Electric bed control method and system based on deep learning algorithm

Abstract

The invention relates to an electric bed control method and system based on a deep learning algorithm. According to the method, dynamic noise suppression processing is conducted on environment audio signals, a personalized voiceprint template is generated by combining user voiceprint feature registration and dynamic updating, snore specificity recognition is conducted on enhanced audio signals through a deep learning model, and an adjustment decision is generated in cooperation with body movement and respiration sensor data. Finally, through a progressive angle adjusting mechanism, the snore of the user and a non-target sound source are accurately distinguished in a complex noise environment, the false triggering rate is effectively reduced, self-adaptive adjustment of the posture of the bed body is achieved, and therefore the sleep breathing quality of the user is improved, and the safety and comfort of the adjusting process are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent sleep assistive device control, and in particular, relates to a control method and system for an electric bed based on a deep learning algorithm. Background Art

[0002] Existing snoring detection technology for electric beds usually relies on a single microphone to collect audio and judge snoring through threshold methods or traditional acoustic features (such as MFCC). However, in a dynamic noise environment (such as air conditioning sound, partner turning over, and environmental conversations), the spectral characteristics of snoring and non-snoring are highly overlapped, resulting in a high false trigger rate. For example, coughing or environmental noise may be misjudged as snoring, causing the bed to be incorrectly adjusted, affecting the user experience.

[0003] In addition, existing deep learning models lack the ability to adapt to users' personalized voiceprint features, making it difficult to distinguish between target users and other sound sources in scenarios where multiple people sleep together. Summary of the invention

[0004] Based on this, it is necessary to provide a control method and system for an electric bed based on a deep learning algorithm to address the above technical problems.

[0005] In a first aspect, the present application provides a control method for an electric bed based on a deep learning algorithm, comprising:

[0006] S1: collects the ambient audio signal on the electric bed, performs dynamic noise suppression on the ambient audio signal, and generates an enhanced audio signal;

[0007] S2: Register and dynamically update the user's voiceprint features based on the enhanced audio signal to generate a personalized voiceprint template;

[0008] S3: Combined with the personalized voiceprint template, the enhanced audio signal is processed for snoring-specific recognition through a deep learning model to generate a snoring probability determination result;

[0009] S4: Based on the snoring probability determination result, the bed adjustment decision instruction is generated in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data;

[0010] S5: According to the bed adjustment decision instruction, the head angle of the electric bed is adaptively adjusted.

[0011] In a second aspect, the present application also provides a control system for an electric bed based on a deep learning algorithm, comprising:

[0012] The audio signal acquisition and enhancement module is used to acquire the ambient audio signal on the electric bed, and perform dynamic noise suppression processing on the ambient audio signal to generate an enhanced audio signal;

[0013] A voiceprint feature processing module is used to register and dynamically update user voiceprint features based on enhanced audio signals to generate personalized voiceprint templates;

[0014] The snoring recognition and determination module is used to combine the personalized voiceprint template, perform snoring-specific recognition processing on the enhanced audio signal through a deep learning model, and generate a snoring probability determination result;

[0015] A bed adjustment decision generation module is used to generate a bed adjustment decision instruction based on the snoring probability determination result, in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data;

[0016] The bed adjustment module is used to adaptively adjust the head angle of the electric bed according to the bed adjustment decision instruction.

[0017] In a third aspect, the present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements a control method for an electric bed based on a deep learning algorithm as in the first aspect.

[0018] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the control method of an electric bed based on a deep learning algorithm as in the first aspect is implemented.

[0019] The above-mentioned control method and system of an electric bed based on a deep learning algorithm, through dynamic noise suppression processing of the ambient audio signal and combining the user voiceprint feature registration and dynamic update to generate a personalized voiceprint template, uses a deep learning model to perform snoring-specific recognition on the enhanced audio signal, and coordinates body motion and respiratory sensor data to generate adjustment decisions, and finally through a progressive angle adjustment mechanism, accurately distinguishes the user's snoring from non-target sound sources in a complex noise environment, effectively reduces the false trigger rate, and realizes adaptive adjustment of the bed posture, thereby improving the user's sleep breathing quality and ensuring the safety and comfort of the adjustment process. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A flowchart of a control method for an electric bed based on a deep learning algorithm provided by the present invention;

[0022] Figure 2 A schematic structural diagram of a control system for an electric bed based on a deep learning algorithm provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0024] refer to Figure 1 , which shows a flow chart of a control method for an electric bed based on a deep learning algorithm provided by the present application, the method comprising the following steps:

[0025] S1: Collect the ambient audio signal on the electric bed, and perform dynamic noise suppression processing on the ambient audio signal to generate an enhanced audio signal.

[0026] Specifically, high-sensitivity audio sensors can be installed at the head, foot, and appropriate positions on both sides of the electric bed. The sensor can capture the ambient audio signals of the electric bed in all directions, including but not limited to the user's snoring, turning over, external environmental noise, etc. The sampling frequency of the audio sensor can be set to 16kHz to ensure that audio information within different frequency ranges can be accurately collected. The collected audio signals are stored in a built-in storage module of the electric bed in digital form, and the storage format can be 16-bit PCM format.

[0027] The collected ambient audio signal is input into the preset dynamic noise suppression algorithm. The algorithm first performs time-frequency analysis on the audio signal to identify the noise components and effective signal components. Noise components usually have characteristics such as randomness and continuity, while effective signals such as snoring have specific frequency and amplitude characteristics. The algorithm uses adaptive filtering technology to adjust the parameters of the filter according to the real-time characteristics of the noise to suppress the noise. For example, in an actual scenario, when the ambient noise around the electric bed is mainly from the traffic noise outside the window, the dynamic noise suppression algorithm can monitor the frequency and amplitude changes of the traffic noise in real time, adjust the parameters of the filter, so that the filter can effectively filter out the traffic noise, thereby generating an enhanced audio signal.

[0028] S2: Register and dynamically update user voiceprint features based on enhanced audio signals to generate personalized voiceprint templates.

[0029] Specifically, when a user uses an electric bed for the first time, the system prompts the user to register the voiceprint feature. In a quiet environment, the user says a series of preset registration passwords according to the system prompts, such as "Hello, electric bed registration voiceprint". The audio sensor collects the audio of the password spoken by the user and inputs it into the voiceprint feature extraction algorithm. Based on deep learning technology, the algorithm can extract a variety of voiceprint features from the audio, including pitch, timbre, frequency response, amplitude envelope, etc. The above features can be combined into a 30-dimensional feature vector to represent the user's voiceprint features. For example, for a male user, the voiceprint feature may be manifested as a lower pitch (base frequency is approximately 85Hz-180Hz), a specific frequency peak, etc.; while for a female user, it may be manifested as a higher pitch (base frequency is approximately 165Hz-255Hz), different frequency response characteristics, etc. The extracted voiceprint feature vector is stored in the voiceprint database of the electric bed and associated with the user's account information.

[0030] As time goes by, the user's voiceprint features may change due to a variety of factors, such as hoarseness caused by a cold, voice changes caused by aging, etc. In order to ensure the accuracy of voiceprint recognition, the present invention sets up a dynamic update mechanism for voiceprint features. During the user's daily use, the user's voice input is continuously monitored. Whenever a voice is detected with a similarity exceeding a certain value with the registered voiceprint features, it is used as a new voiceprint sample for feature extraction and fused and updated with the original voiceprint feature vector. For example, if the user's voice changes due to a cold one day after registering the voiceprint, after monitoring the user's voice with cold characteristics, new voiceprint features will be extracted, the weights and parameters of the relevant features in the original voiceprint feature vector will be adjusted, and an updated personalized voiceprint template will be generated. This dynamic update mechanism enables the voiceprint template to always maintain a high degree of match with the user's current voiceprint features, thereby improving the robustness and accuracy of voiceprint recognition.

[0031] S3: Combined with the personalized voiceprint template, the enhanced audio signal is processed for snoring-specific recognition through a deep learning model to generate a snoring probability determination result.

[0032] Specifically, the enhanced audio signal after dynamic noise suppression processing is input into the deep learning model. The deep learning model can adopt a convolutional neural network (CNN) architecture and is trained with a large amount of audio data labeled with snoring. The model can learn the time-frequency characteristics, waveform patterns, etc. that are unique to snoring. In the actual recognition process, the model can analyze each 0.5-second time segment in the enhanced audio signal to determine whether the segment contains snoring features. For example, snoring usually has a lower frequency (50Hz-350Hz), periodic waveform changes (period is approximately 0.2-0.8 seconds) and specific amplitude modulation features. When the model detects an audio segment that meets the above characteristics, it determines it as a possible snoring part.

[0033] For audio clips that are determined to contain possible snoring, the model further calculates the probability value of the audio clip being snoring. The probability value is calculated based on the degree of correlation between various audio features and snoring learned by the model during the training process. For example, if the frequency characteristics of an audio clip are highly consistent with the standard frequency range of snoring, and the waveform pattern is also similar to the periodic modulation characteristics of snoring, then the model assigns it a higher snoring probability value, such as above 0.9. For some audio clips that have some features similar to snoring but other features are significantly different, the model will give a relatively low snoring probability value, such as between 0.6-0.8, based on the trade-off between similar features and different features. The snoring probability determination result is output in the form of data.

[0034] S4: Based on the snoring probability determination result, the bed adjustment decision instruction is generated in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data.

[0035] Specifically, while obtaining the snoring probability determination result, the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data are coordinated. The body motion sensor can be installed under the mattress of the electric bed, and can monitor the user's body movement in real time, such as the number of turns over, the amplitude of movement, etc. The respiratory monitoring sensor can use non-contact respiratory monitoring technology, such as a radar-based respiratory monitoring sensor, which can be installed near the user's chest or abdomen to monitor the user's respiratory frequency and breathing depth. The sensor data is input into the bed adjustment decision algorithm together with the snoring probability determination result.

[0036] The bed adjustment decision algorithm performs a comprehensive analysis based on the input multi-source data to generate reasonable bed adjustment decision instructions. For example, if the snoring probability judgment result is continuously high (such as the snoring probability of multiple consecutive time segments is above 0.8), and the body motion sensor detects that the user frequently turns over (such as turning over more than 10 times in one hour), and the respiratory monitoring sensor shows that the user's respiratory rate is faster than the normal value (such as more than 20 times / minute), the algorithm determines that the user may have difficulty breathing due to poor sleeping posture, which in turn causes snoring. At this time, the algorithm will generate corresponding bed adjustment decision instructions based on the user's current body position and respiratory status, such as adjusting the angle of the head of the bed to a suitable angle that can relieve breathing difficulties and promote the user to lie on his side. For another example, if the snoring probability judgment result is low, but the body motion sensor detects that the user maintains a fixed and possibly uncomfortable posture for a long time, the algorithm can generate instructions to slightly adjust the angle of the head of the bed or adjust the local softness and hardness of the mattress to improve the user's comfort.

[0037] S5: According to the bed adjustment decision instruction, the head angle of the electric bed is adaptively adjusted.

[0038] Specifically, the head of the electric bed is equipped with a motor-driven adjustment mechanism, which can accurately control the adjustment of the head angle according to the bed adjustment decision instruction. The motor can be a stepper motor, which can achieve high-precision angle control, and its control accuracy can reach 0.9 degrees / step. The adjustment mechanism can be mechanically connected to the bed frame of the electric bed.

[0039] According to the bed adjustment decision instruction, the motor drives the bed head to adjust at an appropriate rate and angle. For example, when the instruction requires the bed head angle to be adjusted to 30 degrees, the motor first calculates the number of steps required to rotate based on the difference between the current bed head angle and the target angle. Then, the motor adjusts at a relatively gentle rate (such as 5 degrees per second) to avoid user discomfort caused by too fast adjustment. During the adjustment process, the system will also monitor the user's body movement and heart rate and other physiological signals in real time. If it is detected that the user has an abnormal reaction due to the adjustment (such as a sudden increase in heart rate by more than 10%, an abnormal increase in body movement amplitude, etc.), the adjustment will be suspended or the adjustment strategy will be adjusted. For example, if the user struggles due to discomfort during the adjustment of the bed head angle, the system will suspend the adjustment and decide whether to continue adjusting or adjust to a more suitable intermediate angle based on the re-evaluated user status and monitoring data. Through the above-mentioned adaptive adjustment strategy, the electric bed can safely and comfortably adjust the bed head angle according to the user's real-time status and needs, and improve the user's sleeping environment.

[0040] The above-mentioned control method of an electric bed based on a deep learning algorithm generates a personalized voiceprint template by dynamically suppressing the ambient audio signal and combining the user's voiceprint feature registration and dynamic update. The deep learning model is used to perform snoring-specific recognition on the enhanced audio signal, and the adjustment decision is generated in coordination with the body motion and breathing sensor data. Finally, through the progressive angle adjustment mechanism, the user's snoring and non-target sound sources can be accurately distinguished in a complex noise environment, the false trigger rate can be effectively reduced, and the adaptive adjustment of the bed posture can be achieved, thereby improving the user's sleep breathing quality and ensuring the safety and comfort of the adjustment process.

[0041] In an optional embodiment, S1 includes the following steps:

[0042] S11: The dual-channel audio signals on the electric bed are collected through dual microphones as environmental audio signals; the dual-channel audio signals are processed for sound source localization based on the generalized sidelobe cancellation algorithm to generate a mainlobe beamforming signal.

[0043] Specifically, in order to capture the audio information on the electric bed more comprehensively and accurately, a high-performance microphone can be installed at the head of each side of the electric bed to form a dual-channel audio acquisition system. The two microphones can simultaneously collect environmental audio signals, including the user's snoring, turning over, external environmental noise and other sounds. The collected dual-channel audio signal is used as the input source of the environmental audio signal, and the sound source is localized and processed through the generalized sidelobe cancellation algorithm to generate a mainlobe beamforming signal. Specifically, the generalized sidelobe cancellation algorithm can effectively suppress the interference of non-target sound sources, highlight the direction of the main sound source, and make the mainlobe beamforming signal more concentrated in the area where the user may make sound, thereby improving the pertinence and accuracy of subsequent audio processing.

[0044] S12: performing noise spectrum estimation processing on the main lobe beamforming signal to obtain an estimated noise spectrum, and performing dynamic compensation processing on the estimated noise spectrum using a minimum mean square error criterion to generate a preliminary noise reduction signal.

[0045] Specifically, for the generated main lobe beamforming signal, the noise spectrum is estimated using a method based on power spectrum density estimation to obtain an estimated noise spectrum. In this process, the frequency domain characteristics of the signal are analyzed and combined with statistical signal processing technology to accurately estimate the spectral information of the ambient noise. Then, the minimum mean square error criterion is used to dynamically compensate the estimated noise spectrum to generate a preliminary noise reduction signal. In this process, the minimum mean square error criterion minimizes the mean square error between the noise-reduced signal and the original clean signal (noise-free signal), thereby ensuring that the preliminary noise reduction signal after noise reduction removes noise components as much as possible while retaining the main features of the speech, thereby further improving the purity and recognizability of the audio signal.

[0046] S13: Perform harmonic reconstruction processing on the preliminary noise reduction signal to generate an enhanced audio signal.

[0047] Specifically, the preliminary noise reduction signal is subjected to harmonic reconstruction. Since effective audio signals such as snoring usually have relatively rich harmonic components, while the harmonic characteristics of noise signals are relatively weak or present different harmonic structures. By analyzing the harmonic information in the preliminary noise reduction signal, the original harmonics are strengthened and supplemented using harmonic enhancement technology and signal synthesis methods to generate an enhanced audio signal. The harmonic reconstruction process can not only effectively enhance the effective components of the audio signal, but also further weaken the residual noise impact, so that the enhanced audio signal more clearly and accurately reflects the user's real audio characteristics.

[0048] In an optional embodiment, S2 includes the following steps:

[0049] S21: When the user uses it for the first time, the voiceprint registration is triggered by a preset guiding voice instruction, the preset continuous sound signal emitted by the user is recorded, and the resonance peak and fundamental frequency trajectory characteristics of the preset continuous sound signal are extracted.

[0050] Specifically, when the user uses the electric bed for the first time, the voiceprint registration program is automatically started. The preset guidance voice instructions are played through the speaker, such as "Please continue to make the sound of 'ah' within three seconds for voiceprint registration", and the corresponding text prompts are displayed on the display screen to ensure that the user is clearly aware of the operation steps. After the user makes the sound according to the guidance, the audio sensor will accurately record the preset continuous sound signal of this three-second period. In addition, the system will automatically perform a preliminary quality test on the recorded audio. If it is detected that the audio signal is too weak (such as energy lower than the preset threshold of -40dBm) or there is obvious noise interference (such as signal-to-noise ratio lower than 10dB), the user will be prompted to re-record.

[0051] For the recorded preset continuous sound signal, an audio feature extraction algorithm based on wavelet transform is used for processing. Wavelet transform can decompose the audio signal into components of different frequency bands, and accurately extract the resonance peaks and fundamental frequency trajectory characteristics in the signal by analyzing the amplitude and frequency changes of the above components. The resonance peak reflects the spectral envelope of the audio signal, which mainly reflects the vocal tract characteristics of the sound-producing body; the fundamental frequency trajectory describes the change law of the fundamental frequency over time, reflecting the pitch change of the pronunciation. For example, for a user who makes a standard "ah" sound, the resonance peak usually appears in a specific frequency range (such as the first resonance peak at 500Hz-1000Hz, and the second resonance peak at 1200Hz-2500Hz), while the fundamental frequency trajectory presents a relatively stable curve with a frequency fluctuating between 100Hz-300Hz.

[0052] S22: Standardize the fundamental frequency trajectory feature to generate a standardized fundamental frequency envelope, and combine the resonance peak and the standardized fundamental frequency envelope into an initial voiceprint template.

[0053] Specifically, the extracted fundamental frequency trajectory features are standardized. The standardization method can adopt z-score standardization, that is, first calculate the mean and standard deviation of the fundamental frequency trajectory, and then transform each fundamental frequency value so that its mean is 0 and the standard deviation is 1. In this way, the differences in fundamental frequency levels of different users and the influence of environmental factors on the fundamental frequency can be eliminated, and a standardized fundamental frequency envelope with a uniform scale can be generated. Then, the extracted resonance peak features are combined with the standardized fundamental frequency envelope to form an initial voiceprint template. The initial voiceprint template is a multi-dimensional feature vector that integrates the user's vocal tract characteristics and pitch change characteristics, which can more comprehensively characterize the user's voiceprint characteristics. For example, the dimension of the initial voiceprint template can be 50 dimensions, of which the first 20 dimensions represent the relevant characteristics of the resonance peak, and the last 30 dimensions represent the standardized fundamental frequency envelope characteristics and their changing trends.

[0054] S23: After valid snoring is detected, the voiceprint features of the current audio clip are extracted. If the matching degree between the voiceprint features and the initial voiceprint template exceeds a preset threshold, the initial voiceprint template is updated through a dynamic update algorithm to generate a personalized voiceprint template.

[0055] Specifically, after valid snoring is detected (such as the snoring probability judgment result is greater than 0.8 and the duration exceeds 10 seconds), the system will immediately extract the voiceprint features of the current audio segment containing the snoring. Use the same feature extraction algorithm as during registration to extract the resonance peak and fundamental frequency trajectory features of the audio segment, and also standardize the fundamental frequency trajectory features to generate a voiceprint feature vector to be matched. Subsequently, calculate the similarity between the voiceprint feature vector to be matched and the initial voiceprint template, and use cosine similarity as a measure of matching. The calculation formula for cosine similarity is:

[0056]

[0057] Among them, A and B are the voiceprint feature vector to be matched and the initial voiceprint template vector respectively. If the calculated similarity exceeds the preset threshold (such as 0.75), it is considered that the voiceprint feature of the current audio clip is sufficiently matched with the initial voiceprint template, indicating that the snoring sound is likely to be emitted by the registered user, and then the initial voiceprint template is updated through the dynamic update algorithm to generate a personalized voiceprint template. The dynamic update algorithm adopts the exponentially weighted moving average method, and its formula is:

[0058] T new =αT old +(1-α)F current ;

[0059] Among them, Tnew It is the updated voiceprint template. old is the initial voiceprint template before the update, F current is the voiceprint feature vector of the current successful match, and α is the smoothing factor, which can be set to 0.9. This method can make the voiceprint template gradually adapt to the slight changes in the user's voiceprint characteristics, while maintaining the stability of the template, ensuring the accuracy and personalization of subsequent voiceprint recognition.

[0060] In an optional embodiment, S3 includes the following steps:

[0061] S31: performing frame-by-frame windowing processing on the enhanced audio signal, and extracting Mel-frequency cepstral coefficients and dynamic difference coefficients of each frame.

[0062] Specifically, the enhanced audio signal after dynamic noise suppression is used as input. In order to better extract audio features, the enhanced audio signal is first subjected to frame windowing. When framing, the frame length can be set to 25 milliseconds, the frame shift can be 10 milliseconds, and the Hamming window can be used as the window function. This method can reduce inter-frame interference and improve the accuracy of frequency domain analysis. Then, the Mel frequency cepstral coefficient (MFCC) is extracted for each frame of audio signal, and the order of the extracted MFCC coefficient is 13, which can effectively characterize the spectral characteristics of the audio signal. At the same time, the dynamic differential coefficient of each frame of audio is calculated, including the first-order differential coefficient and the second-order differential coefficient, to capture the dynamic characteristics of the audio signal over time. For example, for an enhanced audio signal with a duration of 5 seconds, about 500 frames of audio data will be obtained after frame windowing. 13 MFCC coefficients and corresponding dynamic differential coefficients are extracted from each frame of data to form a feature matrix with the dimension of the number of frames × coefficients.

[0063] S32: Input the Mel-frequency cepstral coefficients and dynamic differential coefficients into the pre-trained lightweight convolutional recurrent network and output a high-dimensional feature vector.

[0064] Specifically, the extracted Mel-frequency cepstral coefficients and dynamic differential coefficients are input into a pre-trained lightweight convolutional recurrent network. This network combines the advantages of convolutional neural networks (CNN) and recurrent neural networks (RNN), and can not only extract local spatial features of audio signals, but also capture dynamic changes in time series. The convolution layer of the network can use a 3×3 convolution kernel with a step size of 1. By performing a convolution operation on the feature matrix, a higher-level feature representation is extracted. Subsequently, the recurrent layer adopts a long short-term memory network (LSTM) structure, which can effectively process long-distance dependencies in sequence data. After multiple layers of convolution and recurrent operations, the network outputs a high-dimensional feature vector with a dimension of 256. This high-dimensional feature vector integrates the time-frequency features and dynamic change information of the audio signal, and is a highly generalized and abstract representation of the potential snoring features in the enhanced audio signal.

[0065] S33: Fusing the high-dimensional feature vector with the matching score of the personalized voiceprint template, and outputting a snoring probability determination result.

[0066] Specifically, the generated high-dimensional feature vector is fused with the matching score of the personalized voiceprint template. The personalized voiceprint template contains the user's unique voiceprint feature information. By calculating the similarity between the high-dimensional feature vector and the voiceprint template, cosine similarity can be used as a calculation method for the matching score to obtain a value reflecting the degree of matching between the two. Then, this matching score is compared with a preset threshold. When the matching score exceeds the threshold (for example, 0.7), it indicates that the current audio signal has a high similarity with the user's voiceprint feature. Combined with the snoring feature information contained in the high-dimensional feature vector, it is comprehensively judged that the probability of the audio clip being snoring is high, thereby outputting a higher snoring probability judgment result (for example, between 0.8-1). Conversely, if the matching score is lower than the threshold, a lower snoring probability judgment result is output. The above-mentioned fusion processing method makes full use of the feature information of the audio signal and the personalized voiceprint information of the user, and can more accurately identify snoring.

[0067] In an optional embodiment, S5 includes the following steps:

[0068] S51: According to the type of bed adjustment decision instruction, the angle of the bed head is dynamically adjusted at a preset rate, and the snoring probability determination result is periodically detected during the adjustment process.

[0069] Specifically, according to the type of bed adjustment decision instruction, such as adjusting the head of the bed to a specific angle (for example, 30 degrees) or fine-tuning the angle (such as adjusting 5 degrees each time). The motor dynamically adjusts the head of the bed at a preset rate (for example, 5 degrees per second). During the adjustment process, the system periodically detects the snoring probability determination result every 2 seconds. In this way, the changes in the user's snoring after adjusting the angle can be understood in real time, so as to adjust the subsequent adjustment strategy in time. For example, if the snoring probability determination result continues to decrease during the process of adjusting the head of the bed angle, it means that the current angle adjustment is effective in alleviating snoring, and you can continue to adjust in this direction; conversely, if the snoring probability determination result rises or fluctuates greatly, you need to re-evaluate the adjustment direction and amplitude.

[0070] S52: Dynamically limit the angle adjustment range based on the spinal pressure distribution model, and trigger the protection mechanism when abnormal pressure is detected.

[0071] Specifically, in the process of adjusting the angle of the head of the bed, the system dynamically limits the angle adjustment range based on the spinal pressure distribution model. The spinal pressure distribution model is established based on ergonomic principles, taking into account the changes in spinal pressure at different angles. When the angle of the head of the bed is adjusted to a position that may cause abnormal spinal pressure (for example, the pressure exceeds the preset threshold, such as 1.5 times the normal spinal pressure), the system will automatically trigger a protection mechanism to limit further adjustment of the angle. For example, when the angle of the head of the bed is adjusted to 45 degrees, the spinal pressure distribution model calculates that the spinal pressure may exceed the safe range. The system will suspend the angle adjustment and issue a prompt to remind the user that the current angle may be uncomfortable. At the same time, the angle will be adjusted back to a safe range (such as 30 degrees) to ensure the user's spinal health.

[0072] S53: When the snoring probability determination result drops below a preset threshold, the current angle is recorded as the optimal suppression angle, and the optimal suppression angle is preferentially used in subsequent adjustments.

[0073] Specifically, in the process of continuously monitoring the snoring probability determination results, when it is detected that the snoring probability determination result drops below a preset threshold (for example, from 0.7 to below 0.3), the system records the current head of the bed angle as the optimal suppression angle. The optimal suppression angle is stored in the user's personal data of the system and is given priority in subsequent adjustment processes. For example, if during a certain use, when the head of the bed is adjusted to 25 degrees, the snoring probability determination result drops significantly and remains at a low level, the system will record 25 degrees as the optimal suppression angle. In subsequent use, when the system detects a similar situation, it will give priority to adjusting the head of the bed to 25 degrees to quickly and effectively suppress snoring and improve the user's sleep comfort.

[0074] The above-mentioned control method of an electric bed based on a deep learning algorithm eliminates environmental interference through multi-microphone array beamforming and dynamic noise suppression technology, builds a personalized recognition benchmark in combination with dynamic registration and update of user voiceprint features, uses a lightweight convolutional recurrent network to fuse acoustic features and voiceprint matching to achieve accurate snoring recognition, performs multimodal collaborative decision-making through body motion monitoring and respiratory rhythm verification, adopts progressive angle adjustment combined with spinal pressure feedback to dynamically optimize the lifting strategy, and introduces abnormal load detection and bed-leaving sleep mechanism, ultimately achieving high-specificity snoring recognition and low false trigger rate in a complex acoustic environment, significantly improving respiratory disorders through safe and adaptive bed adjustment, while ensuring user comfort and system reliability, and taking into account both personalized adaptation and long-term use stability.

[0075] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0076] Based on the same inventive concept, the embodiment of the present application also provides a system for implementing the control method of the electric bed based on the deep learning algorithm involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in the control system embodiments of one or more electric beds based on the deep learning algorithm provided below can be referred to the limitations of the control method of an electric bed based on the deep learning algorithm above, and will not be repeated here.

[0077] In an exemplary embodiment, Figure 2 As shown, a control system 20 for an electric bed based on a deep learning algorithm is provided, comprising:

[0078] The audio signal acquisition and enhancement module 21 is used to acquire the ambient audio signal on the electric bed, and perform dynamic noise suppression processing on the ambient audio signal to generate an enhanced audio signal.

[0079] The voiceprint feature processing module 22 is used to register and dynamically update the user's voiceprint features based on the enhanced audio signal to generate a personalized voiceprint template.

[0080] The snoring recognition and determination module 23 is used to combine the personalized voiceprint template, perform snoring-specific recognition processing on the enhanced audio signal through a deep learning model, and generate a snoring probability determination result.

[0081] The bed adjustment decision generating module 24 is used to generate a bed adjustment decision instruction based on the snoring probability determination result, in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data.

[0082] The bed adjustment module 25 is used to adaptively adjust the head angle of the electric bed according to the bed adjustment decision instruction.

[0083] Optionally, the audio signal collection and enhancement module 21 includes:

[0084] The dual-channel audio collection and beamforming unit 211 is used to collect dual-channel audio signals on the electric bed through dual microphones as environmental audio signals; perform sound source localization processing on the dual-channel audio signals based on a generalized sidelobe cancellation algorithm to generate a mainlobe beamforming signal.

[0085] The noise compensation and preliminary noise reduction unit 212 is used to perform noise spectrum estimation processing on the main lobe beamforming signal to obtain an estimated noise spectrum, and perform dynamic compensation processing on the estimated noise spectrum using a minimum mean square error criterion to generate a preliminary noise reduction signal.

[0086] The harmonic reconstruction and audio enhancement unit 213 is used to perform harmonic reconstruction processing on the preliminary noise reduction signal to generate an enhanced audio signal.

[0087] Optionally, the voiceprint feature processing module 22 includes:

[0088] The voiceprint registration feature extraction unit 221 is used to trigger voiceprint registration through a preset guiding voice instruction when the user uses it for the first time, record the preset continuous sound signal emitted by the user, and extract the resonance peak and fundamental frequency trajectory characteristics of the preset continuous sound signal.

[0089] The fundamental frequency standardization and template initialization unit 222 is used to standardize the fundamental frequency trajectory feature, generate a standardized fundamental frequency envelope, and combine the resonance peak and the standardized fundamental frequency envelope into an initial voiceprint template.

[0090] The voiceprint dynamic updating unit 223 is used to extract the voiceprint features of the current audio segment after detecting effective snoring. If the matching degree between the voiceprint features and the initial voiceprint template exceeds a preset threshold, the initial voiceprint template is updated through a dynamic updating algorithm to generate a personalized voiceprint template.

[0091] Optionally, the snoring recognition and determination module 23 includes:

[0092] The frame windowing and feature extraction unit 231 is used to perform frame windowing processing on the enhanced audio signal to extract the Mel-frequency cepstral coefficients and dynamic difference coefficients of each frame.

[0093] The high-dimensional feature mapping unit 232 is used to input the Mel-frequency cepstral coefficients and the dynamic differential coefficients into a pre-trained lightweight convolutional recurrent network and output a high-dimensional feature vector.

[0094] The matching fusion and determination output unit 233 is used to fuse the high-dimensional feature vector with the matching score of the personalized voiceprint template and output a snoring probability determination result.

[0095] Optionally, the bed adjustment module 25 includes:

[0096] The dynamic adjustment and monitoring unit 251 is used to dynamically adjust the angle of the bed head at a preset rate according to the type of the bed adjustment decision instruction, and periodically detect the snoring probability determination result during the adjustment process.

[0097] The spinal pressure protection and angle limitation unit 252 is used to dynamically limit the angle adjustment range based on the spinal pressure distribution model and trigger the protection mechanism when abnormal pressure is detected.

[0098] The optimal suppression angle recording and calling unit 253 is used to record the current angle as the optimal suppression angle when the snoring probability determination result drops below a preset threshold, and give priority to using the optimal suppression angle in subsequent adjustments.

[0099] An embodiment of the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0100] The embodiments of the present application further provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0101] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0102] The above-mentioned embodiments only express several implementation methods of the embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the embodiments of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the embodiments of the present application, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A control method for an electric bed based on a deep learning algorithm, characterized in that: The method comprises: S1: collecting the ambient audio signal on the electric bed, and performing dynamic noise suppression processing on the ambient audio signal to generate an enhanced audio signal; S2: Registering and dynamically updating the user's voiceprint features based on the enhanced audio signal to generate a personalized voiceprint template; S3: combining the personalized voiceprint template, performing snoring-specific recognition processing on the enhanced audio signal through a deep learning model to generate a snoring probability determination result; S4: Based on the snoring probability determination result, in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data, a bed adjustment decision instruction is generated; S5: According to the bed adjustment decision instruction, the head angle of the electric bed is adaptively adjusted.

2. The method according to claim 1, characterized in that The S1 includes: S11: collecting dual-channel audio signals on the electric bed through dual microphones as the environmental audio signals; performing sound source localization processing on the dual-channel audio signals based on a generalized sidelobe cancellation algorithm to generate a mainlobe beamforming signal; S12: performing noise spectrum estimation processing on the main lobe beamforming signal to obtain an estimated noise spectrum, and performing dynamic compensation processing on the estimated noise spectrum using a minimum mean square error criterion to generate a preliminary noise reduction signal; S13: Perform harmonic reconstruction processing on the preliminary noise reduction signal to generate the enhanced audio signal.

3. The method according to claim 1, characterized in that The S2 includes: S21: When the user uses the device for the first time, trigger voiceprint registration through a preset guiding voice command, record a preset continuous sound signal emitted by the user, and extract the formant and fundamental frequency trajectory characteristics of the preset continuous sound signal; S22: performing standardization processing on the fundamental frequency trajectory feature to generate a standardized fundamental frequency envelope, and combining the resonance peak and the standardized fundamental frequency envelope into an initial voiceprint template; S23: After valid snoring is detected, the voiceprint features of the current audio segment are extracted. If the matching degree between the voiceprint features and the initial voiceprint template exceeds a preset threshold, the initial voiceprint template is updated through a dynamic update algorithm to generate the personalized voiceprint template.

4. The method according to claim 1, characterized in that: The S3 includes: S31: performing frame-by-frame windowing processing on the enhanced audio signal to extract Mel-frequency cepstral coefficients and dynamic difference coefficients of each frame; S32: Inputting the Mel-frequency cepstral coefficient and the dynamic differential coefficient into a pre-trained lightweight convolutional recurrent network, and outputting a high-dimensional feature vector; S33: fusing the high-dimensional feature vector with the matching score of the personalized voiceprint template, and outputting the snoring probability determination result.

5. The method according to any one of claims 1 to 4, characterized in that: The S5 includes: S51: dynamically adjusting the angle of the bed head at a preset rate according to the type of the bed adjustment decision instruction, and periodically detecting the snoring probability determination result during the adjustment process; S52: Dynamically limit the angle adjustment range based on the spinal pressure distribution model, and trigger the protection mechanism when abnormal pressure is detected; S53: When the snoring probability determination result drops below a preset threshold, the current angle is recorded as the optimal suppression angle, and the optimal suppression angle is preferentially used in subsequent adjustments.

6. A control system for an electric bed based on a deep learning algorithm, characterized in that: The system comprises: An audio signal acquisition and enhancement module is used to acquire the ambient audio signal on the electric bed, and perform dynamic noise suppression processing on the ambient audio signal to generate an enhanced audio signal; A voiceprint feature processing module, used to register and dynamically update the user's voiceprint features based on the enhanced audio signal to generate a personalized voiceprint template; A snoring recognition and determination module, used to combine the personalized voiceprint template, perform snoring-specific recognition processing on the enhanced audio signal through a deep learning model, and generate a snoring probability determination result; A bed adjustment decision generating module, used to generate a bed adjustment decision instruction based on the snoring probability determination result, in coordination with the body motion sensor monitoring data and the respiratory monitoring sensor monitoring data; The bed adjustment module is used to adaptively adjust the head angle of the electric bed according to the bed adjustment decision instruction.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images