, it ensures that intervention measures have a minimum duration of action, avoiding ineffective interventions that are adjusted and then withdrawn immediately.

[0095] For example, this application could set the entry threshold for medium-risk levels to 0.40 and the exit threshold to 0.32, forming a hysteresis interval with a width of 0.08. Alternatively, it could set the entry threshold for high-risk levels to 0.80 and the exit threshold to 0.70, forming a hysteresis interval with a width of 0.10. Furthermore, this application could dynamically adjust the width of the hysteresis interval based on the historical false trigger frequency of different risk levels; for example, configuring a wider hysteresis for high-risk levels to enhance protection, and configuring a narrower hysteresis for low-risk levels to improve response sensitivity. This application constructs a robust boundary for risk level state transitions based on any of the above methods, supporting the stable execution of sound therapy intervention strategies.

[0096] Example 8: In another optional embodiment, this application also provides that the generation of the target sound scene in step S6 includes: Step 1: Invoke the corresponding scene control parameter group based on the current risk level; The current risk level is any one of low, medium, or high risk. The scene control parameter group is a set of acoustic control parameters pre-bound to the risk level, which includes at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target hold duration. The scene control parameter group is configured during the system initialization phase and can be dynamically updated according to the user's individual profile. In this embodiment, the scene control parameter group does not directly define the complete acoustic signal waveform, but exists in the form of adjustable dimensions for subsequent parameterized driving of the acoustic scene template.

[0097] Step 2: Based on the current output state of the sound scene, determine the parameter combination with the smallest difference from the current sound scene from the scene control parameter group as the target sound scene control parameter; The current output state of the sound scene can refer to the real-time values ​​of the currently playing sound scene in quantifiable dimensions such as volume envelope, frequency band energy ratio, rhythm density, cycle period, and spatial sense. The minimum difference can refer to calculating the Euclidean distance or weighted Manhattan distance between the values ​​of each dimension of the current sound scene and each set of parameters in the scene control parameter group within a preset parameter space, and selecting the set with the smallest distance as the target parameter. The weights of each dimension in the parameter space are set according to their sensitivity to sleep disturbances; for example, the weight of the volume change dimension is higher than that of the rhythm density dimension. In this embodiment, the difference calculation process does not depend on the specific audio decoding format, but is only executed based on the standardized parameter vector. Through this mechanism, the deviation between the target sound scene control parameters and the current state is ensured to be controllable, providing a reasonable starting point for subsequent gradual adjustments.

[0098] Step 3: Retrieve the scene template corresponding to the current risk level from the preset sound scene library; The preset sound scene library includes one or more of the following: white noise scenes, pink noise scenes, brown noise scenes, natural sound scenes, low dynamic range ambient sound scenes, and composite masking scenes. Scene templates are pre-stored audio base materials with fixed structural characteristics. Each template has a recognizable acoustic fingerprint; for example, the power spectral density of a white noise template is flat, while that of a pink noise template is... 3 dB / octave attenuation, brown noise template present With 6 dB / octave attenuation, the natural sound scene template includes typical time-frequency structures of rain, stream, or forest wind sounds. The low dynamic range ambient sound scene template features low RMS volatility and narrowband energy concentration. The composite masking scene template integrates a noise floor layer and a weak rhythmic guidance layer. The invocation action is completed by looking up a table based on the current risk level. Multiple candidate templates can correspond to the same risk level. The specific selection is determined based on the user's historical preference data or the current multimodal signal status. For example, at a medium risk level, if the user has previously responded well to natural sounds and the current breathing rhythm is becoming disordered, the natural sound scene template is invoked. If the current body energy is rising but breathing is stable, the low dynamic range ambient sound template is switched to.

[0099] Step 4: Based on the target sound scene control parameters, the scene template is parametrically adjusted to generate the target sound scene; Parametric adjustment can refer to mapping the aforementioned determined target sound scene control parameters onto the aforementioned invoked scene template, and non-destructively adjusting its original acoustic properties. Mapping methods include, but are not limited to: applying a smooth gain curve to the volume envelope, applying dynamic equalizer coefficients to the frequency band energy distribution, adjusting the pulse trigger interval to the rhythm density, setting a random jitter range to the cycle period, and injecting binaural delay or reverberation feedback to the spatial sense. All adjustments are completed in the parameter domain before audio rendering and do not involve resampling, waveform interpolation, or model reconstruction. This application can, for example, determine the output loudness of the target sound scene based on volume envelope control. This application can also, for example, determine the auditory center of gravity position of the target sound scene based on frequency band energy distribution control. Furthermore, this application can also determine the temporal stability characteristics of the target sound scene based on the coordinated control of rhythm density and cycle period. This application obtains a target sound scene that matches the current risk level and is continuously connected to the current sound scene state based on any of the above methods.

[0100] Example 9: In another optional embodiment, this application also provides the progressive adjustment of the current sound scene in step S6 and the posterior determination and parameter update of micro-awakening events in step S7, including: Step 1: Within the target transition time, gradually adjust the output volume according to the preset gradation rate; The target transition duration (e.g., 5 seconds, 10 seconds, 20 seconds, 30 seconds, 60 seconds, or 90 seconds) can refer to the minimum time span set for completing parameter migration from the current sound scene to the target sound scene. This time span is used to constrain the continuity of the volume change process and prevent auditory startle reflex or cortical arousal caused by sudden volume changes. The preset gradual change rate (e.g., a linear or nonlinear change rate of 0.1 dB to 2.0 dB per second) can refer to the adjustment range of the volume parameter per unit time, and its value is constrained by the user's individual historical tolerance data. In this embodiment, the preset gradual change rate is determined based on the maximum volume change slope that did not induce micro-arousals in the user's past intervention records, and a conservative rate is adopted when the risk level increases, while a slightly higher rate is allowed when the risk level decreases to accelerate recovery. This application can determine the output volume change curve over time by means of continuous interpolation of the volume envelope; it can also determine the phased step value of the output volume within the target transition time by means of piecewise linear mapping; further, it can determine the adjustment rhythm of the output volume by means of dynamically adjusting the gradual rate based on the user's real-time heart rate variability feedback. This application obtains a smooth and controllable volume transition process based on any of the above methods to maintain the stability of the auditory channel and avoid interference with sleep homeostasis due to instantaneous sound pressure changes.

[0101] Step 2: Gradually adjust the energy proportion of different frequency bands according to the preset ratio; Here, different frequency bands (e.g., low frequency 0.1–100 Hz, mid frequency 100–1000 Hz, high frequency 1000–8000 Hz) can refer to acoustic energy distribution ranges divided according to the characteristics of human auditory perception and the sensitivity of neural response during sleep; preset ratios (e.g., each adjustment increases the proportion of low frequency energy by 5%, decreases mid frequency by 3%, and decreases high frequency by 2%, or redistributes it in a progressive manner with a fixed step size) can refer to the relative change in the energy weight of each frequency band within a single parameter update cycle; this ratio setting takes into account both masking effectiveness and physiological compatibility, ensuring that the enhancement of low frequency components provides stable background noise support, while suppressing high frequency burst components to reduce the risk of cortical excitation; This application can, for example, determine the dynamic ratio of energy in each frequency band within the target transition time using a spectral weighted reconstruction method; it can also determine the phased adjustment path of gain in different frequency bands using a method of progressively updating filter bank coefficients; further, it can determine the frequency band energy redistribution strategy by adaptively biasing the center of gravity of intermediate frequency energy based on the drift direction of the snoring dominant frequency. Based on any of the above methods, this application obtains a spectral structure evolution process that meets the needs of sleep maintenance, thereby achieving flexible adaptation to the acoustic environment.

[0102] Step 3: Gradually adjust the rhythm density according to the preset step length; Among them, rhythm density (e.g., the number of times a rhythm unit appears per minute, where a rhythm unit includes pulse-type rhythm points, periodic amplitude modulation peaks, and harmonic cluster repetition intervals) can refer to a quantitative indicator characterizing the strength of the temporal dimension structure in a sound scene; preset step size (e.g., ±0.2 times / minute, ±0.5 times / minute, or ±1.0 times / minute for each adjustment) can refer to the allowable variation in rhythm density for a single parameter update; this step size setting reflects the principle of prioritizing low disturbances—reducing rhythm density during the phase of increased micro-arousal risk to weaken the auditory attention capture effect, and slowly increasing it during the phase of decreased risk to support the reconstruction of the natural sleep rhythm; This application can, for example, determine the gradual change value of rhythm density within the target transition duration by iteratively updating the rhythm template scaling factor; it can also determine the decreasing path of rhythm perception intensity by gradually attenuating the temporal envelope modulation depth; further, it can determine the rhythm parameter adjustment logic by adjusting the direction of rhythm density change in the opposite direction based on the trend of the centroid shift in the body motion signal spectrum. This application obtains a rhythm control process that co-evolves with the user's motion state based on any of the above methods, thereby avoiding conflicts between rhythmic stimulation and autonomic nervous activity.

[0103] Step 4: Perform a crossfade transition between the current sound scene and the target sound scene; Among them, cross-fade-in and cross-fade-out transitions (such as playing the current sound scene in the left channel with exponential decay and playing the target sound scene in the right channel with exponential growth, or smooth switching of dual-channel mixed weights according to an S-shaped function) can refer to achieving a seamless connection between two sound scenes in terms of auditory perception through multi-channel phase coordination or mono-channel time-domain weighted superposition; this transition method does not rely on absolute silence gaps, avoiding alarm reactions caused by sound interruption; its switching function parameters (such as exponential decay / growth with a time constant τ=5 s, or a Sigmoid function with a step width of 10 s) are jointly constrained by the target transition duration and the user's previous sound therapy tolerance records; This application can, for example, determine the mixed output of the current and target sound scenes based on dual-channel amplitude envelope cross-modulation; it can also determine the independent transition rhythm of different frequency bands based on frequency band selective fade-in / fade-out guided by a time-frequency masking model; further, it can determine the transition triggering conditions by dynamically adjusting the fade-in / fade-out start timing based on the slope of blood oxygen saturation change. Based on any of the above methods, this application achieves a seamless, uninterrupted, and abrupt sound scene transition process, thereby eliminating the risk of secondary disturbances caused by traditional switching modes.

[0104] Step 5: In step S7, determine whether a micro-arousal event has occurred based on whether the user exhibits at least one of the following: increased body movement, sudden changes in respiratory rhythm, abnormal increase in heart rate, abnormal changes in snoring spectrum, or decreased sleep stability. Among them, increased body movement (e.g., body movement energy increases by more than 200% from the baseline, or body movement frequency reaches more than 3 times within 3 seconds), abrupt changes in respiratory rhythm (e.g., the standard deviation of the respiratory cycle increases by 150% from the mean of the previous 5 minutes, or the inspiratory-to-expiratory ratio deviates from the normal range by ±30%), abnormal increase in heart rate (e.g., heart rate increases by ≥15 bpm from the baseline value during sleep and lasts for ≥5 seconds), abnormal changes in snoring spectrum (e.g., snoring dominant frequency drift >15 Hz and a sudden increase in the proportion of high-frequency energy ≥40%), and decreased sleep stability (e.g., the sleep stability score calculated by fusing the four-dimensional features of respiration, heart rate, body movement, and snoring decreases by ≥0.3 within 10 seconds) are all objective proxy indicators for the occurrence of micro-arousals; the determination of a micro-arousal event is triggered if any of the above indicators are met, but not all of them need to be met; the determination logic adopts an OR gate fusion structure to ensure sensitivity. This application can determine whether a micro-arousal event has occurred, for example, by using a multi-source physiological signal combined threshold criterion; it can also determine whether a micro-arousal event has occurred by using the statistical significance of feature mutations within a short sliding window; furthermore, it can determine whether a micro-arousal event has occurred by using EEG equivalent surrogate indicators (such as transcranial near-infrared blood oxygen fluctuation patterns) for verification. Based on any of the above methods, this application obtains a highly sensitive, low-false-negative posterior micro-arousal recognition capability, providing a reliable basis for closed-loop correction.

[0105] Step Six: When a micro-awakening event is determined to occur within the prediction window, increase the weight of the abnormal feature corresponding to the current micro-awakening event, and / or increase the entry threshold of the corresponding risk level; Among them, the abnormal features corresponding to the current micro-awakening event (such as the actual exceeding threshold of body kinetic energy, respiratory cycle variability, heart rate increase, snoring frequency drift, or sleep stability score in this judgment) can refer to the specific feature items that were activated and contributed to the judgment result; increasing its weight (such as increasing its original weight from 0.15 to 0.22 in subsequent risk value calculation) can refer to enhancing the contribution of this feature in the weighted fusion analysis; increasing the entry threshold of the corresponding risk level (such as increasing the entry threshold of the medium-high risk level from 0.60 to 0.65) can refer to raising the minimum risk value required to trigger intervention at this level, so as to avoid over-intervention due to false alarms of such precursor signals; this adjustment operation only applies to the features involved in this judgment and their associated risk levels, and does not affect other features or other levels; This application can determine the update amount of abnormal feature weights by back-calculating the correction based on feature contribution; it can also determine the direction and magnitude of the entry threshold adjustment by statistical optimization based on the threshold error rate; further, it can determine the threshold update strategy by dynamically constraining the upper limit of a single threshold adjustment based on the cumulative false alarm rate over multiple nights. Based on any of the above methods, this application obtains parameter adaptive capability oriented towards individual precursor sensitivity, making the system more responsive to real disturbances and better suppressing artifact disturbances.

[0106] Step 7: When no micro-awakening event occurs within the prediction window, maintain the current scene control parameter group unchanged, or lower the entry threshold of the corresponding risk level to form subsequent micro-awakening risk value assessment parameters and sound scene control parameters for individual users.

[0107] Maintaining the current set of control parameters unchanged (e.g., keeping the target volume range, target frequency band energy distribution, target rhythm density, etc., unchanged) can mean confirming that the intervention did not induce any new disturbances, and the system will not adjust the existing configuration for the time being; lowering the entry threshold for the corresponding risk level (e.g., lowering the entry threshold for medium-high risk level from 0.60 to 0.57) can mean appropriately relaxing the triggering conditions for this level to improve the sensitivity of early intervention; the premise of this operation is that the system has accumulated at least 3 predictions of the same level that did not occur, and the most recent event that did not occur was no more than 72 hours ago; all adjustments are written into the user's individual profile module and participate in subsequent risk value calculation and scene mapping; This application can determine the magnitude of the threshold reduction using methods such as Bayesian confidence updates; it can also determine whether to maintain the parameter set unchanged based on the output of a parameter stability assessment model; further, it can determine the threshold update rhythm by predicting the timing of the next adjustment based on overnight parameter drift trends. Based on any of the above methods, this application obtains a robust, gradual, and traceable individualized parameter evolution path, supporting the continuous improvement of long-term intervention effects.

[0108] Figure 2 This is a schematic diagram of the structure of a progressive sound therapy scene generation system for sleep disorder intervention provided in one embodiment of this application, as shown below. Figure 2 As shown, the progressive sound therapy scene generation system 300 for sleep disorder intervention in this embodiment includes: a preprocessing module 301, a fusion analysis module 302, a control analysis module 303, a risk analysis module 304, a scene analysis module 305, and a parameter iteration module 306.

[0109] The preprocessing module 301 is used to acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to a preset sliding time window. The fusion analysis module 302 is used to perform weighted fusion analysis on the sleep stability characteristic parameters and the micro-awakening precursor characteristic parameters based on a preset short-term analysis window and a preset trend analysis window, so as to determine the micro-awakening risk value of micro-awakening events occurring within a preset prediction window in the future; The control analysis module 303 is used to determine the corresponding risk level based on the micro-awakening risk value, and to determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level. The risk analysis module 304 is used to determine whether to enter or exit the risk level based on the micro-awakening risk value and the entry and exit thresholds of the corresponding risk level. The scene analysis module 305 is used to progressively adjust the current sound scene according to the target sound scene control parameters after confirming that the risk level has been entered, so as to generate the target sound scene. The parameter iteration module 306 is used to determine whether a micro-awakening event has occurred based on the user's actual sleep state after the prediction window ends, and to update the risk judgment threshold, feature weights and / or scenario control parameter group used for subsequent micro-awakening risk value assessment based on the determination result.

[0110] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating a progressive sound therapy scenario for sleep disorder intervention, characterized in that, include: S1. Acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to a preset sliding time window; S2. Based on a preset short-term analysis window and a preset trend analysis window, a weighted fusion analysis is performed on the sleep stability characteristic parameters and the micro-awakening precursor characteristic parameters to determine the micro-awakening risk value of micro-awakening events occurring within a preset prediction window in the future. S3. Determine the corresponding risk level based on the micro-awakening risk value, and determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level; S4. Based on the micro-awakening risk value and the corresponding risk level entry and exit thresholds, determine whether to enter or exit the risk level; S5. After confirming that the risk level has been entered, the current sound scene is progressively adjusted according to the target sound scene control parameters to generate the target sound scene. S6. After the prediction window ends, determine whether a micro-awakening event has occurred based on the user's actual sleep state, and update the risk judgment threshold, feature weights and / or scenario control parameter group used for subsequent micro-awakening risk value assessment based on the determination result.

2. The method of claim 1, wherein, The multimodal sleep monitoring data includes at least two of the following: respiratory signals, heart rate or heart rate variability signals, body movement signals, snoring or breathing sound signals, and blood oxygenation signals; The preprocessing in S2 includes timestamp alignment, noise reduction, outlier removal, and data quality scoring. The data quality score is determined based on the signal integrity, continuity, signal-to-noise level, and temporal validity of various types of multimodal sleep monitoring data; For multimodal sleep monitoring data with a data quality score below a preset threshold, they are removed, or the feature parameters extracted from them are downweighted.

3. The method of claim 2, wherein, Step S2 involves extracting the sleep stability feature parameters, including: Based on the respiratory signals, the respiratory cycle variability and respiratory amplitude fluctuation rate are extracted; Based on the heart rate or heart rate variability signal, extract the amplitude of heart rate changes and the degree of heart rate fluctuation; Based on the body movement signals, the body movement frequency and body movement energy are extracted; Extract the main frequency drift and spectral structure change based on the snoring or breathing sound signal; And at least two of the above-mentioned characteristic parameters are used as the sleep stability characteristic parameters.

4. The method of claim 3, wherein, The micro-awakening precursor feature parameters extracted in S2 include: Calculate the difference values ​​of each of the sleep stability characteristic parameters between adjacent sliding time windows; Calculate the slope of change of each of the sleep stability characteristic parameters between multiple consecutive sliding time windows; Determine the degree of synchronous abnormality when different types of sleep stability characteristic parameters simultaneously exceed the corresponding abnormality threshold within the same sliding time window; Based on the difference value, the slope of change, and the degree of synchronization anomaly, the micro-awakening precursor characteristic parameters are generated.

5. The method of claim 4, wherein, Step S3 specifically includes: The micro-awakening precursor characteristic parameters are aggregated according to the short-term analysis window to obtain short-term fluctuation characteristic parameters; The sleep stability characteristic parameters are aggregated according to the trend analysis window to obtain the trend change characteristic parameters. The weighting coefficients are determined based on the data quality scores corresponding to each feature parameter; The short-term fluctuation characteristic parameters and the trend change characteristic parameters are weighted and fused based on the weighting coefficients. The micro-awakening risk value is determined based on the weighted fusion result and the preset risk judgment rules.

6. The method of claim 5, wherein, In step S4, the risk level is divided into at least three levels, including low risk, medium risk and high risk; Pre-set scenario control parameter groups for different risk levels; Each of the scenario control parameter groups includes at least two of the following: target volume range, target frequency band energy distribution, target rhythm density, target transition duration, and target hold duration.

7. The method of claim 6, wherein, In step S5, when the micro-awakening risk value is higher than the entry threshold of the corresponding risk level for at least two consecutive sliding time windows, it is confirmed that the risk level has been entered. When the micro-awakening risk value is lower than the exit threshold of the corresponding risk level for at least two consecutive sliding time windows, it is confirmed that the risk level will be exited. The entry threshold is higher than the exit threshold to suppress frequent switching of the sound scene caused by short-term fluctuations in the micro-awakening risk value.

8. The method of claim 7, wherein, Step S6 generates the target sound scene, including: The corresponding scenario control parameter group is invoked based on the current risk level; Based on the current output state of the sound scene, the parameter combination with the smallest difference from the current sound scene is determined from the scene control parameter group as the target sound scene control parameter; Retrieve the scene template corresponding to the current risk level from the preset sound scene library; The scene template is then parametrically adjusted based on the target sound scene control parameters to generate the target sound scene. The preset sound scene library includes one or more of the following: white noise scene, pink noise scene, brown noise scene, natural sound scene, low dynamic range ambient sound scene, and composite masking scene.

9. The method of claim 8, wherein, Step S6 involves progressively adjusting the current sound scene, including: Within the target transition time, the output volume is gradually adjusted according to the preset gradation rate; The energy proportion of different frequency bands is gradually adjusted according to the preset ratio; The rhythm density is gradually adjusted according to the preset step size; Perform a cross-fade-in / fade-out transition between the current sound scene and the target sound scene; In step S7, it is determined whether a micro-arousal event has occurred based on whether the user exhibits at least one of the following: increased body movement, sudden changes in respiratory rhythm, abnormal increase in heart rate, abnormal changes in snoring spectrum, or decreased sleep stability. When a micro-awakening event is determined to occur within the prediction window, the weight of the abnormal feature corresponding to the current micro-awakening event is increased, and / or the entry threshold for the corresponding risk level is increased. When it is determined that no micro-awakening event has occurred within the prediction window, the current scene control parameter group is kept unchanged, or the entry threshold of the corresponding risk level is lowered, so as to form subsequent micro-awakening risk value assessment parameters and sound scene control parameters for individual users.

10. A progressive sound therapy scene generation system for sleep disorder intervention, characterized in that, The method applied to any one of claims 1-9 includes: The preprocessing module is used to acquire multimodal sleep monitoring data during the user's sleep process, preprocess the multimodal sleep monitoring data, and extract sleep stability feature parameters and micro-awakening precursor feature parameters according to a preset sliding time window. The fusion analysis module is used to perform weighted fusion analysis on the sleep stability characteristic parameters and the micro-arousal precursor characteristic parameters based on a preset short-term analysis window and a preset trend analysis window, so as to determine the micro-arousal risk value of micro-arousal events occurring within a preset prediction window in the future; The control and analysis module is used to determine the corresponding risk level based on the micro-awakening risk value, and to determine the target sound scene control parameters from the preset scene control parameter group corresponding to the risk level. The risk analysis module is used to determine whether to enter or exit the risk level based on the micro-awakening risk value and the corresponding entry and exit thresholds of the risk level. The scene analysis module is used to progressively adjust the current sound scene according to the target sound scene control parameters after confirming that the risk level has been entered, so as to generate the target sound scene. The parameter iteration module is used to determine whether a micro-awakening event has occurred based on the user's actual sleep state after the prediction window ends, and to update the risk judgment threshold, feature weights and / or scenario control parameter group used for subsequent micro-awakening risk value assessment based on the determination result.

Images