Microphone-based wind speed detection and energy consumption compensation method, device and program product

Abstract

The application provides a microphone-based wind speed detection and energy consumption compensation method, device and program product, and relates to the technical field of wearable devices or the technical field of acoustic signal processing. The method comprises: collecting environmental acoustic signals through double-sided microphones built in a wearable device; extracting features of the environmental acoustic signals to obtain feature parameters related to wind speed; the feature parameters at least include power spectral density, spectral flatness of the environmental acoustic signals and signal intensity difference of the double-sided microphone signals; inputting the feature parameters into a deep learning model to output a wind speed prediction value, and correcting a motion energy consumption calculation result based on the wind speed prediction value, the deep learning model being a nonlinear mapping model based on a neural network algorithm. The method is used to improve the accuracy of wind speed detection and energy consumption calculation.

Description

Technical Field

[0001] This application relates to the field of wearable device technology or acoustic signal processing technology, and in particular to a microphone-based wind speed detection and energy consumption compensation method, device and program product. Background Technology

[0002] In scenarios requiring real-time monitoring of exercise energy consumption, such as running, cycling, and skateboarding, wearable devices need to calculate calorie expenditure based on complex and ever-changing environmental wind fields and the exerciser's movement parameters. However, aerodynamic drag can affect the energy consumption calculation set of wearable devices depending on the exercise scenario. Therefore, how to perform wind speed detection and energy consumption compensation based on microphones is crucial.

[0003] Currently, existing methods for calculating exercise energy consumption mainly rely on the classical metabolic equivalent model. This method primarily calculates calorie expenditure based on heart rate, movement speed, altitude changes, and the user's basic physiological parameters (height, weight, age, and gender). However, existing methods for calculating exercise energy consumption do not consider aerodynamic drag, resulting in poor accuracy in wind speed detection and energy consumption calculation. Summary of the Invention

[0004] This application provides a microphone-based wind speed detection and energy consumption compensation method, device, and program product to improve the accuracy of wind speed detection and energy consumption calculation.

[0005] In a first aspect, this application provides a microphone-based method for wind speed detection and energy consumption compensation, comprising:

[0006] Ambient acoustic signals are collected using dual microphones built into the wearable device;

[0007] Feature extraction is performed on the ambient acoustic signal to obtain wind speed-related feature parameters; the feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the two microphone signals of the ambient acoustic signal;

[0008] The feature parameters are input into a deep learning model, which outputs a wind speed prediction value. The calculation result of motion energy consumption is then corrected based on the wind speed prediction value. The deep learning model is a nonlinear mapping model based on a neural network algorithm.

[0009] In one possible implementation, the deep learning model includes a data acquisition layer, a feature extraction layer, a deep learning modeling layer, a model optimization and adaptation layer, a calibration and control layer, and a wind direction and speed determination and output layer.

[0010] The data acquisition layer is used to cover multiple wind speeds, all-angle wind directions, and various common motion postures in a simulated experimental environment, collect wind noise signals, and real-value data of measured wind speed and direction, and construct a training dataset.

[0011] The feature extraction layer is used to extract monophonic wind noise features and binaural spatial features based on the training dataset, and output wind noise correlation features.

[0012] The deep learning modeling layer is used to build a nonlinear mapping relationship based on the wind noise correlation features, preset a unified wind direction angle orientation definition rule, and output the initial prediction result; the initial prediction result includes the initial wind speed value and the initial wind direction angle data.

[0013] The model optimization and adaptation layer is used to adapt the initial prediction results to the equipment structural attributes and establish a differentiated model parameter library; the differentiated model parameter library includes the initial wind speed value, the initial wind direction angle data, equipment attitude sensing data, and motion positioning speed data;

[0014] The calibration and control layer is used to calibrate the environmental reference wind speed, and, in conjunction with the differentiated model parameter library, correct the model parameters of the wind speed vector deviation during the motion process, and output the corrected wind speed and direction correction data.

[0015] The wind direction and wind speed determination output layer is used to combine spatial features with the corrected wind speed and wind direction correction data to determine the wind direction category determination result and the wind speed prediction value.

[0016] In one possible implementation, before performing feature extraction on the environmental acoustic signal to obtain wind speed-related feature parameters, the process includes:

[0017] Calculate the complex coherence coefficient of the dual-microphone signals, and filter out non-wind noise interference based on the complex coherence coefficient.

[0018] In one possible implementation, the feature extraction of the environmental acoustic signal includes at least one of the following:

[0019] Adjust the cutoff frequency of the Butterworth low-pass filter;

[0020] Adjust the window length for the detrending algorithm.

[0021] In one possible implementation, the step of inputting the feature parameters into the deep learning model includes:

[0022] Calculate the deviation rate of the predicted wind speeds on both sides, and determine whether microphone hole obstruction has occurred based on the deviation rate.

[0023] In one possible implementation, the step of calculating the deviation rate of the two-sided wind speed predictions includes:

[0024] The input weights of the deep learning model are adjusted based on the deviation rate to correct the wind speed prediction under occlusion conditions.

[0025] In one possible implementation, the method further includes:

[0026] Air resistance is calculated based on wind speed forecasts, and exercise calorie consumption is adjusted by combining the movement speed and heart rate data of the exercise equipment.

[0027] The parameters of the deep learning model are corrected based on the deviation between the moving speed of the motion device and the predicted wind speed.

[0028] Secondly, this application provides a microphone-based wind speed detection and energy consumption compensation device, comprising:

[0029] The acquisition module is used to acquire ambient acoustic signals through the dual microphones built into the wearable device;

[0030] The extraction module is used to extract features from the ambient acoustic signal to obtain feature parameters related to wind speed; the feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the two microphone signals of the ambient acoustic signal.

[0031] The output module is used to input the feature parameters into the deep learning model, output the wind speed prediction value, and correct the motion energy consumption calculation result based on the wind speed prediction value. The deep learning model is a nonlinear mapping model based on a neural network algorithm.

[0032] Thirdly, this application provides an electronic device, including: a memory and a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, causing the processor to perform various possible embodiments as described in the first aspect above.

[0033] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement various possible embodiments of the first aspect above.

[0034] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements various possible implementations of the first aspect described above.

[0035] This application provides a microphone-based wind speed detection and energy consumption compensation method, apparatus, and program product. It collects environmental acoustic signals using dual microphones built into a wearable device, providing data for subsequent feature extraction. Feature extraction is performed on the environmental acoustic signals to obtain wind speed-related feature parameters, eliminating redundant interference and improving the accuracy of wind speed detection. These feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the dual microphone signals of the environmental acoustic signal. These feature parameters are input into a deep learning model, which outputs a predicted wind speed value. The calculated motion energy consumption is then corrected based on this predicted wind speed value, improving the accuracy of energy consumption calculation. The deep learning model is a nonlinear mapping model based on a neural network algorithm. Attached Figure Description

[0036] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0037] Figure 1 A flowchart illustrating a microphone-based wind speed detection and energy consumption compensation method provided in this application;

[0038] Figure 2 A flowchart illustrating another microphone-based wind speed detection and energy consumption compensation method provided in this application;

[0039] Figure 3 A schematic diagram of the architecture of a microphone-based wind speed detection and energy consumption compensation system provided for this application;

[0040] Figure 4 A flowchart illustrating a wind speed detection and energy consumption compensation system method provided in this application;

[0041] Figure 5 A schematic diagram of a microphone-based wind speed detection and energy consumption compensation device provided in this application;

[0042] Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in this application.

[0043] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0044] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0045] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0046] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0047] First, some of the terms used in this invention will be explained:

[0048] Global Positioning System (GPS) speed: refers to the real-time speed of a moving object calculated using GPS technology.

[0049] In scenarios requiring real-time monitoring of exercise energy consumption, such as running, cycling, and skateboarding, wearable devices need to calculate calorie expenditure based on complex and ever-changing environmental wind fields and the exerciser's movement parameters. However, aerodynamic drag can affect the energy consumption calculation set of wearable devices depending on the exercise scenario. Therefore, how to perform wind speed detection and energy consumption compensation based on microphones is crucial.

[0050] Currently, existing methods for calculating exercise energy consumption mainly rely on the classical metabolic equivalent model. This method primarily calculates calorie expenditure based on heart rate, movement speed, altitude changes, and the user's basic physiological parameters (height, weight, age, and gender). However, existing methods for calculating exercise energy consumption do not consider aerodynamic drag, resulting in poor accuracy in wind speed detection and energy consumption calculation.

[0051] In view of the above-mentioned problems of existing motion energy consumption calculation methods, this application proposes a method that uses dual external microphones, combined with acoustic signal processing and deep learning models, to sense local wind fields and detect wind speed, and then dynamically corrects motion energy consumption calculation by fusing motion state parameters. This method can improve the accuracy of wind speed detection and energy consumption calculation.

[0052] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments.

[0053] The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0054] The entity executing this microphone-based wind speed detection and energy consumption compensation method can be, for example, a detection system. Optionally, the detection system can be any existing electronic device with processing capabilities, such as a terminal or a server. In some embodiments, the detection system can also be deployed in a server cluster or cloud environment. This application does not limit the deployment environment of the detection system.

[0055] Figure 1 This is a flowchart illustrating a microphone-based wind speed detection and energy consumption compensation method provided in this application. Figure 1 As shown, the method includes:

[0056] The S101 collects ambient acoustic signals through dual microphones built into the wearable device.

[0057] For example, a wearable device can be a user-wearable electronic device, such as a true wireless stereo (TWS) headset, a smartwatch, or a mobile phone. For example, an ambient acoustic signal can be an acoustic wave signal generated in the environment by airflow, noise, etc.

[0058] Optionally, the detection system can have dual microphones arranged symmetrically on the left and right, or staggered front and back on the dual microphones. Alternatively, it can use one microphone exposed to collect mid-to-far field ambient sound, while the other microphone built-in to collect near-field airflow sound, thus picking up ambient acoustic signals at different distances in layers.

[0059] Optionally, the detection system can synchronously acquire ambient acoustic signals by setting the same sampling rate, the same pickup gain, and the same filtering frequency band for both microphones. It can also acquire low-frequency acoustic signals related to airflow through one microphone and high-frequency ambient noise signals through the other microphone.

[0060] Optionally, the detection system can continuously collect ambient acoustic signals using dual microphones. The system can also detect changes in ambient airflow intensity and sudden changes in ambient noise loudness, automatically activating the dual microphones for data acquisition. Furthermore, the system can receive user commands via touch, voice, or software to activate the dual microphones and begin acquiring ambient acoustic signals.

[0061] S102, extract features from the ambient acoustic signal to obtain wind speed-related feature parameters; these feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the two microphone signals of the ambient acoustic signal.

[0062] For example, power spectral density can be the density of the energy distribution of an ambient acoustic signal in the frequency domain, used to characterize the strength of wind noise.

[0063] For example, spectral flatness can be the uniformity of the spectral distribution of an ambient acoustic signal, used to distinguish between wind noise and ambient noise.

[0064] For example, the signal strength difference can be used to characterize the wind direction by collecting the amplitude difference of the ambient acoustic signal from both microphones.

[0065] Optionally, the detection system can perform preprocessing on the ambient acoustic signal, including framing, windowing, and pre-emphasis, and then convert it to the frequency domain using Fourier transform to obtain a complex spectrum. Then, the power spectrum is obtained by squaring the magnitude of the spectrum, normalizing the power spectrum along the frequency axis, and calculating the signal energy per unit frequency to obtain the power spectral density. Optionally, the detection system can calculate the arithmetic mean and geometric mean of the power spectrum based on the aforementioned power spectral density, and use the ratio of the geometric mean to the arithmetic mean as a characteristic parameter of spectral flatness.

[0066] Optionally, the detection system can calculate the effective signal strength of the audio frames of both microphones separately, perform a difference calculation on the signal strength of the two microphones, and obtain the signal strength difference between the two microphones.

[0067] In some embodiments, before performing feature extraction on the ambient acoustic signal to obtain wind speed-related feature parameters, the detection system can calculate the complex coherence coefficient of the dual-microphone signals and filter out non-wind noise interference based on the complex coherence coefficient.

[0068] For example, the complex coherence coefficient can be the correlation between the signals from both microphones in the frequency domain, used to distinguish between wind noise and far-field noise.

[0069] For example, non-wind noise interference can be environmental noise that overlaps with the wind noise frequency band but is not generated by airflow, such as traffic horns.

[0070] Optionally, the detection system can employ filters to filter high-frequency environmental noise while preserving the wind noise frequency band. For example, the detection system can use a Butterworth low-pass filter to filter high-frequency environmental noise (such as human voices and traffic noise) while preserving the core wind noise frequency band. A linear de-trending algorithm can also be used to eliminate DC components and slow drift in the signal.

[0071] Optionally, the detection system can divide the pre-processed dual-channel signal into frames of fixed length, calculate the amplitude of each frame, and remove abnormal frames whose amplitude exceeds a preset amplitude threshold. The preset amplitude threshold can be, for example, three times the average amplitude of all frames, and abnormal frames can be, for example, sudden external noise or impact noise caused by the headphones falling.

[0072] Optionally, the detection system can calculate the complex coherence coefficient for each frame. When the complex coherence coefficient ranges from [0,1], a value close to 1 indicates strong coherence between the two signals; a value close to 0 indicates weak coherence.

[0073] Optionally, if the complex coherence coefficient is greater than the coherence determination threshold, the detection system can determine that the frame signal has coherent non-wind noise interference (such as traffic noise) in the core frequency band of wind noise; if the complex coherence coefficient is less than the coherence determination threshold, the detection system can determine that the frame signal is pure wind noise or weakly coherent wind noise signal.

[0074] Optionally, if non-wind noise interference frame signals exist, the detection system can employ an adaptive noise cancellation algorithm, using the coherent components of both signals as reference noise to cancel the original signal and filter out coherent non-wind noise interference. It can also use frame signal interpolation to replace the features of interfering frames with the feature mean of adjacent pure wind noise frames.

[0075] Optionally, the detection system can also adjust the coherence determination threshold according to the ambient noise intensity of the motion scene. For example, if the ambient noise intensity is high, the detection system can lower the coherence determination threshold to improve the sensitivity of interference identification; if the ambient noise intensity is low, the detection system can raise the coherence determination threshold to avoid over-filtering wind noise signals.

[0076] Using the above method, before extracting features from the environmental acoustic signal and obtaining the feature parameters related to wind speed, the complex coherence coefficient of the dual-microphone signals is calculated, and non-wind noise interference is filtered out based on the complex coherence coefficient, thereby improving the accuracy of wind speed detection in environmental noise scenarios.

[0077] As one possible implementation, the detection system can adjust the cutoff frequency of the Butterworth low-pass filter and also adjust the window length of the detrending algorithm.

[0078] Optionally, the detection system can statistically analyze the proportion of high-frequency interference components in the ambient acoustic signal. If there are strong high-frequency noises such as high-frequency abnormal noises or electromagnetic crosstalk, the filter cutoff frequency is lowered; if the environment is clean and the effective frequency band of wind noise is wide, the cutoff frequency is raised.

[0079] Optionally, the detection system can combine extracted wind speed characteristic parameters such as power spectral density and spectral flatness to determine the main frequency band of the current wind noise and set the filter cutoff frequency to the upper limit of the effective frequency band of the wind noise. It can also preset multiple fixed cutoff frequency parameter levels and switch the corresponding frequency level according to the scene type (indoor calm wind, outdoor light wind, outdoor strong wind).

[0080] By adjusting the cutoff frequency of the Butterworth low-pass filter using the methods described above, signal purity is improved. Furthermore, adjusting the window length of the detrending algorithm enhances the accuracy of extracting characteristic parameters such as power spectral density, spectral flatness, and signal strength difference, thereby improving the accuracy of wind speed detection and energy consumption calculation.

[0081] S103, input the feature parameter into the deep learning model, output the wind speed prediction value, and correct the motion energy consumption calculation result based on the wind speed prediction value. The deep learning model is a nonlinear mapping model based on neural network algorithm.

[0082] In some embodiments, the deep learning model includes a data acquisition layer, a feature extraction layer, a deep learning modeling layer, a model optimization and adaptation layer, a calibration and control layer, and a wind direction and wind speed determination and output layer.

[0083] Among them: This data acquisition layer is used to cover multiple wind speeds, all-angle wind directions and various common motion postures in a simulated experimental environment, collect wind noise signals, as well as the true value data of measured wind speed and wind direction, and build a training dataset.

[0084] This feature extraction layer is used to extract monophonic wind noise features and binaural spatial features based on the training dataset, and output wind noise correlation features;

[0085] This deep learning modeling layer is used to build a nonlinear mapping relationship based on the wind noise correlation feature, preset a unified wind direction angle orientation definition rule, and output the initial prediction result; the initial prediction result includes the initial wind speed value and the initial wind direction angle data;

[0086] The model optimization adaptation layer is used to adapt the initial prediction result to the equipment structural attributes and establish a differentiated model parameter library; the differentiated model parameter library includes the initial wind speed value, the initial wind direction angle data, the equipment attitude sensing data, and the motion positioning speed data;

[0087] This calibration and control layer is used to calibrate the environmental reference wind speed, and combined with the differential model parameter library, corrects the model parameters of the wind speed vector deviation during the motion, and outputs the corrected wind speed and direction correction data.

[0088] This wind direction and speed determination output layer is used to combine spatial features with the corrected wind speed and direction correction data to determine the wind direction category determination result and the wind speed prediction value.

[0089] Optionally, the detection system can input the extracted wind speed-related feature parameters, such as power spectral density, spectral flatness, and the difference in signal strength between the two microphones, into a nonlinear mapping deep learning model built based on a neural network algorithm.

[0090] Optionally, the detection system can filter and integrate the input features through a feature extraction layer, fusing monophonic wind noise features and binaural spatial features to form standardized wind noise correlation features. Then, a deep learning modeling layer constructs a nonlinear mapping relationship based on the above wind noise correlation features, completes preliminary calculations according to preset wind direction angle definition rules, and outputs initial prediction results containing initial wind speed values ​​and initial wind direction angle data.

[0091] Then, the initial prediction results are matched with the equipment structural attributes through the model optimization and adaptation layer, and a differentiated model parameter library is constructed by combining the equipment attitude sensing data and motion positioning speed data. Then, the pre-calibrated environmental reference wind speed is retrieved through the calibration and control layer, and the wind speed vector deviation during the motion process is corrected by combining it with the model parameter library to obtain the corrected wind speed and direction correction data.

[0092] Then, by combining the wind direction and speed determination output layer with spatial characteristics, the wind direction category is determined, and a wind speed prediction value is output. The detection system can use this wind speed prediction value to calculate the energy consumption impact of wind resistance on the motion state, and thus correct the original motion energy consumption calculation results.

[0093] As one possible implementation, after inputting the feature parameter into the deep learning model, the detection system can calculate the deviation rate of the predicted wind speed on both sides and determine whether microphone hole obstruction has occurred based on the deviation rate.

[0094] For example, the deviation rate can be the proportion of the difference between the two sides of the wind speed prediction.

[0095] Optionally, the detection system can extract feature parameters such as power spectral density, spectral flatness, and signal strength difference, and feed them into a deep learning model for inference calculations, according to the feature data corresponding to the left and right microphones respectively. Based on the acoustic features of one side, the deep learning model outputs the wind speed prediction value corresponding to the left microphone and the wind speed prediction value corresponding to the right microphone, respectively, to obtain two sets of bilateral wind speed prediction values.

[0096] Optionally, the detection system can use the wind speed prediction value on one side as a benchmark to calculate the relative difference between the two values, or it can calculate the average of the two sets of prediction values ​​and then calculate the degree of deviation of each value from the average.

[0097] Optionally, the detection system can determine whether microphone hole obstruction has occurred based on a preset deviation threshold. For example, the detection system can compare the deviation trends of two prediction values ​​to identify whether the left or right microphone is obstructed. If both sets of wind speed prediction values ​​show a large deviation in the same direction and an abnormally high deviation rate, the detection system can determine that both microphone holes are obstructed simultaneously.

[0098] Optionally, the detection system can determine whether microphone obstruction exists using a fixed deviation threshold. It can also set different deviation thresholds for different scenarios to determine microphone obstruction. Furthermore, multiple deviation thresholds can be used to classify different levels of obstruction, such as slight obstruction, moderate obstruction, and complete obstruction.

[0099] By using the above method, after inputting the feature parameter into the deep learning model, the deviation rate of the predicted wind speed on both sides is calculated, and the microphone hole is blocked based on the deviation rate. This solves the problem of wind speed detection failure caused by masking and improves the accuracy of energy consumption calculation.

[0100] As one possible approach, after calculating the deviation rate of the predicted wind speeds on both sides, the detection system can adjust the input weights of the deep learning model based on the deviation rate to correct the predicted wind speeds under shading conditions.

[0101] For example, input weights can be the contribution coefficients of different feature parameters in a deep learning model. For instance, in the case of unilateral occlusion, the weights of the feature parameters on the occluded side are reduced, while the weights on the unoccluded side are increased.

[0102] Optionally, the detection system can compare the calculated deviation rate of the predicted wind speed on both sides with a preset normal deviation range to determine the current level of obstruction and the degree of signal anomaly. Based on the magnitude of the deviation rate, a pre-stored weight adjustment strategy is matched to determine the adjustment range corresponding to the left and right channel features and various wind speed feature parameters.

[0103] Optionally, the detection system can modify the input weights of features such as the corresponding feature branches of the two microphones, power spectral density, spectral flatness, and signal strength difference at the input layer of the deep learning model. Then, according to the adjusted weights, the features collected from the two channels are weighted and fused, and the reconstructed weighted features are fed into the deep learning model for further computation, outputting the corrected wind speed prediction value after occlusion compensation.

[0104] Optionally, the detection system can linearly adjust the input weights based on the deviation rate, or it can adjust the input weights in stages based on different deviation rate levels such as slight occlusion, moderate occlusion, and severe occlusion. The detection system can adjust the input weights only on the side where occlusion occurs, or it can selectively adjust the weights of power spectral density, spectral flatness, and signal strength difference.

[0105] Using the above method, after calculating the deviation rate of the predicted wind speed on both sides, the input weights of the deep learning model are adjusted according to the deviation rate to correct the predicted wind speed under the shading state, thereby optimizing the accuracy of the predicted wind speed under the shading state.

[0106] As one possible implementation, the detection system can also calculate air resistance based on the predicted wind speed and, combined with the movement speed and heart rate data of the exercise equipment, correct for exercise calorie consumption. Then, based on the deviation between the movement speed of the exercise equipment and the predicted wind speed, the parameters of the deep learning model are adjusted.

[0107] For example, air resistance can be the mechanical resistance generated by airflow on a person. For instance, if air resistance increases in a headwind, the detection system needs to consume more energy.

[0108] Optionally, the detection system can obtain any corrected and accurate wind speed prediction value, substitute it into the air resistance calculation formula, and calculate the real-time air resistance experienced by the user during exercise. Then, it synchronously retrieves the user's movement speed and real-time heart rate data collected by the exercise equipment and integrates it with the calculated air resistance data.

[0109] Optionally, the detection system can also incorporate additional energy expenditure due to air resistance into the basic exercise calorie calculation model, and combine this with the exercise intensity reflected by heart rate to compensate and correct the original estimated exercise calorie expenditure value to obtain the actual calorie expenditure.

[0110] Optionally, the detection system can also compare the measured movement speed of the moving equipment with the predicted wind speed to calculate the relative speed deviation formed by the coupling of the two. Based on the magnitude and fluctuation pattern of this speed deviation, the internal parameters of the deep learning model are iteratively optimized.

[0111] Optionally, the detection system can combine the equivalent frontal area of ​​the human body with a preset drag coefficient to calculate the basic air resistance at different wind speeds. The system can also calculate the magnitude of air resistance by matching the corresponding frontal area and drag coefficient based on the type of movement. Furthermore, the system can distinguish between headwinds, crosswinds, and other winds by using wind direction information identified by dual microphones, and correct for the effective air resistance corresponding to different wind directions.

[0112] Using the above method, air resistance is calculated based on the predicted wind speed, and the exercise calorie consumption is corrected by combining the moving speed of the exercise equipment and heart rate data, thus achieving dynamic compensation for exercise calorie consumption and improving the accuracy of energy consumption calculation. Based on the deviation between the moving speed of the exercise equipment and the predicted wind speed, the parameters of the deep learning model are corrected, improving the prediction accuracy and robustness of the deep learning model.

[0113] In this embodiment, ambient acoustic signals are collected using dual microphones built into the wearable device, providing data for subsequent feature extraction. Feature extraction is performed on these ambient acoustic signals to obtain wind speed-related feature parameters, eliminating redundant interference and improving the accuracy of wind speed detection. These feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the two microphone signals. These feature parameters are then input into a deep learning model, which outputs a predicted wind speed value. Based on this predicted wind speed value, the calculated motion energy consumption is corrected, improving the accuracy of energy consumption calculation. This deep learning model is a nonlinear mapping model based on a neural network algorithm.

[0114] Figure 2 A flowchart illustrating another microphone-based wind speed detection and energy consumption compensation method provided in this application. Figure 2 As shown, the method includes:

[0115] (1) Acoustic signal acquisition;

[0116] Optionally, wind noise signals can be collected using the dual external microphones on both sides of the TWS earbuds. This can be done in real time during exercise, with the sampling frequency synchronized with the earbuds' audio sampling frequency. The collected raw acoustic signals (dual channels) can be transmitted in real time via Bluetooth to the paired sports device (smartwatch / phone), or the wind noise and wind speed model can be calculated on the earbuds and then transmitted to the phone / watch via wind speed information.

[0117] Optionally, the microphone gain can be automatically adjusted for different sports scenarios (running, cycling) to avoid noise interference from headphone vibration and clothing friction during exercise. For example, the gain is reduced by 10%-15% in cycling scenarios, while the gain remains unchanged in running scenarios.

[0118] (2) Signal preprocessing: Filter out non-wind noise interference (such as ambient sound, motion vibration noise, and headphone circuit noise) in the original acoustic signal, and retain the pure wind noise signal to lay the foundation for subsequent feature extraction.

[0119] Optionally, a Butterworth low-pass filter with a cutoff frequency of 500Hz can be used to filter high-frequency environmental noise above 500Hz (such as human voices and traffic noise) while retaining the core frequency band of wind noise from 20Hz to 500Hz (this frequency band is the main frequency band of eddy noise generated by airflow through the earphone cavity).

[0120] Optionally, a linear detrending algorithm can be used to eliminate DC components and slow drifts (such as signal shifts caused by microphone temperature drift) in the signal, ensuring the stability of the wind noise signal.

[0121] Optionally, the preprocessed dual-channel signal is divided into frames of a fixed length, with the frame length set to 20ms-50ms (preferably 30ms) and the frame overlap rate set to 50%, to avoid feature extraction errors caused by signal breakage and improve computational efficiency.

[0122] Optionally, by calculating the amplitude of each frame signal, abnormal frames with amplitudes exceeding a threshold (the threshold is 3 times the average amplitude of all frames) (such as sudden external noise or impact noise caused by the falling of headphones) are removed to ensure the accuracy of subsequent feature extraction.

[0123] Optional non-wind noise interference identification and filtering methods include coherence calculation, interference judgment threshold, interference signal filtering, and dynamic threshold adaptation.

[0124] Optionally, the coherence calculation calculates the complex coherence coefficient of each frame for the preprocessed dual-channel framed signal. The complex coherence coefficient ranges from [0,1]. The closer the value is to 1, the stronger the coherence of the two-sided signal. The closer the value is to 0, the weaker the coherence.

[0125] Optionally, a coherence threshold for non-wind noise interference can be calibrated through wind tunnel experiments and real-world urban scenario testing. When the complex coherence coefficient is set to a threshold, the frame signal is determined to have coherent non-wind noise interference (such as traffic noise) in the 20Hz-500Hz frequency band; when the complex coherence coefficient is less than the set threshold, the frame signal is determined to be a pure wind noise / weakly coherent wind noise signal.

[0126] Optionally, for frame signals determined to have non-wind noise interference, an adaptive noise cancellation algorithm is used to cancel the original signal using the coherent components of the two-sided signals as reference noise, thus filtering out coherent non-wind noise interference. For residual interference that still exists after filtering, a frame signal interpolation replacement method is used to replace the features of the interfering frame with the feature mean of adjacent pure wind noise frames to ensure signal continuity.

[0127] Optionally, the system dynamically adjusts the coherence determination threshold based on the ambient noise intensity of the motion scene (determined by the total signal energy). When the ambient noise intensity is high (such as on urban main roads), the threshold is lowered to improve the sensitivity of interference identification; when the ambient noise intensity is low (such as in the suburbs), the threshold is raised to avoid over-filtering wind noise signals.

[0128] (3) Wind noise feature extraction: Extract features in the wind noise signal that are strongly correlated with wind speed and wind direction, including monophonic wind noise features (for wind speed identification) and stereophonic spatial features (for wind direction identification), as follows:

[0129] For example, monophonic wind noise characteristics may include low-frequency energy characteristics, pseudo-sound characteristics, and spectral flatness characteristics.

[0130] Optionally, the detection system can calculate the power spectral density (PSD) of each frame of signal in the 20Hz-500Hz frequency band, and extract the peak value, mean value, and cumulative energy of the PSD as low-frequency energy features. For example, the greater the wind speed, the faster the airflow passes through the headphone microphone hole, the stronger the eddy noise, and the higher the peak value and cumulative energy of the PSD, with a positive correlation between the two.

[0131] Optionally, the detection system can utilize the frequency shift characteristics related to the Strouhal number (St) to identify eddy noise generated by airflow passing through the headphone cavity and determine pseudo-sound features. The Strouhal number is a dimensionless number, obtained by calculating the product of the characteristic frequency of the eddy noise and the characteristic length of the headphone microphone hole, and then dividing this product by the relative airflow velocity. By extracting the shift of the characteristic frequency of the eddy noise in the signal (compared to a reference frequency in a windless state) and combining it with the pre-calibrated characteristic length value of the headphone microphone hole, the range of relative wind speed can be preliminarily calculated.

[0132] Optionally, the detection system can calculate the spectral flatness (SFM) of each frame of signal. The spectrum of wind noise signals is concentrated in low frequencies and has a low spectral flatness (usually between 0.1 and 0.3), while the spectral flatness of environmental noise (such as human voice) is higher (usually above 0.5). This feature can be used to further distinguish between wind noise and non-wind noise signals and improve the purity of feature extraction.

[0133] For example, two-channel spatial characteristics include interaural level difference (ILD), cross-correlation coefficient (CCC), and interaural phase difference (IPD).

[0134] Optionally, the detection system can calculate the amplitude difference between the signals from the left and right microphones in the same frame. Different wind directions result in different impact intensities of airflow on the two microphones, leading to different signal strength differences. For example, in a headwind, the ILD value of both microphones is smaller (due to uniform airflow impact on both sides); in a crosswind, the amplitude of the microphone on the windward side is greater than that on the leeward side, resulting in a larger ILD value (the difference ranges from 0.5dB to 5dB).

[0135] Optionally, the detection system can calculate the cross-correlation coefficient (CCC) of the signals from both microphones to reflect the similarity between the two signals. Under tailwind / headwind conditions, the airflow impact patterns experienced by both microphones are consistent, resulting in a higher CCC value (typically ≥0.8); under crosswind conditions, the similarity between the two signals decreases, leading to a lower CCC value (typically ≤0.6).

[0136] Optionally, the detection system can calculate the phase difference between the signals from both microphones in the 20Hz-500Hz frequency band. Changes in wind direction will cause different time differences in the arrival of airflow at both microphones, thus generating a phase difference. The incident angle of the wind can be further accurately determined by the phase difference (accuracy up to ±10°).

[0137] (4) Wind speed / direction prediction model establishment: Based on the extracted wind noise features, a nonlinear mapping model is established to realize the accurate calculation of relative wind speed and the determination of wind direction. At the same time, a self-calibration mechanism is introduced to improve the adaptability and accuracy of the model.

[0138] The model training process is as follows:

[0139] Training data acquisition: In the wind tunnel laboratory, wind noise signals from the microphones on both sides of the TWS earphones were collected under different wind speeds (0-15m / s, 0.5m / s interval), different wind directions (0° to 360°, 30° interval), and different sports postures (running, cycling). At the same time, real wind speed and wind direction data were recorded using a professional hot-wire anemometer to build a training dataset (sample size ≥10000 groups).

[0140] Model selection: A deep learning model is adopted. The input is the extracted mono-channel wind noise features (PSD peak value, pseudo-sound feature frequency, spectral flatness) and dual-channel spatial features (ILD, CCC, IPD). The output is the relative wind speed and wind direction angle (0° is directly in front, 180° is directly behind, 90° is right side, and 270° is left side).

[0141] Model optimization: The model was trained using the Adam optimizer with at least 1000 iterations, employing mean squared error as the loss function to ensure that the model's predicted wind speed error was ≤0.3m / s and predicted wind direction error was ≤10°. Simultaneously, corresponding model parameter libraries were established for different headphone models (with varying microphone hole sizes and cavity structures) to improve the model's versatility.

[0142] Optionally, after the system starts, it first uses the accelerometer of the motion device to determine whether the user is in a stationary state (acceleration change ≤ 0.1g, duration ≥ 3s). At this time, the GPS speed is 0, and the wind speed corresponding to the wind noise signal collected by the headphones is the ambient wind speed. This is used as the ambient wind reference value and stored in the device.

[0143] Optionally, after the exercise begins, the relative wind speed output by the model is the vector sum of the ambient wind speed and the athlete's own speed. The model's output offset is dynamically corrected by the difference between the GPS speed (approximately equal to the athlete's own speed) and the relative wind speed. When the GPS speed is stable, if the deviation of the relative wind speed from the sum of the GPS speed and the ambient wind speed exceeds 0.5 m / s, the model's weight parameters are automatically adjusted to ensure the accuracy of wind speed prediction.

[0144] Optionally, the detection system can also automatically call up the corresponding calibration parameters for different sports scenarios (running, cycling). For example, when cycling, the headphones are usually fixed on the ears, making wind noise collection more stable, and the calibration frequency is once every 30 seconds; when running, the headphones may shake slightly, and the calibration frequency is increased to once every 15 seconds to avoid errors caused by shaking.

[0145] Optionally, based on the dual-channel spatial features (ILD, CCC, IPD) and combined with the wind direction angle output by the model, the incident direction of the wind is determined. The incident direction of the wind includes headwind, tailwind, crosswind (left / right), and oblique wind. For example, headwind can be a wind direction angle between 160° and -200°, where the relative wind speed is greater than the GPS speed; tailwind can be a wind direction angle between 340° and -20°, where the relative wind speed is less than the GPS speed; right-side wind can be a wind direction angle between 70° and -110°; left-side wind can be a wind direction angle between 250° and -290°; oblique wind can be any other angle range, further subdivided according to the wind direction angle value (such as left headwind, right tailwind, etc.).

[0146] Optionally, when the microphone hole is blocked by a hat, scarf, hair, etc., it will change the characteristic length of the microphone hole and the wind noise spectrum distribution, resulting in a significant deviation between the wind speed prediction value of one side and the other side. By utilizing the consistency difference of the wind speed predictions on both sides, the occlusion status can be determined and targeted corrections can be made.

[0147] Optionally, the detection system can input the extracted monophonic wind noise features from the left and right microphones into the wind speed prediction sub-model to obtain the predicted wind speed on the left and right sides, and calculate the bilateral wind speed deviation rate. This bilateral wind speed deviation rate is calculated by dividing the absolute value of the difference between the predicted wind speed on the left and the predicted wind speed on the right by the maximum value of the predicted wind speed on the left or the right.

[0148] Optionally, the detection system can calibrate the shading state judgment threshold through wind tunnel experiments. When the bilateral wind speed deviation rate is ≤15%, it is judged as an unshading state, and the wind speed / direction prediction result of bilateral feature fusion is used; when 15% < bilateral wind speed deviation rate ≤40%, it is judged as a slight shading state on one side; when the bilateral wind speed deviation rate >40%, it is judged as a severe / complete shading state on one side; if the bilateral wind speeds are close to 0 and the wind noise energy drops sharply but the GPS speed is >5m / s, it is judged as a partial shading state on both sides.

[0149] Optionally, if the system determines that there is slight unilateral occlusion, the detection system can reduce the weight of the wind noise features on the occluded side (the side where the predicted wind speed deviates from the overall trend) by a factor of 0.3-0.5, increase the weight of the features on the unoccluded side, and re-integrate the features from both sides to predict wind speed / direction.

[0150] Optionally, if the system determines that the signal is severely or completely blocked on one side, it can automatically switch to the single-side effective mode, discard the signal from the blocked side, and use only the monophonic wind noise characteristics and motion state parameters (GPS speed, heading) of the unblocked side to calculate the wind speed / direction. At the same time, it can call the model parameter library of the single-side detection to ensure the prediction accuracy.

[0151] Optionally, if the system determines that there is partial occlusion on both sides, it can trigger wind noise feature compensation. Based on the mapping relationship between GPS speed and wind noise features in historical windless / light wind scenarios, the system can perform amplitude and spectrum compensation on the currently collected weak wind noise features before performing model prediction.

[0152] When a device is determined to have severe unilateral occlusion or partial occlusion on both sides for a duration of ≥5 seconds, the detection system can send a wearing check prompt to the user via vibration or text from the motion device (watch / phone) (e.g., "The left microphone may be blocked, please check the wearing status of the headphones"). At the same time, the system records the duration of occlusion in the software to provide a basis for subsequent data correction.

[0153] When the deviation rate of wind speed on both sides drops to within 15% and lasts for ≥3s, the detection system can determine that the occlusion is lifted, automatically switch back to the dual-side feature fusion mode, and restore the basic self-calibration logic without requiring manual operation by the user.

[0154] (5) Motion status fusion: Combine multi-dimensional data from sports devices (watches / phones) to determine the user's motion status, provide accurate status basis for energy consumption compensation, and avoid compensation deviations in different sports scenarios.

[0155] Optionally, the detection system can use the three-axis accelerometer and gyroscope of the sports equipment to extract motion features (such as cadence, arm swing amplitude, and acceleration variation patterns) to determine whether the user is in a running state (cadence of 120-180 steps / minute, with large acceleration fluctuations) or a cycling state (no obvious cadence, stable acceleration), with an accuracy rate of ≥95%.

[0156] Optionally, the detection system can obtain the user's real-time speed, heading (compared with wind direction), and altitude changes via GPS; obtain real-time heart rate (beats / minute) via a heart rate sensor; and calculate the basal metabolic rate based on the user's pre-input basic information (height, weight, age, gender).

[0157] Optionally, if the GPS speed remains stable (fluctuation ≤1m / s, duration ≥5s), the wind noise energy collected by the headphones decreases sharply (reduced by ≥30% compared to the average wind noise energy in the previous 5s), and the user's movement posture is cycling (following others) or running (following others), the system determines that the user has entered the "following others" state (being in the wake of others, with significantly reduced air resistance), and records the time of entering this state, its duration, and the percentage reduction in wind noise energy as the basis for subsequent energy consumption compensation.

[0158] (6) Energy consumption compensation calculation: Based on the identified relative wind speed, wind direction and motion state, calculate the impact of air resistance on energy consumption, establish an energy consumption correction model, and output accurate heat consumption data.

[0159] Optionally, the detection system can use the classic metabolic equivalent of energy (MET) model to calculate basal energy consumption. The calculation formula is as follows: ,in, Based on energy consumption, Exercise metabolic equivalent (during running) During cycling, between 8 and 12. (Adjust dynamically based on heart rate, between 6 and 10). User's weight (kg) The duration of the motion is in hours (h).

[0160] Optionally, the detection system can calculate the instantaneous air resistance experienced during movement based on the relative wind speed. The calculation formula is as follows: ,in, Relative wind speed, The coefficient of air resistance (when running) Between 0.8 and 1.0, while cycling Between 0.5 and 0.7, preset according to the motion posture). The frontal area (m²) is estimated based on the user's height and weight during running. In an area of ​​0.5 to 0.7 m², while cycling (in 0.3 to 0.5 m²) Air density (kg / m³, default value is 1.225 kg / m³, can be dynamically adjusted according to altitude).

[0161] Optionally, the detection system can be based on the formula Converting air resistance into mechanical power increment ,in, GPS speed (m / s) The unit is W. A positive value indicates the extra power required (headwind), and a negative value indicates the reduced power (tailwind / riding).

[0162] Optionally, the detection system can be based on the formula ,Will Converted into an increase in heat consumption (1W = 0.86kcal / h). According to the formula... Calculate precise energy consumption .

[0163] If it's a headwind scenario If the value is positive, the energy consumption is corrected upwards, with the correction ratio adjusted based on the ratio of relative wind speed to GPS speed. In a tailwind scenario, If the value is negative, energy consumption is adjusted downwards by 15%-40% (dynamically adjusted based on the ratio of relative wind speed to GPS speed). For riding / running scenarios, energy consumption is further reduced by 5%-15% on top of the tailwind correction (adjusted based on the reduction ratio of wind noise energy). At the same time, the percentage of time the user spends in the resistance reduction zone is recorded as an indicator of exercise performance (e.g., "30% of riding time saves 25% of energy consumption").

[0164] (7) Data output.

[0165] The obtained precise data, such as energy consumption, relative wind speed, wind direction and angle, exercise status (running / cycling / following), energy consumption correction ratio, wearing occlusion status, and environmental noise intensity level, are output in real time to the display screen of sports equipment, watches, and mobile phones or to the accompanying software application; at the same time, historical data (including the original data of occlusion period and noise scene) are stored for subsequent sports analysis and model optimization.

[0166] Figure 3 A schematic diagram of the architecture of a microphone-based wind speed detection and energy consumption compensation system is provided for this application, as shown below. Figure 3 As shown, the microphone-based wind speed detection and energy consumption compensation system includes:

[0167] The TWS earbuds feature dual external microphones: left earbud microphone 1, left earbud microphone 2, right earbud microphone 1, and right earbud microphone 2. These microphones are used to collect ambient wind noise and acoustic signals, providing raw data for wind speed detection. The earbuds' Bluetooth module establishes a wireless connection with the smartwatch / phone for data transmission. The earbud gain adjustment module adaptively adjusts the microphone signal gain according to different sports scenarios (such as running and cycling) to optimize wind noise collection.

[0168] The wind speed processing module is a compatible core processing unit that can be deployed on TWS earphones or smartwatches / phones. When deployed on TWS earphones, this module directly processes the raw audio captured by the microphone, extracting wind noise features, calculating wind speed / direction, and transmitting core information such as wind speed and direction to the smartwatch / phone via the earphone's Bluetooth module. When deployed on a smartwatch / phone, this module receives the raw audio signal transmitted by the earphone's Bluetooth module and performs subsequent wind speed detection and calculation on the smart device.

[0169] On the smartwatch / phone, the multi-sensor system integrates a three-axis accelerometer, gyroscope, GPS, heart rate sensor, etc., to collect information such as motion status, speed, location, and physiological parameters, providing auxiliary data for wind speed calibration and energy consumption compensation. The motion processing module, as the core control unit, is responsible for coordinating multi-sensor data, motion status recognition, and overall process scheduling. The core processing module includes a wind speed processing module; when deployed on the smartwatch / phone, this module performs wind speed / direction calculations, model inference, and energy consumption compensation algorithms. The Bluetooth module communicates with the Bluetooth module of the TWS earphone, receiving raw audio signals or core wind speed information and issuing control commands. The storage module stores wind speed model parameters, historical motion data, user information, and calculation results.

[0170] The interactive module's display module can show real-time information such as wind speed, wind direction, energy consumption, and exercise status. The notification module uses vibration and text to alert the user when the microphone is blocked or malfunctions. The accompanying software application provides a user interface that supports information input, historical data querying, model parameter updates, and exercise report generation.

[0171] Figure 4 A flowchart illustrating a wind speed detection and energy consumption compensation system method provided in this application is shown below. Figure 4 As shown, the method includes:

[0172] System initialization: Completes the input of basic physiological information of users, Bluetooth pairing of headphones and smartwatches / mobile phones, loading of wind speed / wind direction prediction models, and initial setting of microphone gain.

[0173] Ambient wind baseline calibration: The user's stationary state is determined by the accelerometer of the motion device, the wind noise signal at this time is collected and the ambient wind speed is calculated, which serves as the baseline value for subsequent dynamic correction of the wind speed model.

[0174] Motion state recognition and real-time data acquisition: Acoustic signals are collected by external microphones on both sides of the earphone, motion state parameters are collected by a multi-sensor module, and the original audio signal or core wind speed information is transmitted directionally by a Bluetooth transmission module according to the deployment location of the wind speed processing module.

[0175] Signal preprocessing: The acquired acoustic signals are sequentially subjected to Butterworth low-pass filtering, linear detrending, framing, and outlier removal to filter out non-wind noise interference and retain pure wind noise signals.

[0176] Wind noise feature extraction: Extract mono-channel wind noise features (power spectral density, pseudo-sound features, spectral flatness) and dual-channel spatial features (signal strength difference, cross-correlation coefficient, phase difference) to provide feature basis for wind speed / direction calculation.

[0177] Wind speed / direction calculation and self-calibration: Input wind noise characteristics into the model to obtain relative wind speed and wind direction angle, combine GPS speed to dynamically correct model parameters, and simultaneously complete the determination of headwind, tailwind, and crosswind.

[0178] Motion state fusion and special state recognition: Complete running / cycling motion posture recognition, and follow-cycling / running state recognition based on GPS speed stability and wind noise energy changes, while determining the headphone microphone occlusion state and completing targeted model correction.

[0179] Energy consumption compensation calculation: First, the basic energy consumption is calculated through the MET model, and then the air resistance and energy consumption increment are calculated by combining the relative wind speed. The energy consumption is dynamically corrected according to different scenarios such as headwind, tailwind, and riding / running, so as to obtain accurate exercise energy consumption.

[0180] Data output and storage: The system outputs precise data such as energy consumption, wind speed, wind direction, and motion status to the interactive output module in real time. It also stores historical motion data and model calibration parameters for subsequent motion analysis and model optimization.

[0181] In some embodiments, during the system initialization step, the user inputs basic physiological information (height, weight, age, gender) through a companion software application, and the system automatically calculates the basal metabolic rate and initial... Value (default for running) =10, default for cycling =8). The TWS earbuds are paired with a smartwatch or mobile phone via Bluetooth, confirming that the microphone is working properly and the gain is at the default value. Then, the detection system loads the pre-trained wind speed / direction prediction model (calling the corresponding model parameter library according to the earbud model) to complete the initialization.

[0182] Optionally, in the environmental wind benchmark calibration step, the system uses the smartwatch's accelerometer to determine if the user is stationary (acceleration change ≤ 0.1g, duration ≥ 3s). If stationary, the dual microphones on both sides of the earphone collect environmental acoustic signals in real time for 3 seconds, completing signal preprocessing (filtering, detrending, framing, outlier removal). Then, wind noise features are extracted, input into the wind speed / direction prediction model, and the relative wind speed at this time is output as the environmental wind speed, stored in the watch as a benchmark for subsequent dynamic correction.

[0183] Optionally, during exercise, real-time data acquisition and processing steps are performed. If the user starts exercising (running / cycling), the user can manually select or the system can automatically identify the exercise posture (based on accelerometer and gyroscope data; a cadence of 120-180 steps / minute is considered running, while no cadence and stable acceleration are considered cycling). The dual microphones on the headphones acquire ambient acoustic signals in real-time, framed every 30ms, and synchronously transmitted to the smartwatch. The smartwatch then preprocesses the received signals (Butterworth low-pass filtering, cutoff frequency 500Hz; linear detrending; framing; outlier removal), extracting mono-channel wind noise features (PSD peak value, pseudo-sound characteristic frequency, spectral flatness) and dual-channel spatial features (ILD, CCC, IPD). The smartwatch obtains real-time speed and heading via GPS and real-time heart rate via a heart rate sensor, dynamically adjusting... Value (the higher the heart rate, the better) The larger the value, the adjustment range is 8-12 for running and 6-10 for cycling.

[0184] Optionally, in the wind speed / direction calculation and self-calibration step, the detection system inputs the extracted wind noise features into the wind speed / direction prediction model, outputs the relative wind speed and wind direction angle, calculates the difference between the relative wind speed and the sum of the GPS speed and the ambient wind speed. If the difference exceeds 0.5 m / s, the system automatically adjusts the model's weight parameters and recalculates the relative wind speed to ensure prediction accuracy. The system determines the wind's incident direction (headwind, tailwind, crosswind, oblique wind) based on the wind direction angle and outputs this information synchronously to the display module.

[0185] Optionally, in the follow-riding / running state recognition step, the system monitors the stability of GPS speed in real time (fluctuation ≤1m / s, duration ≥5s), and simultaneously detects the wind noise energy collected by the headphones. Compared with the average wind noise energy in the previous 5s, it determines whether there is a sudden decrease of ≥30%. If all three conditions are met simultaneously—"stable GPS speed," "sudden decrease in wind noise energy," and "movement posture is cycling / running"—the system determines that the user has entered the "follow-riding / running" state, and records the entry time, duration, and percentage reduction in wind noise energy. If any one of the conditions is not met, it is determined to be a non-follow-riding / running state, and is handled as a regular headwind / tailwind scenario.

[0186] Optionally, in the energy consumption compensation calculation and output step, the system uses a formula... Calculate basic energy consumption ; through formula ( For running, use 0.9; for cycling, use 0.6. Use 0.6m² for running and 0.4m² for cycling; Calculate air resistance (using 1.225 kg / m³). ; through formula Calculate drag power ; through formula Calculate energy consumption increment ; through formula Calculate precise energy consumption .

[0187] If the wind angle is between 160° and 200°, it is considered a headwind scenario, and the angle is adjusted upwards by 6% for every 1% of body weight resistance. Value; when the wind direction angle is between 340° and -20°, it is considered a tailwind scenario, with a correction ratio of 15%-40%; for riding / running scenarios, in addition to the tailwind correction, the precise energy consumption is further reduced by 5%-15%. Simultaneously, it records the proportion of the drag reduction zone. The obtained precise data, such as energy consumption, relative wind speed, wind direction angle, motion status, and correction ratio, are displayed in real time on the smartwatch screen, stored locally, and synchronized to the accompanying app.

[0188] Optionally, after the user stops exercising, the system stops data collection, calculates statistical data such as total energy consumption, average wind speed, maximum wind speed, and percentage of time spent riding / running during the exercise, outputs an exercise analysis report, and saves the latest calibration parameters of the model to provide a basis for the next use.

[0189] Figure 5 This application provides a schematic diagram of a microphone-based wind speed detection and energy consumption compensation device. Figure 5As shown, the microphone-based wind speed detection and energy consumption compensation device 500 may include: a data acquisition module 501, an extraction module 502, and an output module 503. Among them,

[0190] Acquisition module 501 is used to acquire ambient acoustic signals through the dual microphones built into the wearable device;

[0191] The extraction module 502 is used to extract features from the environmental acoustic signal to obtain feature parameters related to wind speed; the feature parameters include at least the power spectral density, spectral flatness and signal strength difference of the two microphone signals of the environmental acoustic signal;

[0192] The output module 503 is used to input the feature parameter into the deep learning model, output the wind speed prediction value, and correct the motion energy consumption calculation result based on the wind speed prediction value. The deep learning model is a nonlinear mapping model based on a neural network algorithm.

[0193] Optionally, the extraction module 502 is also used to extract features from the environmental acoustic signal, calculate the complex coherence coefficient of the dual-microphone signals before obtaining the feature parameters related to wind speed, and filter out non-wind noise interference based on the complex coherence coefficient.

[0194] Optionally, the extraction module 502 is further configured to perform feature extraction on the environmental acoustic signal, including at least one of the following:

[0195] Adjust the cutoff frequency of the Butterworth low-pass filter;

[0196] Adjust the window length for the detrending algorithm.

[0197] Optionally, the output module 503 is also used for the deep learning model to include a data acquisition layer, a feature extraction layer, a deep learning modeling layer, a model optimization and adaptation layer, a calibration and control layer, and a wind direction and wind speed determination output layer.

[0198] Among them: This data acquisition layer is used to cover multiple wind speeds, all-angle wind directions and various common motion postures in a simulated experimental environment, collect wind noise signals, as well as the true value data of measured wind speed and wind direction, and build a training dataset.

[0199] This feature extraction layer is used to extract monophonic wind noise features and binaural spatial features based on the training dataset, and output wind noise correlation features;

[0200] This deep learning modeling layer is used to build a nonlinear mapping relationship based on the wind noise correlation feature, preset a unified wind direction angle orientation definition rule, and output the initial prediction result; the initial prediction result includes the initial wind speed value and the initial wind direction angle data;

[0201] The model optimization adaptation layer is used to adapt the initial prediction result to the equipment structural attributes and establish a differentiated model parameter library; the differentiated model parameter library includes the initial wind speed value, the initial wind direction angle data, the equipment attitude sensing data, and the motion positioning speed data;

[0202] This calibration and control layer is used to calibrate the environmental reference wind speed, and combined with the differential model parameter library, corrects the model parameters of the wind speed vector deviation during the motion, and outputs the corrected wind speed and direction correction data.

[0203] This wind direction and speed determination output layer is used to combine spatial features with the corrected wind speed and direction correction data to determine the wind direction category determination result and the wind speed prediction value.

[0204] Optionally, the output module 503 is also used to calculate the deviation rate of the predicted wind speed on both sides after inputting the feature parameter into the deep learning model, and to determine whether microphone hole occlusion has occurred based on the deviation rate.

[0205] Optionally, the output module 503 is also used to calculate the deviation rate of the predicted wind speed on both sides, and then adjust the input weights of the deep learning model according to the deviation rate to correct the predicted wind speed under the shading state.

[0206] Optionally, the output module 503 is also used to calculate air resistance based on the predicted wind speed and, in conjunction with the movement speed and heart rate data of the exercise equipment, correct the calorie consumption during exercise. Based on the deviation between the movement speed of the exercise equipment and the predicted wind speed, the parameters of the deep learning model are corrected.

[0207] The microphone-based wind speed detection and energy consumption compensation device provided in this application embodiment can execute the methods provided in any of the above method embodiments, and its implementation principle and technical effect are similar. This embodiment will not be described in detail here.

[0208] Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in this application. Figure 6 As shown, the electronic device 600 includes a memory 601, a processor 602, and a communication interface 603. The memory 601, processor 602, and communication interface 603 are communicatively connected to each other. For example, the memory 601, processor 602, and communication interface 603 can be connected via a network. Alternatively, the electronic device 600 may also include a bus 604. The memory 601, processor 602, and communication interface 603 are communicatively connected to each other via the bus 604. Figure 6 The electronic device 400 is a device that uses a memory 601, a processor 602, and a communication interface 603 to communicate with each other via a bus 604.

[0209] The memory 601 may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 601 may store programs, and when the program stored in the memory 601 is executed by the processor 602, the processor 602 and the communication interface 603 are used to execute the methods provided in any of the foregoing embodiments. The memory may also store data required for at least one method in this embodiment.

[0210] The processor 602 may be a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), graphics processing unit (GPU), or one or more integrated circuits.

[0211] Processor 602 can also be an integrated circuit chip with signal processing capabilities. In implementation, the methods disclosed in the embodiments of this application can be completed by the integrated logic circuits in the hardware of processor 602 or by instructions in software form. The processor 602 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments below of this application. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments below of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 401, and processor 402 reads the information in memory 401 and, in conjunction with its hardware, completes the methods disclosed in the embodiments of this application.

[0212] The communication interface 603 uses transceiver modules, such as, but not limited to, transceivers, to enable communication between the electronic device 600 and other devices or communication networks. For example, a dataset can be acquired through the communication interface 603.

[0213] When the aforementioned electronic device 600 includes a bus 604, the bus 604 may include a path for transmitting information between various components of the electronic device 600 (e.g., memory 601, processor 602, communication interface 603).

[0214] This application also provides a computer-readable storage medium, which may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. Specifically, the computer-readable storage medium stores program instructions, which are used in the methods described in the above embodiments.

[0215] This application also provides a program product including executable instructions stored in a readable storage medium. At least one processor of an electronic device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the electronic device to perform the methods provided in the various embodiments described above.

[0216] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0217] It should be further noted that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0218] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0219] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0220] When integrated units / modules are implemented in hardware, the hardware can be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor can be any suitable hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC, etc. Unless otherwise specified, the storage unit can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc.

[0221] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0222] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0223] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0224] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A microphone-based wind speed detection and energy consumption compensation method, characterized in that, The method includes: Ambient acoustic signals are collected using dual microphones built into the wearable device; Feature extraction is performed on the ambient acoustic signal to obtain wind speed-related feature parameters; the feature parameters include at least the power spectral density, spectral flatness, and signal strength difference between the two microphone signals of the ambient acoustic signal; The feature parameters are input into a deep learning model, which outputs a wind speed prediction value. The calculation result of motion energy consumption is then corrected based on the wind speed prediction value. The deep learning model is a nonlinear mapping model based on a neural network algorithm.

2. The method of claim 1, wherein, The deep learning model includes a data acquisition layer, a feature extraction layer, a deep learning modeling layer, a model optimization and adaptation layer, a calibration and control layer, and a wind direction and wind speed determination and output layer. The data acquisition layer is used to cover multiple wind speeds, all-angle wind directions, and various common motion postures in a simulated experimental environment, collect wind noise signals, and real-value data of measured wind speed and direction, and construct a training dataset. The feature extraction layer is used to extract monophonic wind noise features and binaural spatial features based on the training dataset, and output wind noise correlation features. The deep learning modeling layer is used to build a nonlinear mapping relationship based on the wind noise correlation features, preset a unified wind direction angle orientation definition rule, and output the initial prediction result; the initial prediction result includes the initial wind speed value and the initial wind direction angle data. The model optimization and adaptation layer is used to adapt the initial prediction results to the equipment structural attributes and establish a differentiated model parameter library; the differentiated model parameter library includes the initial wind speed value, the initial wind direction angle data, equipment attitude sensing data, and motion positioning speed data; The calibration and control layer is used to calibrate the environmental reference wind speed, and, in conjunction with the differentiated model parameter library, correct the model parameters of the wind speed vector deviation during the motion process, and output the corrected wind speed and direction correction data. The wind direction and wind speed determination output layer is used to combine spatial features with the corrected wind speed and wind direction correction data to determine the wind direction category determination result and the wind speed prediction value.

3. The method of claim 1, wherein, Before performing feature extraction on the environmental acoustic signal to obtain wind speed-related feature parameters, the process includes: Calculate the complex coherence coefficient of the dual-microphone signals, and filter out non-wind noise interference based on the complex coherence coefficient.

4. The method of claim 1, wherein, The feature extraction of the environmental acoustic signal includes at least one of the following: Adjust the cutoff frequency of the Butterworth low-pass filter; Adjust the window length for the detrending algorithm.

5. The method of claim 1, wherein, The step of inputting the feature parameters into the deep learning model includes: Calculate the deviation rate of the predicted wind speeds on both sides, and determine whether microphone hole obstruction has occurred based on the deviation rate.

6. The method of claim 5, wherein, After calculating the deviation rate of the predicted wind speeds on both sides, the following is included: The input weights of the deep learning model are adjusted based on the deviation rate to correct the wind speed prediction under occlusion conditions.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Air resistance is calculated based on wind speed forecasts, and exercise calorie consumption is adjusted by combining the movement speed and heart rate data of the exercise equipment. The parameters of the deep learning model are corrected based on the deviation between the moving speed of the motion device and the predicted wind speed.

8. An electronic device, comprising: include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-8.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images