Bluetooth earphone intelligent protection method based on acoustic characteristics and Bluetooth earphone

Abstract

The invention provides a Bluetooth headset intelligent protection method based on acoustic characteristics and a Bluetooth headset, and relates to the technical field of nursing appliances for old people, the method comprises the following steps: collecting original sound signals of three types of sound of original sound signals of normal bathroom activities and carrying out noise reduction processing; extracting time sequence combination features corresponding to the water flow sound, the human body activity sound and the voice sound to obtain a sound combination mode; establishing an acoustic behavior model of normal bathroom activities based on a sound combination mode; matching and comparing a new sound combination mode corresponding to a new original sound signal collected in real time with the acoustic behavior model, and determining an abnormal feature corresponding to the new original sound signal; and determining a danger level according to the triggering number and the duration of the abnormal features, and executing a preset emergency scheme corresponding to the danger level. By implementing the method, the Bluetooth headset can accurately identify the abnormal condition for early warning by analyzing the sequential relationship and the interaction characteristics of the sound, and a passive protection mode that a traditional scheme depends on manual key pressing is avoided.

Description

Technical Field

[0001] This application relates to the technical field of elderly care appliances, and particularly to an intelligent protection method for Bluetooth headsets based on acoustic features and a Bluetooth headset. Background Art

[0002] With the continuous deepening of social aging, the safety protection problem of elderly people living alone has become increasingly prominent. As an essential place in daily life, the bathroom has become one of the places where the elderly are most likely to have accidents due to its humid environmental characteristics and slippery floor conditions. Especially for the elderly group who are used to wearing Bluetooth headsets to listen to the radio or music while taking a bath, how to use the devices they wear to achieve safety protection has become an urgent problem to be solved.

[0003] In related technologies, the protection solutions for the safety of the elderly in the bathroom mainly adopt an emergency call button system. Such a system installs waterproof buttons on the bathroom wall. When the elderly have an accident, they can send a distress signal to their family or the rescue center by pressing the button. To improve safety, some systems also install auxiliary facilities such as anti-slip mats and handrails in the bathroom to reduce the risk of accidents.

[0004] However, this passive safety protection solution has obvious deficiencies. First, when the elderly have an accident such as falling, it is often difficult to reach the emergency button on the wall in time; second, some elderly people may experience short-term confusion after an accident and are difficult to take proactive help-seeking behaviors; finally, this solution is difficult to achieve early warning of dangerous situations and can only trigger the rescue mechanism after an accident occurs, delaying the best rescue time. Summary of the Invention

[0005] This application provides an intelligent protection method for Bluetooth headsets based on acoustic features and a Bluetooth headset, which is used to solve the problem of how to use the Bluetooth headsets that the elderly wear daily to achieve real-time monitoring of the bathroom activity state and proactive safety protection, so as to timely detect abnormal situations and automatically trigger emergency measures.

[0006] In a first aspect, this application provides an intelligent protection method for Bluetooth headsets based on acoustic features, which is applied to Bluetooth headsets. The method includes: Collecting the original sound signals of normal bathroom activities through the microphone array of the Bluetooth headset. The original sound signals include three types of sounds: water flow sounds, human activity sounds, and voice sounds; Performing hierarchical noise reduction processing on the original sound signals to obtain noise reduction signals, and then separating and extracting the water flow sounds, the human activity sounds, and the voice sounds from the noise reduction signals; Extract the corresponding sequential combination features of the water flow sound, the human activity sound, and the voice sound to obtain a sound combination pattern, where the sound combination pattern includes the duration feature of the water flow sound, the sequential correlation feature between the human activity sound and the water flow sound, and the interaction feature between the voice sound and the water flow sound and the human activity sound; Based on the sound combination pattern, establish an acoustic behavior model for normal bathroom activities, where the acoustic behavior model is used to characterize the sequential relationship of various sounds in normal bathroom activities; Match and compare the new sound combination pattern corresponding to the newly collected original sound signal in real time with the acoustic behavior model to determine the abnormal features corresponding to the new original sound signal; Determine the danger level according to the triggering number and duration of the abnormal features, and execute the preset emergency plan corresponding to the danger level.

[0007] Through the above embodiments, the bathroom activity sounds are collected by the microphone array of the Bluetooth headset, and are subjected to hierarchical noise reduction and sound separation, the sequential combination features of the water flow sound, the human activity sound, and the voice sound are extracted, and an acoustic behavior model for normal bathroom activities is established. This solution does not require additional equipment installation, and can achieve safety protection by using the Bluetooth headset that the elderly wear daily. By analyzing the sequential relationship and interaction features of the sounds, abnormal situations can be accurately identified, avoiding the passive protection mode that the traditional solution relies on manual buttons. At the same time, a hierarchical early warning mechanism is adopted, and the danger level is determined according to the triggering number and duration of the abnormal features, which can give an early warning in time before an accident occurs, greatly improving the timeliness and effectiveness of the protection.

[0008] In some embodiments, the step of extracting the corresponding sequential combination features of the water flow sound, the human activity sound, and the voice sound to obtain a sound combination pattern specifically includes: Perform sequential segmentation on the water flow sound, the human activity sound, and the voice sound to obtain a sound event sequence with time stamps; Based on the sound event sequence, construct a state transition matrix, where the rows and columns of the state transition matrix respectively represent the current sound state and the next moment sound state, and the matrix elements represent the corresponding transition probabilities; Construct a sequential dependence graph through the sound event sequence and the state transition matrix, where the nodes of the sequential dependence graph represent sound events, the edges represent the sequential correlation between events, and the edge weights represent the dependence intensity; Determine the typical path features according to the sequential dependence graph, where the typical path features include path length, node type distribution, and edge weight distribution; Perform weighted fusion on the transition probability features of the state transition matrix and the typical path features to obtain the sound combination pattern.

[0009] Through the above embodiments, the Bluetooth headset segments the sound in time series and constructs a state transition matrix to establish a time series dependence graph between sound events, which can more accurately depict the evolution law of various sounds in normal bathroom activities. By adopting the weighted fusion method of typical path features and transition probability features, it not only considers the local transition characteristics of sound events but also takes into account the global time series evolution pattern, making the constructed sound combination pattern have stronger expression ability and discrimination ability, and can more accurately identify abnormal sound combination patterns.

[0010] In some embodiments, before the step of matching the new sound combination pattern corresponding to the newly collected original sound signal with the acoustic behavior model to determine the abnormal features corresponding to the newly collected original sound signal, it further includes: Establish a threshold adjustment matrix corresponding to different preset time periods, where the threshold adjustment matrix includes a daytime threshold coefficient and a nighttime threshold coefficient; Determine the noise intensity coefficient corresponding to the newly collected original sound signal; Perform weighted fusion on the threshold adjustment matrix and the noise intensity coefficient to obtain a real-time matching threshold, which is used to match and compare the new sound combination pattern with the acoustic behavior model.

[0011] Through the above embodiments, the Bluetooth headset solves the problem that a fixed threshold is difficult to adapt to the differences in sound characteristics in different time periods by establishing a threshold adjustment matrix for different preset time periods and dynamically adjusting the threshold in combination with the real-time noise intensity coefficient. This solution can adaptively adjust the matching threshold according to the environmental noise characteristics during the day and at night, which not only ensures that there are no false alarms during normal activities during the day but also ensures sensitive detection at night, significantly improving the accuracy of the system in different scenarios.

[0012] In some embodiments, the step of determining the noise intensity coefficient corresponding to the newly collected original sound signal specifically includes: Collect multiple environmental background noise samples within a preset time window; Calculate the mean and standard deviation of the environmental background noise samples; Establish a noise evaluation model based on the mean and the standard deviation; Input the newly collected original sound signal into the noise evaluation model to obtain the noise intensity coefficient.

[0013] Through the above embodiments, the Bluetooth headset can accurately evaluate the noise level of the current environment by collecting multiple environmental background noise samples and establishing a noise evaluation model. This solution not only considers the average intensity of the noise but also introduces the standard deviation to characterize the fluctuation characteristics of the noise, making the calculation of the noise intensity coefficient more reasonable. This noise evaluation method based on statistical characteristics can effectively filter out the interference of environmental noise and improve the reliability of abnormal detection.

[0014] In some embodiments, the step of matching and comparing the new sound combination pattern corresponding to the newly acquired real-time original sound signal with the acoustic behavior model to determine the abnormal features corresponding to the new original sound signal specifically includes: continuously collecting a plurality of the new sound combination patterns within a preset sliding time window; Calculating the abnormal score for each of the new sound combination patterns to obtain an abnormal score sequence; Based on the abnormal score sequence, establishing an abnormal development trend curve; Determining the abnormal features according to the change rate of the abnormal development trend curve and the corresponding abnormal score.

[0015] Through the above embodiments, the Bluetooth headset continuously collects new sound combination patterns by using a sliding time window. By establishing an abnormal development trend curve, the evolution process of the abnormal state can be effectively tracked. This solution not only focuses on the instantaneous degree of abnormality but also considers the development trend of the abnormal state. By analyzing the change rate of the trend curve, potential risks can be discovered earlier, and preventive measures can be taken in advance. This dynamic tracking mechanism significantly improves the early warning ability and protection effect of the system.

[0016] In some embodiments, the step of calculating the abnormal score for each of the new sound combination patterns to obtain an abnormal score sequence specifically includes: Respectively extracting the timing feature, spectral feature, and energy feature of each of the new sound combination patterns; Comparing the timing feature, spectral feature, and energy feature with the real-time matching threshold respectively, and calculating the abnormal score corresponding to each of the new sound combination patterns based on a preset feature weight; Sorting the abnormal scores according to the time sequence of the new sound combination patterns to obtain an abnormal score sequence.

[0017] Through the above embodiments, the Bluetooth headset realizes multi-dimensional abnormal evaluation by simultaneously extracting the timing feature, spectral feature, and energy feature and performing comprehensive scoring based on a preset feature weight. This solution makes full use of various feature information of the sound signal, considering both the variation law in the time domain and the frequency domain characteristics and energy distribution, making the abnormal score more comprehensive and accurate. By sorting the score sequence in time sequence, the evolution process of the abnormal state can be clearly reflected.

[0018] In some embodiments, before the step of determining the risk level according to the trigger quantity and duration of the abnormal features and executing the preset emergency plan corresponding to the risk level, it further includes: Receiving the user's re-setting of the preset emergency plans corresponding to different risk levels.

[0019] Through the above embodiments, the Bluetooth headset provides high system customization flexibility by allowing users to reset the preset emergency solutions corresponding to different risk levels. This solution can set personalized emergency response strategies according to the actual situations of different elderly people and family needs, such as choosing whether to automatically alarm, whether to notify specific contacts, etc. This customizable emergency solution not only meets the personalized needs of different users but also improves the practicality and applicability of the system.

[0020] In a second aspect, the present application provides a Bluetooth headset, which includes: one or more processors and a memory; The memory is coupled to the one or more processors. The memory is used to store computer program code, and the computer program code includes computer instructions. The one or more processors call the computer instructions so that the Bluetooth headset can implement an intelligent protection method for a Bluetooth headset based on acoustic features provided by the above embodiments, which will not be elaborated here.

[0021] In a third aspect, the present application provides a computer-readable storage medium, including instructions. When the instructions run on a Bluetooth headset, the Bluetooth headset can implement an intelligent protection method for a Bluetooth headset based on acoustic features provided by the above embodiments, which will not be elaborated here.

[0022] In a fourth aspect, the present application provides a computer program product. When the computer program product runs on a Bluetooth headset, the Bluetooth headset can implement an intelligent protection method for a Bluetooth headset based on acoustic features provided by the above embodiments, which will not be elaborated here.

[0023] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Utilize the Bluetooth headset that the elderly wear daily to achieve bathroom safety protection. Collect and separate three types of sound signals, namely water flow sound, human activity sound, and voice sound, through a microphone array, and construct an acoustic behavior model by combining the temporal combination features and interaction relationships of these sounds. This solution does not require additional equipment installation and can achieve safety monitoring by making full use of existing equipment, which not only protects user privacy but also avoids the passive protection mode that relies on manual buttons in traditional solutions, making the safety protection more proactive and timely.

[0024] 2. Adopt a multi-dimensional feature analysis mechanism based on a sound temporal dependence graph, characterize the evolution law of sound events through a state transition matrix, and perform weighted fusion in combination with typical path features to achieve an accurate modeling of the acoustic pattern of normal bathroom activities. At the same time, a day-night dynamic threshold adjustment and a noise assessment model are introduced, enabling the system to adaptively adjust the detection sensitivity according to the environmental characteristics of different periods, significantly improving the accuracy and reliability of anomaly detection.

[0025] 3. A dynamic risk assessment mechanism based on the abnormal development trend curve is proposed. By analyzing the change rate and duration of the abnormal score sequence, early warning of dangerous situations is achieved. Combined with customizable multi-level emergency plans, the system can not only detect potential risks in a timely manner and automatically take preventive measures, but also flexibly configure emergency response strategies according to the actual needs of different users, greatly improving the practicability and adaptability of the protection plan. Brief Description of the Drawings

[0026] Figure 1 is a schematic flowchart of a method for intelligent protection of a Bluetooth headset based on acoustic features in an embodiment of the present application; Figure 2 is another schematic flowchart of a method for intelligent protection of a Bluetooth headset based on acoustic features in an embodiment of the present application; Figure 3 is a schematic structural diagram of an entity device of a Bluetooth headset in an embodiment of the present application. Detailed Embodiments

[0027] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and claims of the present application, the singular forms "a", "an", "the", "above", "said", "this" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations including one or more of the listed items.

[0028] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as implying or indicating relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "plurality" is two or more.

[0029] For ease of understanding, the following describes the process of the method provided in this embodiment. Please refer to Figure 1 , which is a schematic flowchart of a method for intelligent protection of a Bluetooth headset based on acoustic features in an embodiment of the present application.

[0030] S101. Collect the original sound signal of normal bathroom activities through the microphone array of the Bluetooth headset.

[0031] The Bluetooth headset uses a built-in microphone array (such as a dual-microphone or four-microphone array) to directionally collect bathroom sounds by means of beamforming technology. Specifically, after the Bluetooth headset enters the learning mode, the user performs normal activities (such as taking a shower, scrubbing, speaking), and the microphone synchronously collects multi-channel raw sound signals at a sampling rate of 44.1 kHz and a precision of 16 bits. The array calculates the sound source direction through the time difference of arrival (TDOA) and phase difference, combines adaptive filtering to suppress noise in non-target directions, and collects a mixed signal containing water flow sounds (200 Hz - 8 kHz), human activity sounds (100 Hz - 5 kHz), and speech sounds (80 Hz - 4 kHz), and synchronously records timestamps to provide a timing basis for subsequent analysis.

[0032] S102. Perform hierarchical noise reduction processing on the raw sound signal to obtain a noise-reduced signal, and then separate and extract the water flow sound, human activity sound, and speech sound from the noise-reduced signal.

[0033] The raw signal first undergoes hierarchical noise reduction: in the first stage, based on the subspace noise reduction algorithm, the bathroom background noise pre-collected (such as pipe sounds, ventilation fan sounds) is used to construct a noise subspace, and the environmental noise is suppressed through spectral subtraction; in the second stage, a deep learning speech enhancement model (such as DCCRN) is used to further eliminate residual speech interference to obtain a noise-reduced signal. Subsequently, an improved independent component analysis (ICA) algorithm is used, combined with features such as the broadband of water flow sounds, the low-frequency pulses of human activity sounds, and the fundamental frequency of speech sounds for blind source separation, and then a recurrent neural network with attention mechanism (RNN-ATT) is introduced to optimize the separation result, and the three types of sounds, namely water flow sound, human activity sound, and speech sound, are accurately extracted.

[0034] S103. Extract the corresponding timing combination features of the water flow sound, human activity sound, and speech sound to obtain a sound combination pattern.

[0035] For the three types of separated sounds, the Bluetooth headset divides the continuous signal into independent events (such as the start and stop of water flow sounds, human activity actions) through dynamic time warping (DTW), marks the timestamps to form an event sequence. Then, the state transition probability between events is statistically calculated to construct a state transition matrix, a timing dependence graph is constructed with events as nodes and the timing correlation intensity as edge weights, and the high-frequency typical paths are extracted through depth-first search (DFS), and the path length, node distribution, etc. are recorded. Finally, the state transition probability and the typical path features are fused according to weights to form a combined pattern vector reflecting the timing correlation of sounds.

[0036] S104. Establish an acoustic behavior model of normal bathroom activities based on the sound combination pattern.

[0037] The Bluetooth headset can collect data on the sound combination patterns of multiple groups of users' normal activities, train using the Hidden Markov Model (HMM), set 3 states (corresponding to three types of sounds), with the observation probability represented by the Gaussian Mixture Model (GMM) and the transition probability based on typical path feature statistics. During training, different-length sequences are aligned through Dynamic Time Warping (DTW) to enhance the robustness of the model, and then the Bayesian Information Criterion (BIC) is used to determine the optimal model order to avoid overfitting. Finally, the trained HMM parameters (initial state probability, transition matrix, observation probability) are stored in the headset memory to form an acoustic behavior model for normal activities.

[0038] S105. Match and compare the new sound combination pattern corresponding to the newly collected original sound signal in real time with the acoustic behavior model to determine the abnormal features corresponding to the new original sound signal.

[0039] After the Bluetooth headset collects sounds in real time, it extracts the features of the new sound combination pattern according to the steps. Specifically, a threshold adjustment matrix is selected according to the current time period (such as night, day), and the matching threshold is determined in combination with the real-time noise intensity coefficient (calculating the background noise mean and standard deviation). The matching degree between the real-time feature and the model is calculated through the Log Likelihood Ratio (LLR), and if it exceeds the threshold, it is marked as abnormal. If an abnormality is continuously detected 3 times, the change rate of the abnormal score sequence is further analyzed. When the rate exceeds 0.5 / second and the score > 70 points, it is determined as an emergency abnormality to clarify the abnormal features.

[0040] S106. Determine the danger level according to the number of triggers and the duration of the abnormal features, and execute the preset emergency plan corresponding to the danger level.

[0041] The danger level is divided according to the number of abnormal triggers and the duration: 1 abnormal occurrence within 1 hour and the duration < 1 minute is level one (the headset vibrates to remind); 2 abnormal occurrences within 30 minutes and the duration is 1 - 3 minutes is level two (play the voice prompt "Please confirm safety"); 3 abnormal occurrences within 10 minutes and the duration > 3 minutes is level three (automatically call the emergency contact). After determining the level, information is sent or a call is made by connecting to the mobile phone via Bluetooth to call the API. If the mobile phone has no signal, the built-in LTE module of the headset is activated to send the location to the cloud to execute the corresponding emergency plan.

[0042] In the above embodiment, the Bluetooth headset collects bathroom activity sounds through a microphone array, performs hierarchical noise reduction and sound separation, extracts the sequential combination features of water flow sounds, human activity sounds, and voice sounds, and establishes an acoustic behavior model for normal bathroom activities. This solution does not require additional equipment installation and can achieve safety protection by using the Bluetooth headset that the elderly wear daily. By analyzing the temporal relationship and interaction features of sounds, abnormal situations can be accurately identified, avoiding the passive protection mode that traditional solutions rely on manual buttons. At the same time, a hierarchical warning mechanism is adopted to determine the danger level according to the number of triggered abnormal features and the duration of persistence, which can give a timely warning before an accident occurs, greatly improving the timeliness and effectiveness of protection.

[0043] The following is a further and more specific process description of the method provided in this embodiment. Please refer to Figure 2 , which is another process schematic diagram of an intelligent protection method for Bluetooth headsets based on acoustic features in the embodiments of the present application.

[0044] S201. Perform temporal segmentation on water flow sounds, human activity sounds, and voice sounds to obtain a sound event sequence with timestamps.

[0045] After the Bluetooth headset completes the separation and extraction of water flow sounds, human activity sounds, and voice sounds, it conducts temporal segmentation on these three types of sound signals. In this process, the Bluetooth headset first analyzes the time-domain waveform of each type of sound signal and identifies the start and end moments of sound events through a preset energy threshold. For example, for water flow sounds, when the signal energy exceeds the set low energy threshold (such as 0.3 times the maximum energy value), it is determined that the water flow sound starts; when the signal energy continuously remains below this threshold for a certain duration (such as 0.5 seconds), it is determined that the water flow sound ends. For human activity sounds and voice sounds, corresponding thresholds are set for judgment according to their respective energy change characteristics.

[0046] After determining the boundaries of sound events, the Bluetooth headset adds accurate timestamps to each event. The accuracy of the timestamps can reach the millisecond level to accurately record the specific moment when the sound event occurs. These sound events with timestamps are arranged in chronological order to form a sound event sequence. This sequence not only completely records the distribution of various sound events on the time axis but also provides basic data for subsequent analysis of the temporal relationship between sounds.

[0047] Optionally, a method based on short-time energy analysis is adopted to detect the boundaries of sound events. By performing short-time Fourier transform (STFT) on the sound signal, the time-domain signal is converted to the frequency domain, and then the energy of each time segment is calculated. At the same time, the dynamic threshold adjustment technology is used to adaptively adjust the energy threshold according to the energy characteristics of different types of sound signals and the changes in ambient noise, so as to improve the accuracy of sound event detection. The generation of timestamps depends on the high-precision clock module inside the Bluetooth headset to ensure accurate time recording.

[0048] S202. Construct a state transition matrix based on the sound event sequence and construct a temporal dependence graph in combination with the sound event sequence.

[0049] Based on the sound event sequence obtained in step S201, the Bluetooth headset starts to construct a state transition matrix. First, determine the sound states. Here, the sound of running water, the sound of human activities, and the sound of speech are regarded as different states respectively. Then, count the number of times of transitioning from one state to another in all sound events. For example, count the number of times of transitioning from the running water sound state to the human activity sound state, and the number of times of transitioning from the human activity sound state to the speech sound state, etc. Then, calculate the transition probability according to the number of transitions. The transition probability is equal to the number of times of transitioning from one state to another divided by the total number of times this state appears. Arranging these transition probabilities according to the corresponding states forms the state transition matrix.

[0050] When constructing the temporal dependence graph, the sound events are used as nodes, and the edge between two nodes indicates that there is a chronological association between these two sound events. The weight of the edge is determined according to the time interval and the frequency of occurrence between the sound events. If two sound events often occur successively within a short time interval, then the weight of the edge between them is larger; otherwise, the weight is smaller. In this way, the temporal dependence relationship between various sound events can be intuitively displayed.

[0051] Taking the scenario of an elderly person taking a bath as an example, after a period of sound data collection and analysis, the obtained state transition matrix is as follows (assuming only considering the transitions between three sound states): Current state \ Next state Sound of water flow Sound of human activities Voice sound Sound of water flow 0.7 0.2 0.1 Sound of human activities 0.3 0.5 0.2 Voice sound 0.1 0.4 0.5 The above state transition matrix can be understood as follows: after the sound of running water appears, there is a 70% probability that the next sound state is still the sound of running water, a 20% probability of transitioning to the human activity sound state, and a 10% probability of transitioning to the speech sound state.

[0052] In another embodiment, when constructing the temporal dependence graph, if the sound of water flow and the sound of human activities often appear successively within a short period of time, the weight of the edge between them may be set to 0.8; while the time interval between the sound of water flow and the sound of speech is long and infrequent, the weight of the edge between them may be set to 0.3, so as to clearly show the strength of the temporal dependence relationship between sound events.

[0053] S203. Determine the typical path features based on the temporal dependence graph, and perform weighted fusion with the transition probability features of the state transition matrix to obtain the sound combination pattern.

[0054] The Bluetooth headset searches for typical paths in the constructed temporal dependence graph through a specific algorithm. Among them, the typical path refers to the path with a high frequency of occurrence and representativeness in the graph, which reflects the common evolution order of sound events in normal bathroom activities. When searching for typical paths, factors such as the length of the path, the distribution of node types passed through, and the distribution of edge weights can be selected as screening conditions. For example, a path with a moderate length, passing through nodes of different types of sound events and having a large edge weight is more likely to be recognized as a typical path.

[0055] After determining the typical path features, set appropriate weights for the typical path features and the transition probability features respectively, and then perform weighted fusion on the two to finally obtain the sound combination pattern. This sound combination pattern combines the temporal dependence relationship of sound events and the state transition probability information, and can more comprehensively and accurately describe the sound characteristics in normal bathroom activities.

[0056] Optionally, when determining the typical path features, use the depth-first search (DFS) or breadth-first search (BFS) algorithm to traverse the temporal dependence graph, and combine the path evaluation function to screen out the typical paths. The path evaluation function comprehensively considers factors such as the length of the path, the diversity of node types, and the edge weight.

[0057] In the weighted fusion process, use the mathematical method of weighted summation to calculate the typical path feature vector and the transition probability feature vector according to the preset weights. The setting of the weights can be optimized through machine learning algorithms. For example, use a large amount of sound data of normal bathroom activities for training to determine the optimal weight value, which is not limited here.

[0058] S204. Collect multiple environmental background noise samples within a preset time window, and calculate the mean and standard deviation of the environmental background noise samples.

[0059] After the Bluetooth headset enters the working state, it periodically collects environmental background noise samples according to a preset time window. Among them, the duration of the preset time window is usually determined according to the actual application scenario and noise characteristics, such as being set to 5 seconds or 10 seconds. Within each time window, the Bluetooth headset uses its built-in microphone array to continuously collect sound signals in the environment. These sound signals contain various possible background noises, such as the water flow sound of pipes in the bathroom and the operation sound of ventilation equipment.

[0060] After collecting multiple environmental background noise samples, the Bluetooth headset processes these samples. First, each sample is transformed from the time domain to the frequency domain, which is usually achieved through the fast Fourier transform (FFT). After the transformation to the frequency domain, the noise amplitude at each frequency component is calculated separately. Then, the mean and standard deviation of all samples are calculated based on these amplitude data. Among them, the mean reflects the average intensity level of the environmental background noise, while the standard deviation reflects the degree of fluctuation of the noise amplitude around the average value.

[0061] S205. Establish a noise evaluation model based on the mean and standard deviation, and input the new original sound signal into the noise evaluation model to obtain the noise intensity coefficient.

[0062] The Bluetooth headset establishes a noise evaluation model based on the mean and standard deviation of the environmental background noise samples calculated in step S204. Among them, the noise evaluation model can adopt a model based on statistical distribution. For example, it is assumed that the environmental background noise follows a normal distribution. The parameters of the normal distribution are determined using the mean and standard deviation, thereby constructing a model that can describe the characteristics of the environmental noise.

[0063] After establishing the noise evaluation model, the newly collected original sound signal in real time is input into this model. The new original sound signal also undergoes preprocessing steps similar to those of the environmental background noise samples, is transformed to the frequency domain, and the amplitudes of each frequency component are obtained. The model calculates the deviation degree of the new original sound signal at each frequency relative to the background noise based on the amplitude data of the input signal and the established noise distribution characteristics. By synthesizing the deviation conditions of each frequency, a noise intensity coefficient is finally obtained. This coefficient can quantify the noise intensity in the new original sound signal and provide a basis for subsequent threshold adjustment and anomaly detection.

[0064] S206. Perform weighted fusion on the noise intensity coefficient corresponding to the new original sound signal and the threshold adjustment matrix corresponding to the preset time period to obtain the real-time matching threshold.

[0065] The Bluetooth headset has pre-stored threshold adjustment matrices corresponding to different preset time periods, and these matrices contain daytime threshold coefficients and nighttime threshold coefficients. The environmental noise characteristics vary in different time periods. For example, the environment is relatively quiet at night with lower background noise, while there may be more external interferences and higher noise levels during the day. Therefore, it is necessary to adjust the matching threshold according to different time periods to improve the accuracy of the system.

[0066] After obtaining the noise intensity coefficient of the new original sound signal, it is weighted and fused with the threshold adjustment matrix corresponding to the preset time period. The process of weighted fusion is to calculate the noise intensity coefficient and the coefficients in the threshold adjustment matrix according to the preset weights.

[0067] For example, assume the noise intensity coefficient is k, the daytime threshold coefficient is a, the nighttime threshold coefficient is b, and the weights are w1 and w2 (w1 + w2 = 1). If the current time is during the day, the formula for calculating the real-time matching threshold T is: T = w1 × k × a + w2 × a; If it is at night, the formula for calculating the real-time matching threshold T is: T = w1 × k × b + w2 × b.

[0068] Through such a weighted fusion method, the Bluetooth headset dynamically generates a suitable real-time matching threshold according to the current environmental noise intensity and time period characteristics, which is used for subsequent matching and comparison of the new sound combination pattern and the acoustic behavior model.

[0069] S207: Continuously collect multiple new sound combination patterns within the preset sliding time window, and extract the temporal characteristics, spectral characteristics, and energy characteristics of the new sound combination patterns.

[0070] The Bluetooth headset continuously and uninterruptedly collects new sound data according to the preset sliding time window. The size and sliding step of the sliding time window are set according to actual application requirements. For example, the window size is set to 10 seconds and the sliding step is set to 1 second, which can ensure data continuity while capturing sound changes in a timely manner. Within each window, the Bluetooth headset processes the collected sound signal, and through a series of operations such as temporal segmentation, constructing a state transition matrix, constructing a temporal dependence graph, and feature fusion mentioned above, a new sound combination pattern is obtained.

[0071] After obtaining the new sound combination pattern, the Bluetooth headset starts to extract its various features. When extracting the timing features, analyze the distribution law of sound events on the time axis, including the start time, duration, interval time, etc. of the events; the spectral features are obtained by performing a Fourier transform on the sound signal to acquire the energy distribution of the signal at different frequencies, so as to analyze the frequency composition characteristics of the sound; the energy features reflect the intensity information of the sound by calculating the energy magnitude of the sound signal. These features comprehensively describe the characteristics of the sound combination pattern from different dimensions, providing rich data support for the subsequent abnormal score calculation.

[0072] S208. Compare the timing features, spectral features, and energy features with the real-time matching thresholds respectively, and calculate the abnormal score corresponding to each new sound combination pattern based on the preset feature weights.

[0073] The Bluetooth headset compares the timing features, spectral features, and energy features of the new sound combination pattern extracted in step S207 with the real-time matching thresholds obtained in step S206 respectively. For the timing features, compare the differences between the time parameters (such as duration, interval time, etc.) of the sound events and the thresholds; for the spectral features, compare the deviation degree of the energy distribution in different frequency bands from the thresholds; for the energy features, directly compare the calculated energy value with the size relationship of the thresholds.

[0074] At the same time, based on the preset feature weights, comprehensively calculate these comparison results. For example, if the timing feature weight is set to 0.4, the spectral feature weight is set to 0.4, and the energy feature weight is set to 0.2, during the comparison process, if the score corresponding to the deviation degree of the timing features of a certain combination pattern from the threshold is 80 points, the spectral feature score is 70 points, and the energy feature score is 60 points, then the abnormal score of this combination pattern is 80×0.4 + 70×0.4 + 60×0.2 = 72 points. In this way, by comprehensively considering the differences between various features and the thresholds, a score that can comprehensively reflect the abnormal degree of the sound combination pattern is obtained.

[0075] S209. Sort the abnormal scores according to the time sequence of the new sound combination patterns to obtain an abnormal score sequence.

[0076] The Bluetooth headset arranges the abnormal scores of each new sound combination pattern calculated in step S208 in the order of the acquisition time of their corresponding sound combination patterns. In this way, an abnormal score sequence that changes with time is formed, and this sequence can clearly show the change of the abnormal degree of the sound combination pattern at different times. By analyzing this sequence, it can be understood how the abnormal situation develops over time, whether it gradually worsens, remains stable, or alleviates, providing a more intuitive data basis for subsequent judgment of abnormal features.

[0077] S210. Establish an abnormal development trend curve based on the abnormal score sequence, and determine the abnormal characteristics according to the change rate of the abnormal development trend curve and the corresponding abnormal score.

[0078] Based on the abnormal score sequence obtained in step S209, the Bluetooth headset plots an abnormal development trend curve with time as the horizontal axis and the abnormal score as the vertical axis. By analyzing the curve, its change rate is calculated. Among them, the change rate can be obtained by calculating the ratio of the difference in abnormal scores between adjacent time points to the time interval. For example, at two adjacent time points t1 and t2, the abnormal scores are s1 and s2 respectively, and the time interval is Δt, then the change rate v = (s2 - s1) / Δt.

[0079] Determine the abnormal characteristics according to the change rate of the abnormal development trend curve and the corresponding abnormal score. If the change rate is large and the abnormal score exceeds a certain threshold, it may indicate that the abnormal situation is deteriorating rapidly. For example, if the change rate exceeds 0.5 points / second and the abnormal score is greater than 70 points, it can be determined that an emergency abnormal characteristic appears; if the change rate is small but the abnormal score continues to be higher than the normal range, it may indicate the existence of a potential and gradually developing abnormal situation. In this way, the abnormal characteristics are comprehensively judged, and the abnormal situation in the bathroom activity is more accurately identified.

[0080] S211. Receive the user's resetting of the preset emergency plans corresponding to different danger levels.

[0081] The Bluetooth headset provides a human-computer interaction interface. For example, it is connected to a supporting application on the mobile phone via Bluetooth, or it has simple operation buttons and a display screen by itself. The user can enter the setting interface through these interfaces to reset the preset emergency plans corresponding to different danger levels. The user can, according to their own needs, choose whether to turn on the automatic alarm function, set specific contacts to be notified when alarmed, adjust the content and method of voice prompts, etc. The Bluetooth headset receives these setting information of the user and stores it in the internal memory. When the danger level is determined according to the abnormal characteristics later, the corresponding operations will be executed according to the emergency plan reset by the user, so as to achieve personalized safety protection.

[0082] The Bluetooth headset in the embodiment of the present invention is an electronic device. Figure 3 The schematic diagram of the architecture of the electronic device suitable for implementing the embodiment of the present invention is shown.

[0083] It should be noted that Figure 3 The electronic device shown is only an example and should not bring any restrictions to the functions and usage scopes of the embodiments of the present invention.

[0084] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions (computer programs), or by controlling related hardware through instructions (computer programs). The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. The electronic device of this embodiment includes a storage medium and a processor. Among them, multiple instructions are stored in the storage medium, and the instructions can be loaded by the processor to execute any step of the method provided by the embodiment of the present invention.

[0085] Specifically, the storage medium and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more signal lines. The computer execution instructions for implementing the data access control method are stored in the storage medium, including at least one software function module that can be stored in the storage medium in the form of software or firmware. The processor executes various functional applications and data processing by running the software programs and modules stored in the storage medium. The storage medium can be, but is not limited to, a random access storage medium (Random Access Memory, abbreviated as RAM), a read-only storage medium (Read Only Memory, abbreviated as ROM), a programmable read-only storage medium (Programmable Read-Only Memory, abbreviated as PROM), an erasable read-only storage medium (Erasable Programmable Read-Only Memory, abbreviated as EPROM), an electrically erasable read-only storage medium (Electric Erasable Programmable Read-Only Memory, abbreviated as EEPROM), etc. Among them, the storage medium is used to store programs, and the processor executes the programs after receiving the execution instructions.

[0086] Furthermore, the software programs and modules in the above storage medium may further include an operating system, which may include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and may communicate with various hardware or software components to provide a running environment for other software components. The processor can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, including a central processing unit (Central Processing Unit, abbreviated as CPU), a network processor (Network Processor, abbreviated as NP), etc., which can implement or execute the various methods, steps and logic flow block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0087] Since the instructions stored in the storage medium can execute the steps in any of the methods provided in the embodiments of the present invention, the beneficial effects of any of the methods provided in the embodiments of the present invention can be achieved. For details, refer to the previous embodiments and will not be elaborated here.

[0088] As described above, only the specific preferred embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A smart protection method for Bluetooth headsets based on acoustic features, applied to Bluetooth headsets, characterized in that, The method includes: Collecting the original sound signals of normal bathroom activities through the microphone array of the Bluetooth headset, where the original sound signals include three types of sounds: water flow sound, human activity sound, and voice sound; Performing hierarchical noise reduction processing on the original sound signals to obtain noise-reduced signals, and then separating and extracting the water flow sound, the human activity sound, and the voice sound from the noise-reduced signals; Extracting the corresponding temporal combination features of the water flow sound, the human activity sound, and the voice sound to obtain a sound combination pattern, where the sound combination pattern includes the duration feature of the water flow sound, the temporal correlation feature between the human activity sound and the water flow sound, and the interaction feature between the voice sound and the water flow sound and the human activity sound; Establishing an acoustic behavior model of normal bathroom activities based on the sound combination pattern, where the acoustic behavior model is used to characterize the temporal relationship of various sounds in normal bathroom activities; Matching and comparing the new sound combination pattern corresponding to the newly collected original sound signals in real time with the acoustic behavior model to determine the abnormal features corresponding to the new original sound signals; Determining the risk level according to the trigger number and duration of the abnormal features, and executing the preset emergency plan corresponding to the risk level.

2. The method according to claim 1, wherein The step of extracting the corresponding temporal combination features of the water flow sound, the human activity sound, and the voice sound to obtain a sound combination pattern specifically includes: Performing temporal segmentation on the water flow sound, the human activity sound, and the voice sound to obtain a sound event sequence with timestamps; Constructing a state transition matrix based on the sound event sequence, where the rows and columns of the state transition matrix represent the current sound state and the next moment sound state respectively, and the matrix elements represent the corresponding transition probabilities; Constructing a temporal dependence graph through the sound event sequence and the state transition matrix, where the nodes of the temporal dependence graph represent sound events, the edges represent the temporal correlation between events, and the edge weights represent the dependence intensity; Determining the typical path features according to the temporal dependence graph, where the typical path features include path length, node type distribution, and edge weight distribution; Performing weighted fusion on the transition probability features of the state transition matrix and the typical path features to obtain the sound combination pattern.

3. The method according to claim 1, wherein Before the step of matching and comparing the new sound combination pattern corresponding to the newly collected original sound signals in real time with the acoustic behavior model to determine the abnormal features corresponding to the new original sound signals, it further includes: Establishing a threshold adjustment matrix corresponding to different preset time periods, where the threshold adjustment matrix includes a daytime threshold coefficient and a nighttime threshold coefficient; Determining the noise intensity coefficient corresponding to the new original sound signals; Performing weighted fusion on the threshold adjustment matrix and the noise intensity coefficient to obtain a real-time matching threshold, where the real-time matching threshold is used to perform matching and comparison on the new sound combination pattern and the acoustic behavior model.

4. The method according to claim 3, characterized in that The step of determining the noise intensity coefficient corresponding to the new original sound signals specifically includes: Collecting multiple environmental background noise samples within a preset time window; Calculating the mean and standard deviation of the environmental background noise samples; Establishing a noise evaluation model based on the mean and the standard deviation; Input the new original sound signal into the noise evaluation model to obtain the noise intensity coefficient.

5. The method according to claim 1, wherein The step of matching and comparing the new sound combination pattern corresponding to the newly collected original sound signal with the acoustic behavior model to determine the abnormal feature corresponding to the new original sound signal specifically includes: Continuously collect a plurality of the new sound combination patterns within a preset sliding time window; Calculate the abnormal score of each of the new sound combination patterns to obtain an abnormal score sequence; Based on the abnormal score sequence, establish an abnormal development trend curve; Determine the abnormal feature according to the change rate of the abnormal development trend curve and the corresponding abnormal score.

6. The method according to claim 5, wherein The step of calculating the abnormal score of each of the new sound combination patterns to obtain an abnormal score sequence specifically includes: Extract the time series feature, frequency spectrum feature, and energy feature of each of the new sound combination patterns respectively; Compare the time series feature, frequency spectrum feature, and energy feature with the real-time matching threshold respectively, and calculate the abnormal score corresponding to each of the new sound combination patterns based on a preset feature weight; Sort the abnormal scores according to the time sequence of the new sound combination patterns to obtain an abnormal score sequence.

7. The method according to claim 1, wherein Before the step of determining the danger level according to the trigger number and duration of the abnormal feature and executing the preset emergency plan corresponding to the danger level, it further includes: Receiving the user's re-setting of the preset emergency plan corresponding to different danger levels.

8. A Bluetooth headset, characterized in that, The Bluetooth headset includes: one or more processors and a memory; The memory is coupled to the one or more processors, and the memory is used to store computer program code. The computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the Bluetooth headset to execute the method according to any one of claims 1-7.

9. A computer-readable storage medium, comprising instructions, characterized in that, When the instruction runs on the Bluetooth headset, it enables the Bluetooth headset to execute the method according to any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product runs on the Bluetooth headset, it enables the Bluetooth headset to execute the method according to any one of claims 1-7.

Images