Snoring sound recognition intervention system and method

Abstract

The application relates to a snoring sound identification intervention system and method, wherein the snoring sound identification intervention system comprises a data acquisition module, a snoring sound identification module and a snoring sound intervention module; the data acquisition module is used for acquiring heart impact snoring sound vibration data of a user; the snoring sound identification module is used for generating a corresponding snoring sound feature vector based on the heart impact snoring sound vibration data, inputting the snoring sound feature vector into a trained snoring sound classification model, and outputting a snoring sound judgment result and a snoring sound intensity level; the snoring sound intervention module is used for generating a corresponding intervention scheme based on the snoring sound intensity level when the snoring sound judgment result is that snoring sound is identified, and adjusting an intelligent pillow used by the user based on the intervention scheme. Through the application, the problem of high snoring sound misjudgment rate is solved.

Description

Technical Field

[0001] This application relates to the field of intelligent sleep health device technology, and in particular to snoring recognition and intervention systems and methods. Background Technology

[0002] As people pay increasing attention to sleep health, snoring, as a key factor affecting sleep quality, has made its recognition and intervention technology a core function of smart sleep devices. Snoring originates from airflow vibrations caused by narrowing or obstruction of the airway during sleep. It not only disturbs others' rest, but long-term heavy snoring may also induce health risks such as sleep apnea. Among related technologies, methods using audio recognition and manual intervention suffer from a high rate of false positives for snoring.

[0003] Currently, no effective solution has been proposed to address the high rate of false positives for snoring in related technologies. Summary of the Invention

[0004] This application provides a snoring recognition intervention system and method to at least solve the problem of high snoring misjudgment rate in related technologies.

[0005] In a first aspect, embodiments of this application provide a sound recognition intervention system, the system comprising: a data acquisition module, a snoring recognition module, and a snoring intervention module;

[0006] The data acquisition module is used to collect the user's cardiac impact and snoring vibration data;

[0007] The snoring recognition module is used to generate a corresponding snoring feature vector based on the cardiac impact snoring vibration data, and input the snoring feature vector into the trained snoring classification model to output the snoring judgment result and snoring intensity level.

[0008] The snoring intervention module is used to generate a corresponding intervention plan based on the snoring intensity level when the snoring judgment result indicates that snoring has been detected, and to adjust the smart pillow used by the user based on the intervention plan.

[0009] In some embodiments, the data acquisition module includes an audio sensing unit; the snoring recognition module includes a cardiac impact vibration component analysis unit and an audio wake-up control module;

[0010] The cardiac impact vibration component analysis unit is used to analyze the cardiac impact snoring vibration data, obtain a preliminary snoring judgment result, and trigger the audio wake-up control module when the preliminary snoring judgment result is suspected snoring.

[0011] The audio sensing unit is used to collect sound data in response to the triggering of the audio wake-up control module;

[0012] The snoring recognition module is also used to generate the snoring feature vector based on the cardiac impact snoring vibration data and the sound data.

[0013] In some embodiments, the snoring recognition module is further configured to analyze the cardiac impact snoring vibration data using a preset time-series period detection algorithm to identify the vibration signal period and determine the preliminary snoring judgment result based on the vibration signal period.

[0014] In some embodiments, the snoring recognition module is further configured to extract key audio features of the corresponding snoring based on the sound data, and extract corresponding time-frequency domain features based on the cardiac impact snoring vibration data.

[0015] The snoring recognition module is further configured to fuse the key audio features of snoring and the time-frequency domain features according to preset weights to generate the snoring feature vector.

[0016] In some embodiments, the snoring recognition module is further configured to extract corresponding initial key audio features of snoring based on the sound data, and reduce the dimensionality of the initial key audio features of snoring to a preset dimension using a principal component analysis algorithm to obtain the key audio features of snoring.

[0017] In some embodiments, the snoring recognition module is further configured to perform hardware noise reduction and digital conversion on the sound data to obtain standard audio data, and input the standard audio data into the Mel spectrogram architecture to generate a Mel spectrogram;

[0018] The snoring recognition module is also used to extract key audio features of the initial snoring based on the Mel spectrogram; the key audio features of the initial snoring include Mel frequency cepstral coefficients, spectral flatness, and fundamental frequency.

[0019] In some embodiments, the snoring recognition module is further configured to input the snoring feature vector into the snoring classification model to obtain the snoring judgment result;

[0020] The snoring recognition module is further configured to, when the snoring judgment result indicates that snoring has been detected, analyze the snoring feature vector using the decibel regression algorithm of the snoring classification model to obtain snoring intensity data, and determine the corresponding snoring intensity level based on the snoring intensity data.

[0021] In some embodiments, the data acquisition module is also used to acquire cardiac impact snoring vibration data after intervention;

[0022] The snoring recognition module is also used to obtain snoring intensity data after intervention based on the cardiac impact snoring vibration data after intervention;

[0023] The snoring intervention module is also used to dynamically adjust the intervention plan based on the snoring intensity data after the intervention and the snoring intensity data before the intervention;

[0024] The snoring intervention module is also used to optimize the intervention parameters corresponding to different intensities of snoring based on the user's historical comfort score and historical intervention parameters using a nonlinear fitting algorithm.

[0025] In some embodiments, the data acquisition module is further configured to acquire initial cardiac impact snoring vibration data and remove interference signals from the initial cardiac impact snoring vibration data using a preset adaptive bandpass filtering algorithm to obtain the cardiac impact snoring vibration data.

[0026] Secondly, embodiments of this application provide a snoring recognition and intervention method, the method comprising:

[0027] Collect the user's cardiac impact snoring vibration data; generate the corresponding snoring feature vector based on the cardiac impact snoring vibration data;

[0028] The snoring feature vector is input into the trained snoring classification model, which outputs the snoring judgment result and the snoring intensity level.

[0029] When the snoring sound is detected, a corresponding intervention plan is generated based on the snoring sound intensity, and the smart pillow used by the user is adjusted based on the intervention plan.

[0030] Compared to related technologies, the snoring recognition and intervention system and method provided in this application include: a data acquisition module, a snoring recognition module, and a snoring intervention module. The data acquisition module collects cardiac impact snoring vibration data from the user. The snoring recognition module generates corresponding snoring feature vectors based on the cardiac impact snoring vibration data, inputs the snoring feature vectors into a trained snoring classification model, and outputs a snoring judgment result and a snoring intensity level. The snoring intervention module generates a corresponding intervention plan based on the snoring intensity level when the snoring judgment result indicates snoring has been detected, and adjusts the smart pillow used by the user based on the intervention plan. This solves the problem of high snoring misjudgment rate.

[0031] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0032] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0033] Figure 1 This is a structural block diagram of a snoring recognition and intervention system according to an embodiment of this application;

[0034] Figure 2 This is a flowchart of a snoring recognition intervention method according to an embodiment of this application. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application. Furthermore, it is understood that although the efforts made in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, modifications to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0036] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0037] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application means two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The terms “first,” “second,” “third,” etc., used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0038] With increasing public awareness of sleep health, snoring, a significant factor affecting sleep quality, has made its recognition and intervention a core function of smart sleep devices. Snoring originates from airway narrowing or obstruction during sleep, causing airflow vibrations. It not only disturbs others' rest but, in the long term, heavy snoring can also lead to health risks such as sleep apnea. Currently, snoring-related smart devices on the market mainly revolve around two core functions: "snoring recognition" and "intervention and regulation," with applications covering areas such as home sleep monitoring and health management. Users have increasingly urgent needs for accurate recognition, effective intervention, and sleep-friendly features. Existing technological solutions suffer from the following core problems, making it difficult to meet users' comprehensive needs for "accurate recognition + efficient intervention + sleep-friendly features":

[0039] First, snoring recognition has low accuracy, weak anti-interference capabilities, and poor adaptability: single audio recognition is easily affected by environmental noise, resulting in a high false judgment rate; multi-sensor schemes have not formed an effective feature fusion mechanism, failing to fully leverage the complementary advantages of different sensor data; complex algorithms consume a lot of computing power, making them difficult to adapt to low-computing-power embedded devices; and they have not been fine-tuned based on individual user snoring characteristics, resulting in insufficient adaptability to different users.

[0040] Secondly, the intervention programs are not targeted enough and lack sleep-friendliness: the overall adjustment cannot accurately adapt to the narrow airway, and the intervention effect is limited; the programs are not differentiated according to the snoring intensity level, resulting in over-intervention for mild snoring or under-intervention for severe snoring; the adjustment range and speed are unreasonable, the awakening rate is high, and it interferes with the user's sleep continuity.

[0041] Finally, there is a lack of a closed-loop optimization mechanism: the effects of the intervention are not monitored and feedback is not monitored in real time, and the strategy cannot be dynamically adjusted according to changes in snoring after the intervention; historical intervention data of users is not accumulated, and the program cannot be optimized through long-term learning, making it difficult to continuously improve the intervention effect.

[0042] To address the aforementioned problems, this embodiment provides a snoring recognition and intervention system. Figure 1 This is a structural block diagram of a snoring recognition and intervention system according to an embodiment of this application, such as... Figure 1 As shown, the system includes: a data acquisition module 11, a snoring recognition module 12, and a snoring intervention module 13;

[0043] Data acquisition module 11 is used to collect the user's cardiac impact and snoring vibration data;

[0044] The snoring recognition module 12 is used to generate corresponding snoring feature vectors based on cardiac impact snoring vibration data, and input the snoring feature vectors into the trained snoring classification model to output snoring judgment results and snoring intensity levels.

[0045] The snoring intervention module 13 is used to generate a corresponding intervention plan based on the snoring intensity level when the snoring judgment result is that snoring is detected, and to adjust the smart pillow used by the user based on the intervention plan.

[0046] The data acquisition module 11 is designed to collect data on cardiac impact and snoring vibrations during the user's sleep. This module is equipped with a high-precision cardiac impact sensing unit. By deploying a Ballistocardiography Sensor Array (BCG sensor array) within the core support area of ​​the smart pillow, it accurately captures the head and neck vibration components caused by snoring. Simultaneously, it incorporates a low-noise signal amplification module and an adaptive bandpass filtering algorithm (e.g., 0.1Hz-50Hz) to effectively eliminate irrelevant interference signals such as environmental vibrations and bed shaking, ensuring high fidelity of the collected cardiac impact and snoring vibration data and providing reliable raw data support for subsequent snoring recognition.

[0047] The snoring recognition module 12 is responsible for converting cardiac impact snoring vibration data into effective decisions. After acquiring the cardiac impact snoring vibration data transmitted by the data acquisition module 11, this module extracts the time-frequency domain features of the vibration data and generates corresponding snoring feature vectors. To ensure recognition accuracy and efficiency, this module has a built-in trained lightweight snoring classification model. After inputting the generated feature vectors into the model, it can not only quickly output snoring judgment results, but also analyze snoring intensity data through a decibel regression algorithm, ultimately determining different levels of snoring intensity (e.g., Level 1: 15dB-30dB, mild snoring; Level 2: 30dB-50dB, moderate snoring; Level 3: ≥50dB, heavy snoring), achieving accurate recognition and classification of snoring. The snoring classification model can be a lightweight support vector machine classification model (lightweight SVM classification model).

[0048] The snoring intervention module 13 is responsible for converting the recognition results into targeted intervention actions. Upon receiving the snoring recognition module 12's output, which indicates the snoring has been detected and its corresponding intensity level, this module generates a differentiated intervention plan based on the snoring intensity level and the user's historical intervention effect data (such as the optimal adjustment parameters for a certain intensity of snoring). The intervention strategies exhibit significant individual differences for different intensities of snoring. For example, for level 1 mild snoring, only a small adjustment is made to the corresponding support area under the head, subtly altering the airway angle through a slight height change to alleviate mild snoring. For level 2 moderate snoring, the adjustment range is expanded to the head and neck support areas, with a larger adjustment amplitude, simultaneously adapting to the head and neck support height to further optimize airway patency. For level 3 severe snoring, multi-area coordinated adjustment is initiated, adjusting to the maximum extent by raising the head pillow height, lowering the neck height, raising the left side of the head, and lowering the right side, forcing the user to sleep on their right side (if the user prefers to sleep on their left side, the left side is lowered and the right side is raised), thereby effectively alleviating severe snoring. This module precisely executes adjustment actions by driving a high-precision stepper motor and multi-stage telescopic rods inside the smart pillow. While ensuring a ≥80% reduction rate for medium-to-high intensity snoring, it keeps the user wake-up rate below 5%, balancing the intervention effect with sleep-friendliness.

[0049] It should be noted that the smart pillow has multiple independent height adjustment mechanisms inside. According to the head and neck support needs, the pillow surface is divided into multiple independent adjustment units such as upper, middle and lower, left, middle and right. Each unit is equipped with a high-precision stepper motor and multi-stage telescopic rods, with adjustment accuracy ≤1mm, adjustment range ≥5cm (covering different user support needs), and minimum adjustment speed ≤1mm / s (avoiding discomfort caused by rapid adjustment).

[0050] The aforementioned modules achieve data exchange and collaborative operation through a standardized communication protocol. The data acquisition module 11 and the snoring recognition module 12 transmit acquired data via the SPI protocol (Serial Peripheral Interface), while the snoring recognition module 12 and the snoring intervention module 13 interact via the UART protocol (Universal Asynchronous Receiver / Transmitter), exchanging recognition results and control commands to ensure the real-time performance and stability of the entire system. Furthermore, the system possesses dynamic feedback and optimization capabilities. The snoring intervention module 13 continuously monitors changes in snoring intensity after intervention through the data acquisition module 11, dynamically adjusting intervention parameters based on the effect; simultaneously, it accumulates historical user intervention data and comfort scores, continuously optimizing intervention strategies corresponding to different snoring intensities through a nonlinear fitting algorithm.

[0051] In addition, the system may include a user interaction module, which allows users to easily interact with the system via a mobile app or mini-program. Specific functions include: users can input basic personal information such as age, height, and weight, as well as information on snoring levels (no, mild, moderate, severe), providing a reference for personalized intervention strategies; users can view real-time data on snoring intensity and intervention records (including adjustment time, adjustment parameters, and intervention effects) during sleep, intuitively understanding the dynamics of sleep and intervention; after waking up, users can rate the intervention effect and sleep comfort on a scale of 1 to 5. This rating data will be transmitted to the snoring intervention module as historical feedback data to help optimize subsequent intervention strategies. If the user does not actively rate the system, the system will automatically generate a comfort score based on a sleep quality assessment model (combining sleep duration, number of body movements, etc.), ensuring the completeness of feedback data and the continuity of the optimization process.

[0052] The snoring recognition and intervention system provided in the above embodiments collects cardiac impact snoring vibration data through a data acquisition module, and analyzes and processes the snoring feature vectors using a trained snoring classification model through a snoring recognition module. This not only significantly improves the accuracy of snoring recognition and effectively reduces the false judgment rate of traditional audio recognition methods, but also quickly outputs snoring judgment results and multi-level intensity levels, providing a precise basis for targeted intervention. At the same time, the snoring intervention module generates an appropriate intervention plan based on the snoring intensity level and adjusts the smart pillow, overcoming the shortcomings of insufficient intervention targeting and easy sleep disturbance in existing technologies. It effectively alleviates snoring while taking into account sleep friendliness, successfully solving the core problems of low snoring recognition accuracy and poor intervention effect in related technologies, and providing more reliable and efficient technical support for intelligent sleep health monitoring and snoring intervention.

[0053] In some embodiments, the data acquisition module includes an audio sensing unit; the snoring recognition module includes a cardiac impact vibration component analysis unit and an audio wake-up control module;

[0054] The cardiac impact vibration component analysis unit is used to analyze cardiac impact snoring vibration data, obtain preliminary snoring judgment results, and trigger the audio wake-up control module when the preliminary snoring judgment result is suspected snoring.

[0055] An audio sensing unit is used to collect sound data in response to the triggering of the audio wake-up control module;

[0056] The snoring recognition module is also used to generate snoring feature vectors based on cardiac impact snoring vibration data and sound data.

[0057] The aforementioned data acquisition module includes not only a cardiac impact sensing unit but also an audio sensing unit. The audio sensing unit complements the cardiac impact and snoring vibration data acquisition function, constructing a dual-modal data acquisition system that provides richer raw data support for subsequent accurate identification. The aforementioned snoring recognition module further includes a cardiac impact vibration component analysis unit and an audio wake-up control module.

[0058] The cardiac impact vibration component analysis unit, acting as a pre-screening device for snoring recognition, plays a crucial role in initial judgment and triggering wake-up. After receiving cardiac impact snoring vibration data from the data acquisition module, this unit performs in-depth analysis using a preset time-series periodic detection algorithm. It accurately identifies the periodic characteristics of the vibration signal and, combined with preset discrimination thresholds (such as an energy percentage greater than 20% in the 10Hz–30Hz frequency band and an amplitude standard deviation greater than 1.5 times the normal state), determines whether the current vibration is a suspected snoring vibration, thus forming a preliminary snoring judgment result. When the preliminary judgment result indicates suspected snoring, the unit immediately triggers the audio wake-up control module, initiating the subsequent audio data acquisition process. This design of screening before wake-up effectively avoids ineffective operation of the audio sensing unit.

[0059] The audio sensing unit employs a low-power sleep mode and an on-demand wake-up mode. By default, it operates in low-power sleep mode (power consumption < 5mA) to reduce overall system energy consumption and extend device battery life. Upon receiving a trigger signal from the audio wake-up control module, the unit responds quickly (wake-up response time < 100ms), accurately collecting sound data for suspected snoring periods. The collection duration is adapted to the snoring cycle (e.g., 3-8 seconds per slice) to ensure the integrity of the sound data. After collection, the audio sensing unit preprocesses the sound data using hardware noise reduction circuitry to filter out ambient noise, then digitizes it into standard audio data, ensuring a signal-to-noise ratio ≥ 35dB. It then quickly enters sleep mode, awaiting the next wake-up command.

[0060] After acquiring cardiac impact snoring vibration data and preprocessed sound data, the snoring recognition module initiates a dual-modal feature fusion process. On one hand, it extracts corresponding time-frequency domain features from the cardiac impact snoring vibration data to capture the physical vibration changes during snoring; on the other hand, it extracts audio features from the sound data. Then, the module deeply fuses the two types of features according to a preset weighting rule (e.g., both cardiac impact vibration features and audio features have a weight of 0.5), generating a snoring feature vector that combines physical vibration and acoustic attributes, providing high-precision feature support for subsequent snoring judgment and intensity classification.

[0061] The above embodiments, by adding an audio sensing unit to the data acquisition module and setting up a cardiac impact vibration component analysis unit and an audio wake-up control module in the snoring recognition module, construct an identification system that first performs vibration pre-judgment, then performs precise audio acquisition, and finally achieves dual-modal feature fusion, realizing accurate identification throughout the entire process from signal screening to feature fusion. Based on the in-depth analysis of cardiac impact snoring vibration data, the cardiac impact vibration component analysis unit can accurately screen suspected snoring signals and trigger audio wake-up, avoiding the ineffective operation of the audio sensing unit and significantly reducing system energy consumption. After responding to the wake-up, the audio sensing unit selectively collects sound data, combines it with cardiac impact snoring vibration data to form a dual-modal data source, and then fuses it through the snoring recognition module to generate a feature vector, effectively eliminating ambient noise without vibration and interference vibrations without acoustic features, significantly improving the anti-interference ability and accuracy of snoring recognition, while solving the problem of high misjudgment rate in traditional single-sensor recognition, providing highly reliable feature support for subsequent snoring intensity classification and targeted intervention, and adapting to the low-power operation requirements of low-computing-power embedded devices.

[0062] In some embodiments, the snoring recognition module is also used to analyze cardiac impact snoring vibration data through a preset time-series period detection algorithm to identify the vibration signal period and determine a preliminary snoring judgment result based on the vibration signal period.

[0063] The core logic of the temporal periodicity detection algorithm lies in the multi-dimensional analysis of the temporal characteristics of cardiac impact snoring vibration data. The algorithm first segments the preprocessed vibration data, sets a reasonable time window based on the physiological characteristics of snoring, and then extracts the periodic parameters of the vibration signal in each segment through techniques such as autocorrelation analysis and peak detection. These parameters include key information such as period duration and peak interval stability. For example, the algorithm identifies recurring peak features in the vibration signal, calculates the time interval between adjacent peaks, and thus determines the periodicity of the signal. Simultaneously, it filters out irregular peaks caused by random interference, ensuring the reliability of periodicity identification.

[0064] After identifying the vibration signal cycle, the snoring recognition module compares the extracted cycle parameters with a preset snoring cycle threshold range to determine the initial snoring judgment result. Based on the physiological mechanism of snoring, the preset threshold focuses on the common cycle range of snoring vibrations while allowing for a reasonable fluctuation range to account for individual differences. When the identified vibration signal cycle falls within the preset threshold range and the cycle stability meets the set conditions (e.g., the duration fluctuation amplitude of multiple consecutive cycles is less than a preset proportion), the module determines the vibration as a suspected snoring vibration and generates a preliminary judgment result of "suspected snoring." If the vibration signal has no clear cycle characteristics, or the cycle parameters exceed the preset threshold range, it is determined to be non-snoring interference and the subsequent audio wake-up process is not triggered. This achieves accurate screening of suspected snoring, ensuring the targeting of subsequent identification and avoiding energy waste caused by ineffective wake-ups.

[0065] The above embodiments introduce a preset time-series periodic detection algorithm into the snoring recognition module to conduct targeted analysis of cardiac impact snoring vibration data, accurately capture the periodic characteristics of the vibration signal, and fully utilize the regular periodic attributes of snoring vibration to effectively remove non-snoring signals such as environmental vibrations and random interference without periodic characteristics, significantly improving the accuracy of the preliminary snoring judgment results. At the same time, the design of determining the preliminary snoring judgment result based on the vibration signal period can accurately screen out suspected snoring signals, providing a scientific basis for the on-demand wake-up of the audio sensing unit, avoiding energy waste caused by the ineffective operation of the audio sensing unit. This not only ensures the targeting and anti-interference capability of snoring recognition, but also lays a high-quality pre-screening foundation for dual-modal feature fusion, further improving the overall accuracy of snoring recognition and the low power consumption characteristics of the system operation.

[0066] In some embodiments, the snoring recognition module is also used to extract key audio features of the corresponding snoring based on sound data, and to extract corresponding time-frequency domain features based on cardiac impact snoring vibration data.

[0067] The snoring recognition module is also used to fuse key audio features and time-frequency domain features of snoring according to preset weights to generate a snoring feature vector.

[0068] Specifically, for the sound data, the snoring recognition module first performs preprocessing optimization: hardware noise reduction and digital conversion are sequentially performed on the collected sound data to filter out irrelevant signals such as environmental noise and circuit interference, resulting in high-fidelity standard audio data; then, based on the standard audio data, key audio features corresponding to snoring are extracted. These features can comprehensively characterize the timbre, energy distribution, and pitch variation patterns of snoring. To reduce subsequent computational complexity, the module also performs dimensionality reduction processing on the audio features, ultimately obtaining concise and effective key audio features for snoring.

[0069] For cardiac impact snoring vibration data, the snoring recognition module focuses on the physical vibration characteristics it reflects, extracting the corresponding time-frequency domain features. Through time-series analysis and frequency-domain decomposition techniques, the module captures key information such as amplitude changes, frequency distribution, and energy proportions of the vibration signal at different time windows, focusing on mining vibration features related to snoring and accurately depicting the subtle vibration patterns of the head and neck during snoring. These time-frequency domain features can effectively distinguish snoring vibrations from environmental vibrations, user turning over, and other non-snoring vibrations, providing reliable physical feature support for snoring recognition.

[0070] After extracting the two types of features, the snoring recognition module performs feature fusion according to a preset weighting rule. For example, the weights of both the key audio features of snoring and the time-frequency domain features of cardiac impact vibration are set to 0.5, and this weight can be adjusted as needed. The module deeply fuses the two types of features through weighted summation, generating a snoring feature vector that combines acoustic and physical vibration attributes. This fused vector can comprehensively and accurately represent the core features of snoring, effectively eliminating ambient noise without vibration and interference vibrations without acoustic features, providing high-quality feature basis for subsequent input into the snoring classification model for judgment and grading.

[0071] In the above embodiments, the snoring recognition module extracts key features from sound data and cardiac impact snoring vibration data respectively and fuses them according to preset weights, which significantly improves the comprehensiveness and accuracy of the snoring feature vector: the key audio features of snoring extracted from sound data can accurately characterize the acoustic properties of snoring, and the time-frequency domain features extracted from cardiac impact snoring vibration data can effectively reflect the physical vibration characteristics of snoring. The complementary fusion of the two types of features successfully eliminates environmental noise without vibration and interference vibration without acoustic features, greatly reducing the misjudgment rate of traditional single feature recognition; at the same time, the reasonable allocation of preset weights ensures that the two types of features play a full role in the recognition process, and the generated snoring feature vector can more comprehensively and accurately characterize the essential features of snoring, providing high-quality data support for subsequent snoring judgment and intensity classification, and further strengthening the system's anti-interference ability and recognition reliability.

[0072] In some embodiments, the snoring recognition module is further configured to extract corresponding initial key audio features of snoring based on sound data, and reduce the dimensionality of the initial key audio features of snoring to a preset dimension through principal component analysis algorithm to obtain key audio features of snoring.

[0073] Specifically, in the initial stage of extracting key audio features of snoring, the snoring recognition module first preprocesses the collected sound data. First, it filters out irrelevant signals such as environmental noise and circuit interference through hardware noise reduction circuitry. Then, it converts the analog sound signal into standard audio data through digital conversion, ensuring high fidelity. Subsequently, based on the standard audio data, it extracts the initial key audio features of snoring. However, the initially extracted audio features have a high dimensionality. If directly used for subsequent calculations, it would increase the computational cost of the algorithm, making it difficult to adapt to the operating requirements of low-computing-power embedded devices such as STM32 (a 32-bit embedded microcontroller). Therefore, the snoring recognition module introduces Principal Component Analysis (PCA) to reduce the dimensionality of the initial key audio features of snoring. This algorithm analyzes the variance contribution of the feature vectors, selects the principal components that play a core role in snoring recognition, eliminates redundant information and noise interference, and compresses the initial high-dimensional features to a preset dimension (e.g., within 40 dimensions).

[0074] In the above embodiments, the dimensionality reduction process not only significantly reduces the computational complexity of subsequent feature fusion and model inference, but also reduces the system's computing power consumption and energy consumption, ensuring the stable operation of the system on low-computing-power devices; at the same time, it retains the core information in the audio features that is strongly correlated with snoring recognition, avoids the curse of dimensionality caused by excessively high feature dimensions, ensures the effectiveness and recognizability of key audio features of snoring, and lays a high-quality foundation for subsequent fusion with BCG vibration time-frequency domain features.

[0075] In some embodiments, the snoring recognition module is also used to perform hardware noise reduction and digital conversion on the sound data to obtain standard audio data, and input the standard audio data into the Mel spectrogram architecture to generate a Mel spectrogram;

[0076] The snoring recognition module is also used to extract key audio features of the initial snoring based on the Mel spectrogram; the key audio features of the initial snoring include Mel frequency cepstral coefficients, spectral flatness, and fundamental frequency.

[0077] Specifically, after the sound data is acquired, the snoring recognition module first initiates a preprocessing phase. The first step involves denoising the raw sound data using a hardware noise reduction circuit. This circuit is specifically designed to filter out common disturbances in sleep scenarios (such as air conditioning noise, ambient noise from outside the window, and noise generated by slight bed shaking), significantly reducing the intensity of irrelevant signals. Subsequently, a digital conversion is performed, transforming the denoised analog sound signal into standard audio data that can be processed by a computer, ensuring a consistent data format and meeting accuracy standards, thus providing high-quality input data for subsequent spectrogram generation.

[0078] The preprocessed standard audio data is then input into the Mel spectrogram architecture to generate the corresponding Mel spectrogram. The core advantage of the Mel spectrogram lies in its frequency scale's high degree of alignment with the perceptual characteristics of the human auditory system, enabling it to more accurately capture the differences in sensitivity of the human ear to different frequencies. Snoring, as a typical human physiological acoustic signal, has its key characteristics more clearly presented in the Mel spectrogram. Through this architecture, the standard audio data is transformed into a two-dimensional spectrogram that intuitively reflects the frequency distribution and energy changes of snoring, converting abstract sound signals into quantifiable visual feature carriers.

[0079] Based on the generated Mel spectrogram, the snoring recognition module further extracts key audio features of the initial snoring sound. These features comprehensively characterize the acoustic essence of snoring from different dimensions. Among them, Mel frequency cepstral coefficients (MFCCs), as a core feature characterizing sound timbre, can effectively capture the spectral envelope information of snoring. Typically, 12-16 dimensions are extracted to balance feature richness and computational efficiency. Spectral flatness describes the uniformity of energy distribution in the snoring spectrum and can distinguish snoring from irregular noise. Fundamental frequency focuses on the pitch characteristics of snoring. Combining the physiological mechanism of snoring, fundamental frequency data in the 50Hz-200Hz range is extracted to accurately reflect the pitch and variation patterns of snoring. These three types of features complement each other, together forming an initial feature set that can completely characterize the acoustic properties of snoring.

[0080] The above embodiments, through the snoring recognition module, sequentially perform hardware noise reduction, digital conversion, Mel spectrogram generation, and initial key audio feature extraction on the sound data, significantly improving the effectiveness and recognizability of snoring features: hardware noise reduction effectively filters out irrelevant signals such as environmental noise and circuit interference; digital conversion ensures the uniformity and accuracy of the data format; and the Mel spectrogram architecture's adaptability to human ear-sensitive frequencies allows the generated Mel spectrogram to clearly present the acoustic characteristics of snoring. Based on the three initial key audio features extracted from this spectrogram—Mel frequency cepstral coefficients, spectral flatness, and fundamental frequency—the core acoustic attributes of snoring are comprehensively characterized from multiple dimensions such as timbre, energy distribution, and pitch, effectively distinguishing snoring from irregular interference noise. This provides high-fidelity, high-recognition original feature support for subsequent feature reduction, fusion, and model recognition. At the same time, the standardized processing flow further ensures the stability and consistency of feature extraction, helping the system improve overall recognition accuracy and anti-interference capabilities.

[0081] In some embodiments, the snoring recognition module is also used to input the snoring feature vector into the snoring classification model to obtain the snoring judgment result;

[0082] The snoring recognition module is also used to analyze the snoring feature vector through the decibel regression algorithm of the snoring classification model when the snoring judgment result is that snoring is detected, to obtain snoring intensity data, and to determine the corresponding snoring intensity level based on the snoring intensity data.

[0083] Specifically, in the snoring detection stage, the snoring recognition module inputs a snoring feature vector, which integrates acoustic and physical vibration characteristics, into a pre-defined snoring classification model. This classification model employs a lightweight SVM (Support Vector Machine) algorithm architecture and, after training with a large number of snoring samples and interference signal samples, possesses strong feature discrimination capabilities. The model performs rapid inference on the input feature vector, and by judging the matching degree between the vector and the pre-defined snoring feature space, outputs a clear snoring judgment result, effectively eliminating irrelevant interference such as coughing, turning over, and environmental noise.

[0084] When the snoring detection result indicates that snoring has been detected, the snoring detection module will initiate an intensity grading process. Based on the decibel regression algorithm built into the snoring classification model, it will perform in-depth analysis of the snoring feature vector to accurately calculate the actual intensity data of the current snoring, thereby achieving a quantitative assessment of the snoring loudness.

[0085] After obtaining snoring intensity data, the module determines the corresponding snoring intensity level according to a preset grading standard. For example, a three-level grading result can be generated: 15dB-30dB is classified as Level 1 (mild snoring), 30dB-50dB as Level 2 (moderate snoring), and ≥50dB as Level 3 (heavy snoring). This grading method not only reflects the differences in snoring intensity in actual sleep scenarios but also provides a clear grading basis for the subsequent intervention module to formulate differentiated strategies, ensuring the targeted and reasonable nature of the intervention actions.

[0086] In the above embodiments, the classification model, through deep analysis of dual-modal fusion features, can accurately distinguish snoring from interference signals such as coughing, turning over, and environmental noise, significantly reducing the misjudgment rate of traditional single-feature recognition and ensuring the reliability of snoring judgment results. After snoring is detected, the quantification analysis of feature vectors through decibel regression algorithms can accurately obtain snoring intensity data. Then, based on preset standard intensity levels, a refined assessment of snoring intensity can be achieved, providing a precise basis for subsequent differentiated intervention. At the same time, thanks to the lightweight model architecture and efficient algorithm logic, the entire judgment and classification process responds quickly (≤0.5 seconds) and consumes low computing power, making it compatible with low-computing-power embedded devices such as STM32 for stable operation. This not only meets the real-time recognition requirements but also further improves the system's recognition accuracy and practicality.

[0087] In some embodiments, the data acquisition module is also used to acquire cardiac impact snoring vibration data after intervention;

[0088] The snoring recognition module is also used to obtain snoring intensity data after intervention based on cardiac impact snoring vibration data after intervention;

[0089] The snoring intervention module is also used to dynamically adjust the intervention plan based on the snoring intensity data after intervention and the snoring intensity data before intervention;

[0090] The snoring intervention module is also used to optimize the intervention parameters corresponding to different intensities of snoring based on the user's historical comfort score and historical intervention parameters through a nonlinear fitting algorithm.

[0091] Specifically, the data acquisition module remains operational throughout the intervention process, specifically collecting post-intervention data on the vibration of the user's snoring. This data undergoes the same preprocessing as the pre-intervention data, using an adaptive bandpass filtering algorithm to remove interference signals such as environmental vibrations and operating noise from the adjustment mechanism. This ensures that the data accurately reflects the actual vibration of the snoring after the intervention, providing high-fidelity raw data support for subsequent effect evaluation.

[0092] After receiving the vibration data following the intervention, the snoring recognition module uses the same feature extraction and grading logic as before the intervention to perform in-depth analysis of the data. By extracting the vibration time-frequency domain features, it generates a snoring feature vector after the intervention, which is then input into a lightweight SVM classification model and a decibel regression algorithm to accurately obtain the snoring intensity data after the intervention, achieving a standardized comparison of intensity data before and after the intervention.

[0093] Based on snoring intensity data before and after intervention, the snoring intervention module initiates a dynamic adjustment process, forming a real-time feedback loop. The module quantitatively compares the differences between the two: if the snoring intensity decreases by ≥20% after several prognoses, the intervention is deemed effective, and the current adjustment state is maintained and continuously monitored; if the decrease is between 10% and 20%, it indicates that the intervention direction is feasible but the parameters need optimization, and the intervention is performed again after fine-tuning the adjustment amplitude, speed, and other parameters; if the snoring intensity actually increases, the intervention direction is immediately reversed; if the decrease is still <10% after 5 consecutive adjustments, the current plan is deemed ineffective for the user or the snoring scenario, and the intervention is automatically stopped to avoid ineffective adjustments interfering with the user's sleep.

[0094] Meanwhile, the snoring intervention module also optimizes strategies based on long-term accumulated historical data, constructing a personalized adaptation system tailored to each individual user. The module links and stores the user's historical comfort scores for each sleep session (including user-submitted scores of 1-5 points or system-generated scores) with corresponding historical intervention parameters (such as snoring intensity level, adjustment area, amplitude, and speed). It then uses a non-linear fitting algorithm to deeply mine this correlated data. The algorithm analyzes the mapping relationship between different snoring intensities and optimal intervention parameters, as well as the correlation between intervention parameters and comfort scores. By continuously iterating and updating the intervention parameters corresponding to different snoring intensities and updating the weight allocation rules, subsequent intervention plans are better suited to individual user characteristics, achieving a steady improvement in long-term intervention effects.

[0095] The above embodiments continuously collect cardiac impact and snoring vibration data after intervention through a data acquisition module. The snoring intensity data after intervention is analyzed by a snoring recognition module. The snoring intervention module then dynamically adjusts the intervention plan based on the intensity data before and after intervention. Simultaneously, based on the user's historical comfort score and historical intervention parameters, a nonlinear fitting algorithm optimizes the intervention parameters corresponding to different snoring intensities, constructing a closed-loop system from real-time feedback to dynamic adjustment and long-term optimization. This system can both adjust the plan in real-time according to the intervention effect, avoiding ineffective adjustments or over-intervention and ensuring that the relief rate of medium-to-high intensity snoring remains stable at ≥80%, and iteratively optimize intervention parameters through historical data, making the intervention strategy more tailored to individual user characteristics and effectively reducing the user's wake-up rate (controlled within 5%). Furthermore, this mechanism overcomes the limitations of traditional intervention plans that lack feedback optimization, significantly improving the system's intervention targeting, sleep-friendly nature, and long-term adaptability, further enhancing the system's practical value for stable operation on low-computing-power embedded devices.

[0096] In some embodiments, the data acquisition module is further configured to acquire initial cardiac impact snoring vibration data and remove interference signals from the initial cardiac impact snoring vibration data using a preset adaptive bandpass filtering algorithm to obtain cardiac impact snoring vibration data.

[0097] Specifically, when collecting cardiac impact snoring vibration data, the data acquisition module first obtains initial cardiac impact snoring vibration data containing various signals such as snoring vibration, environmental vibration, bed shaking, and user turning over through the BCG sensor array. Then, it activates a preset adaptive bandpass filtering algorithm. This algorithm adjusts the filtering parameters for the typical frequency range of snoring vibration (such as 0.1Hz-50Hz), accurately selecting effective vibration signals related to snoring within this frequency band, while efficiently eliminating irrelevant signals such as environmental interference, random noise, and non-snoring physiological vibrations outside the frequency band. Finally, it obtains high-fidelity, high-purity cardiac impact snoring vibration data, providing reliable raw data support for subsequent preliminary snoring judgment, feature extraction, and fusion recognition.

[0098] The design of the above embodiment uses an adaptive bandpass filtering algorithm to accurately denoise the initial cardiac impact snoring vibration data, effectively removing non-snoring-related interference signals, significantly improving the purity and reliability of the cardiac impact snoring vibration data, and avoiding the misleading effect of interference signals on subsequent recognition processes. At the same time, the adaptive characteristics of the algorithm can adapt to signal changes in different sleep scenarios, ensuring that effective signals can still be stably screened in complex environments, further strengthening the system's anti-interference ability and helping to achieve a comprehensive snoring recognition accuracy of ≥90%.

[0099] This embodiment also provides a snoring recognition and intervention method. Figure 2 This is a flowchart of a snoring recognition intervention method according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps:

[0100] Step S201: Collect the user's cardiac impact snoring vibration data; generate the corresponding snoring feature vector based on the cardiac impact snoring vibration data;

[0101] Step S202: Input the snoring feature vector into the trained snoring classification model and output the snoring judgment result and snoring intensity level;

[0102] Step S203: When the snoring judgment result indicates that snoring has been detected, a corresponding intervention plan is generated based on the snoring intensity, and the smart pillow used by the user is adjusted based on the intervention plan.

[0103] First, the multi-sensor data collaborative acquisition and feature vector construction process is initiated. When the user's head touches the surface of the smart pillow, the device's built-in BCG sensor array continuously captures initial cardiac impact snoring vibration data from the head and neck. This data includes various signals such as snoring vibrations, environmental vibrations, bed shaking, and user turning over. Subsequently, the initial data is denoised using a preset adaptive bandpass filtering algorithm (e.g., 0.1Hz-50Hz), accurately removing irrelevant interference signals, extracting pure vibration data related to snoring, and extracting time-frequency domain features reflecting the physical vibration patterns.

[0104] Next, the system performs intelligent snoring recognition and intensity grading. The constructed snoring feature vector is input into a trained snoring classification model (such as a lightweight SVM classification model). This model, trained with a large number of snoring samples and interference signal samples, possesses strong feature discrimination capabilities, quickly eliminating non-snoring signals such as coughing, environmental noise, and turning over, and outputting a clear snoring judgment result. If the judgment result indicates snoring is detected, the system will activate the built-in decibel regression algorithm to quantify and analyze the core parameters related to sound intensity in the feature vector, and, combined with preset intensity grading standards, accurately determine the snoring intensity level. For example, 15dB-30dB is level 1 (mild snoring), 30dB-50dB is level 2 (moderate snoring), and ≥50dB is level 3 (heavy snoring). The entire recognition and grading process is rapid, taking ≤0.5 seconds, with a comprehensive recognition accuracy of ≥90%, ensuring both real-time performance and result reliability.

[0105] Finally, differentiated intervention plans are generated and implemented. Upon receiving the judgment result of the identified snoring and its corresponding intensity level, the system generates a targeted intervention plan based on the snoring intensity and the user's historical intervention data (including optimal adjustment parameters for snoring of similar intensity, comfort scores, etc.). For example, for level 1 mild snoring, only small-amplitude, slow adjustments are made to the corresponding support area under the head, slightly optimizing the airway angle through minor height changes; for level 2 moderate snoring, the adjustment range is expanded to the head and neck support areas, using moderate-amplitude adjustments to simultaneously adapt to the head and neck support height, further improving airway patency; for level 3 heavy snoring, multi-area coordinated adjustment is initiated, using maximum-amplitude adjustments such as raising the head and adjusting the side sleeping direction to effectively alleviate airway obstruction. After the solution is generated, the system will transmit the control signal to the zoned adjustment actuator inside the smart pillow, driving the high-precision stepper motor and telescopic rod of the corresponding area to move. The adjustment accuracy is ≤1mm and the minimum adjustment speed is ≤1mm / s. While ensuring a ≥80% reduction rate for medium and high intensity snoring, the system will control the user wake-up rate to within 5%, achieving a balance between intervention effect and sleep friendliness.

[0106] In the above steps, by collecting cardiac impact snoring vibration data and generating corresponding snoring feature vectors, these vectors are input into a trained snoring classification model to obtain snoring judgment results and snoring intensity levels. Based on these intensity levels, corresponding intervention plans are generated and the smart pillow is adjusted. This approach effectively eliminates interference from environmental noise and coughing based on the physical vibration characteristics of cardiac impact signals, significantly reducing the misjudgment rate of traditional single-audio recognition and improving the accuracy and anti-interference capability of snoring recognition. Furthermore, by using a differentiated intervention strategy adapted to snoring intensity levels, it avoids the problem of insufficient targeting in overall adjustment. While ensuring a ≥80% relief rate for medium-to-high intensity snoring, it keeps the user wake-up rate below 5% through precise and gentle adjustment actions. At the same time, the process is adapted to a lightweight algorithm architecture, reducing computing power consumption and enabling stable operation on low-computing-power embedded devices such as STM32. This achieves a synergy of accurate recognition, graded intervention, and sleep-friendly features, effectively solving the core problems of low recognition accuracy, poor intervention targeting, and easy sleep disturbance in existing technologies.

[0107] In addition, in conjunction with the snoring recognition and intervention methods in the above embodiments, this application embodiment can provide a storage medium for implementation. The storage medium stores a computer program; when the computer program is executed by a processor, it implements any of the snoring recognition and intervention methods in the above embodiments.

[0108] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0109] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0110] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A snoring recognition and intervention system, characterized in that, The system includes: a data acquisition module, a snoring recognition module, and a snoring intervention module; The data acquisition module is used to collect the user's cardiac impact and snoring vibration data; The snoring recognition module is used to generate a corresponding snoring feature vector based on the cardiac impact snoring vibration data, and input the snoring feature vector into the trained snoring classification model to output the snoring judgment result and snoring intensity level. The snoring intervention module is used to generate a corresponding intervention plan based on the snoring intensity level when the snoring judgment result indicates that snoring has been detected, and to adjust the smart pillow used by the user based on the intervention plan.

2. The snoring recognition and intervention system according to claim 1, characterized in that, The data acquisition module includes an audio sensing unit; the snoring recognition module includes a cardiac impact vibration component analysis unit and an audio wake-up control module. The cardiac impact vibration component analysis unit is used to analyze the cardiac impact snoring vibration data to obtain a preliminary snoring judgment result, and to trigger the audio wake-up control module when the preliminary snoring judgment result is suspected snoring. The audio sensing unit is used to collect sound data in response to the triggering of the audio wake-up control module; The snoring recognition module is also used to generate the snoring feature vector based on the cardiac impact snoring vibration data and the sound data.

3. The snoring recognition and intervention system according to claim 2, characterized in that, The snoring recognition module is also used to analyze the cardiac impact snoring vibration data through a preset time-series period detection algorithm to identify the vibration signal period, and determine the preliminary snoring judgment result based on the vibration signal period.

4. The snoring recognition and intervention system according to claim 2, characterized in that, The snoring recognition module is also used to extract key audio features of the corresponding snoring based on the sound data, and to extract corresponding time-frequency domain features based on the cardiac impact snoring vibration data. The snoring recognition module is further configured to fuse the key audio features of snoring and the time-frequency domain features according to preset weights to generate the snoring feature vector.

5. The snoring recognition and intervention system according to claim 4, characterized in that, The snoring recognition module is also used to extract the corresponding initial key audio features of snoring based on the sound data, and to reduce the dimensionality of the initial key audio features of snoring to a preset dimension through a principal component analysis algorithm to obtain the key audio features of snoring.

6. The snoring recognition and intervention system according to claim 5, characterized in that, The snoring recognition module is also used to perform hardware noise reduction and digital conversion on the sound data to obtain standard audio data, and input the standard audio data into the Mel spectrogram architecture to generate a Mel spectrogram; The snoring recognition module is also used to extract key audio features of the initial snoring based on the Mel spectrogram; the key audio features of the initial snoring include Mel frequency cepstral coefficients, spectral flatness, and fundamental frequency.

7. The snoring recognition and intervention system according to claim 1, characterized in that, The snoring recognition module is also used to input the snoring feature vector into the snoring classification model to obtain the snoring judgment result; The snoring recognition module is further configured to, when the snoring judgment result indicates that snoring has been detected, analyze the snoring feature vector using the decibel regression algorithm of the snoring classification model to obtain snoring intensity data, and determine the corresponding snoring intensity level based on the snoring intensity data.

8. The snoring recognition and intervention system according to claim 7, characterized in that, The data acquisition module is also used to collect cardiac impact snoring vibration data after intervention; The snoring recognition module is also used to obtain snoring intensity data after intervention based on the cardiac impact snoring vibration data after intervention; The snoring intervention module is also used to dynamically adjust the intervention plan based on the snoring intensity data after the intervention and the snoring intensity data before the intervention; The snoring intervention module is also used to optimize the intervention parameters corresponding to different intensities of snoring based on the user's historical comfort score and historical intervention parameters using a nonlinear fitting algorithm.

9. The snoring recognition and intervention system according to claim 1, characterized in that, The data acquisition module is also used to acquire initial cardiac impact snoring vibration data, and to remove interference signals from the initial cardiac impact snoring vibration data through a preset adaptive bandpass filtering algorithm to obtain the cardiac impact snoring vibration data.

10. A snoring detection and intervention method, characterized in that, The method includes: Collect the user's cardiac impact snoring vibration data; generate the corresponding snoring feature vector based on the cardiac impact snoring vibration data; The snoring feature vector is input into the trained snoring classification model, which outputs the snoring judgment result and the snoring intensity level. When the snoring sound is detected, a corresponding intervention plan is generated based on the snoring sound intensity, and the smart pillow used by the user is adjusted based on the intervention plan.

Images