A method and system for adaptive adjustment of noise cancellation parameters in Bluetooth headphones

By extracting the sealing state characteristics and low-frequency coherence characteristics of the feedback microphone and feedforward microphone signals of Bluetooth headsets, and combining them with motion sensors to identify instantaneous movements, the problem of misjudgment of sealing state in Bluetooth headset noise reduction technology has been solved, achieving stability and comfort in noise reduction effect.

CN123123776APending Publication Date: 2026-11-17GUANGDONG HUAZHUANG TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611407866.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-11
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing Bluetooth headphone noise cancellation technology cannot accurately judge the continuous changes in the sealing state, resulting in a high misjudgment rate, reduced noise cancellation effect, ear pressure discomfort, or amplified wind noise.

Method used

By acquiring the feedback microphone signal and feedforward microphone signal from the Bluetooth headset, the sealing status characteristics and low-frequency coherence characteristics are extracted. Combined with the motion sensor to identify whether the wearer is in a momentary motion state, the noise reduction parameter adaptive adjustment component is activated, and the noise reduction parameters are adjusted according to the sealing status level.

Benefits of technology

It enables accurate judgment of changes in sealing status under low power consumption conditions, ensuring the stability and comfort of noise reduction effect, and avoiding misjudgment and parameter adjustment interference caused by instantaneous action.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123776A_ABST
    Figure CN123123776A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for adaptive adjustment of noise reduction parameters in Bluetooth headsets, relating to the field of intelligent sensor technology. The method includes: acquiring feedback microphone signals and feedforward microphone signals, and extracting sealing state features; extracting low-frequency coherence features based on the feedforward and feedback microphone signals; acquiring motion signals and identifying whether the wearer is in a state of instantaneous motion; activating the noise reduction parameter adaptive adjustment component when the sealing state features and low-frequency coherence features meet preset sealing change conditions and the wearer is not in a state of instantaneous motion; obtaining the current sealing state level based on the sealing state features and low-frequency coherence features; matching the target noise reduction parameters according to the current sealing state level and adjusting the noise reduction parameters. This solves the technical problem in existing noise reduction schemes that cannot accurately determine the continuous change in sealing state, leading to high misjudgment rates, decreased noise reduction effect, ear pressure discomfort, or amplified wind noise.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and specifically to a method and system for adaptive adjustment of noise reduction parameters in Bluetooth headsets. Background Technology

[0002] Currently, the active noise cancellation technology of Bluetooth headphones mainly uses the microphone inside the headphones to collect ambient noise, which is then processed by a noise reduction filter to generate an inverse noise signal. This signal is then played by the speaker to cancel out the ambient noise entering the ear, providing users with a quiet listening environment.

[0003] However, since the seal between the earcups and the wearer's head may change continuously due to various factors during the wearing of noise-canceling headphones, existing noise-canceling solutions have high power consumption due to adaptive noise cancellation, or rely on manual switching or single threshold judgment, which can easily misjudge instantaneous actions as changes in the seal. They cannot accurately determine whether the seal has changed continuously under low power conditions and adjust the noise cancellation parameters in time, resulting in problems such as reduced noise cancellation effect, ear pressure discomfort or amplified wind noise. Summary of the Invention

[0004] This invention provides a method and system for adaptive adjustment of noise reduction parameters in Bluetooth headphones, which solves the technical problem that existing noise reduction solutions cannot accurately determine the continuous changes in the sealing state, resulting in high misjudgment rate, reduced noise reduction effect, ear pressure discomfort or amplified wind noise.

[0005] In a first aspect, the present invention provides a method for adaptive adjustment of noise reduction parameters in Bluetooth headphones, the method comprising:

[0006] Acquire the feedback microphone signal and feedforward microphone signal of the Bluetooth headset, and extract the sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone;

[0007] Based on the feedforward microphone signal and the feedback microphone signal, low-frequency coherence features are extracted;

[0008] Motion signals are acquired by a motion sensor installed in the Bluetooth headset, and the wearer is identified as being in a state of instantaneous motion based on the motion signals.

[0009] When the sealing state feature and the low-frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state, the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset is activated.

[0010] The sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state characteristics and the low-frequency coherence characteristics.

[0011] The noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjusts the noise reduction parameter according to the target noise reduction parameter.

[0012] Secondly, the present invention provides a noise reduction parameter adaptive adjustment system for Bluetooth headphones, the system comprising:

[0013] A sealing state feature extraction module is used to acquire the feedback microphone signal and the feedforward microphone signal of the Bluetooth headset, and extract sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone;

[0014] A low-frequency coherence feature extraction module is used to extract low-frequency coherence features based on the feedforward microphone signal and the feedback microphone signal;

[0015] The instantaneous motion state recognition module is used to acquire motion signals through a motion sensor installed in the Bluetooth headset, and to identify whether the wearer is in an instantaneous motion state based on the motion signals.

[0016] The adjustment component wake-up module is used to wake up the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset when the sealing state feature and the low frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state.

[0017] A sealing status level output module is used by the sealing status classification agent in the noise reduction parameter adaptive adjustment component to output the current sealing status level based on the sealing status characteristics and the low-frequency coherence characteristics.

[0018] The noise reduction parameter adjustment module is used by the noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component to match the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjust the noise reduction parameter according to the target noise reduction parameter.

[0019] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0020] This invention provides a method and system for adaptive adjustment of noise reduction parameters in Bluetooth headsets. First, it acquires the feedback microphone signal and feedforward microphone signal from the Bluetooth headset, extracts sealing state features including at least the low-frequency energy of the feedback microphone, and extracts low-frequency coherence features. Second, by setting a motion sensor within the Bluetooth headset to acquire motion signals and identify whether the wearer is in a state of instantaneous movement, it can shield against misjudgments of the sealing state when motion disturbances occur. Third, when the sealing state features and low-frequency coherence features meet preset sealing change conditions within multiple consecutive time windows, and the wearer is not in a state of instantaneous movement, the noise reduction parameter adaptive adjustment component is activated to ensure the reliability of sealing change detection. Further, the sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state features and low-frequency coherence features, providing a reliable basis for accurate matching of noise reduction parameters. Finally, the noise reduction parameter adjustment module matches the target noise reduction parameter from a preset noise reduction parameter mapping table according to the current sealing state level, smoothly transitions the current noise reduction parameter to the target noise reduction parameter, and applies it to the noise reduction filter of the Bluetooth headset, ensuring a smooth transition in noise reduction performance and listening experience while adapting to sealing changes. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method for adaptive adjustment of noise reduction parameters in a Bluetooth headset provided by the present invention.

[0022] Figure 2 This is a logical schematic diagram of a method for adaptive adjustment of noise reduction parameters in Bluetooth headphones provided by the present invention;

[0023] Figure 3 This is a schematic diagram of the structure of a Bluetooth headset noise reduction parameter adaptive adjustment system provided by the present invention.

[0024] In the attached diagram, the components represented by each number are as follows:

[0025] 11. Sealing status feature extraction module; 12. Low frequency coherence feature extraction module; 13. Instantaneous action status recognition module; 14. Adjustment component wake-up module; 15. Sealing status level output module; 16. Noise reduction parameter adjustment module. Detailed Implementation

[0026] This invention provides a method and system for adaptive adjustment of noise reduction parameters in Bluetooth headphones, which solves the technical problem that existing noise reduction solutions cannot accurately determine the continuous changes in the sealing state, resulting in high misjudgment rate, reduced noise reduction effect, ear pressure discomfort or amplified wind noise.

[0027] The present invention will now be described in detail with reference to the accompanying drawings.

[0028] Example 1, as Figure 1 , Figure 2As shown, the present invention provides a method for adaptive adjustment of noise reduction parameters in Bluetooth headphones, the method comprising:

[0029] S100: Acquire the feedback microphone signal and feedforward microphone signal of the Bluetooth headset, and extract the sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone;

[0030] In this embodiment of the invention, the feedback microphone signal is an electroacoustic signal collected by a feedback microphone located inside the earcup of the Bluetooth headset; the feedforward microphone signal is an electroacoustic signal collected by a feedforward microphone located outside the earcup of the Bluetooth headset.

[0031] Specifically, the Bluetooth headset uses its built-in feedback microphone to collect acoustic signals from the inside of the earcups, obtaining the feedback microphone signal; and uses a feedforward microphone to collect ambient noise signals from the outside of the earcups, obtaining the feedforward microphone signal. After acquiring the feedback microphone signal, the low-frequency component is extracted through low-pass filtering, and the low-frequency energy of the feedback microphone is calculated from this low-frequency component.

[0032] Step S100 in the method of this embodiment of the invention includes:

[0033] Feedback microphone signals are obtained through a feedback microphone located inside the earcups of the Bluetooth headset, and feedforward microphone signals are obtained through a feedforward microphone located outside the earcups of the Bluetooth headset.

[0034] The feedback microphone signal is low-pass filtered to obtain the low-frequency feedback microphone signal;

[0035] Based on the low-frequency signal from the feedback microphone, extract the low-frequency energy of the feedback microphone and the frequency band energy distribution in multiple frequency bands;

[0036] The low-frequency energy of the feedback microphone and the frequency band energy distribution in the multiple frequency bands are used as the characteristics of the sealed state.

[0037] In this embodiment of the invention, firstly, when the Bluetooth headset is in working condition, a feedback microphone signal is obtained through a feedback microphone located inside the earcup of the Bluetooth headset. This signal includes residual noise components after noise reduction processing by the headset and changes in acoustic characteristics caused by changes in wearing status. Then, a feedforward microphone signal is obtained through a feedforward microphone located outside the earcup of the Bluetooth headset.

[0038] Specifically, the energy of the low-frequency component in the feedback microphone signal is related to the degree of sealing of the headphones. When the headphones are well sealed to the ear canal, the low-frequency sound waves are attenuated less in the coupling cavity, and the low-frequency energy collected by the feedback microphone is relatively high. When there is a leak in the seal, the low-frequency sound waves propagate outward through the leakage path, and the low-frequency energy collected by the feedback microphone is reduced accordingly.

[0039] Secondly, the feedback microphone signal is low-pass filtered to obtain the low-frequency signal from the feedback microphone. Low-pass filtering is a signal processing operation that allows signal components below the cutoff frequency to pass through while attenuating signal components above the cutoff frequency.

[0040] Specifically, a preset cutoff frequency is used, which can be set according to the actual application scenario, for example, 500 Hz. The feedback microphone signal is input to a low-pass filter, which filters the feedback microphone signal, preserving signal components with frequencies below the cutoff frequency and significantly attenuating signal components with frequencies above the cutoff frequency. Since changes in the headphone's sealing condition mainly affect the transmission characteristics of low-frequency acoustic signals, while having a relatively small impact on high-frequency signals, extracting the low-frequency components from the feedback microphone signal can more accurately reflect changes in the sealing condition.

[0041] Next, based on the low-frequency signal from the feedback microphone, the low-frequency energy of the feedback microphone and the frequency band energy distribution across multiple frequency bands are extracted. The low-frequency energy of the feedback microphone is a time-domain energy measure of the low-frequency signal, used to quantify the overall intensity level of the signal. The frequency band energy distribution describes the distribution of signal energy within each sub-band after dividing the spectrum of the low-frequency signal from the feedback microphone into multiple frequency sub-bands, used to characterize the energy features of the signal in different frequency ranges.

[0042] Finally, the extracted low-frequency energy from the feedback microphone and the frequency band energy distributions on multiple frequency bands are combined to form a sealed state feature vector.

[0043] Step S100 in this embodiment of the invention involves extracting the low-frequency energy of the feedback microphone and the frequency band energy distribution across multiple frequency bands based on the low-frequency signal from the feedback microphone, including:

[0044] The low-frequency signal of the feedback microphone is divided into multiple short frames, and the energy of the low-frequency signal of the feedback microphone in each short frame is calculated to obtain the low-frequency energy sequence of the feedback microphone.

[0045] Based on the low-frequency energy sequence of the feedback microphone, calculate the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude between adjacent short frames, and use the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude as the low-frequency energy of the feedback microphone.

[0046] Spectral analysis is performed on the low-frequency signal of the feedback microphone to obtain the frequency band energy distribution of the low-frequency signal of the feedback microphone in multiple frequency bands, and the frequency band energy distribution in the multiple frequency bands is taken as the frequency band energy distribution.

[0047] In this embodiment of the invention, firstly, the low-frequency signal of the feedback microphone is divided into multiple short-time frames, and the energy of the low-frequency signal of the feedback microphone in each short-time frame is calculated to obtain the low-frequency energy sequence of the feedback microphone.

[0048] Specifically, the low-frequency signal from the feedback microphone is divided in the time domain to obtain multiple short frames. Each short frame contains a preset number of signal sampling points, for example, 256 sampling points per short frame. A preset overlap length can be set between adjacent short frames, such as 50% overlap of sampling points. For each short frame, the mean square energy of the low-frequency signal from the feedback microphone within that frame is calculated to obtain the corresponding low-frequency energy value for that frame.

[0049] Preferably, for each short frame after framing, the time-domain energy of the low-frequency signal from the feedback microphone within that short frame is calculated. For N sampling points within a short frame, the low-frequency energy value E is calculated by squaring the amplitude of each sampling point and then summing the results.

[0050]

[0051] Where, x i Let be the amplitude of the i-th sampling point. Arrange the low-frequency energy values ​​of all short frames in chronological order to form a feedback microphone low-frequency energy sequence. Then, calculate the time average of this feedback microphone low-frequency energy sequence: the time average of the feedback microphone low-frequency energy = the sum of the feedback microphone low-frequency energies / the number of feedback microphone low-frequency energies.

[0052] Secondly, based on the low-frequency energy sequence of the feedback microphone, the absolute value or square of the difference between the low-frequency energy values ​​between adjacent short time frames is calculated, and the values ​​are accumulated and averaged to obtain the energy fluctuation amplitude. This energy fluctuation amplitude is used to characterize the degree of drastic change in the sealing state over time.

[0053] Preferably, for the energy fluctuation amplitude, the energy change between adjacent short-time frames is calculated, and the energy change ΔE is calculated. j =|E j+1 -E j | where j = 1, 2, ..., M-1, and then the arithmetic mean of all energy changes is taken to obtain the average energy fluctuation amplitude E. f .

[0054]

[0055] Where M-1 represents the number of energy changes. Subsequently, the time average value of the low-frequency energy of the feedback microphone and the amplitude of energy fluctuations are combined to form the low-frequency energy of the feedback microphone.

[0056] Finally, spectral analysis was performed on the low-frequency signal from the feedback microphone to obtain the frequency band energy distribution of the low-frequency signal in multiple frequency bands, and the frequency band energy distribution in multiple frequency bands was taken as the frequency band energy distribution.

[0057] Specifically, a spectral analysis is performed on the low-frequency signal from the feedback microphone, transforming it from the time domain to the frequency domain. Spectral analysis can be achieved using a Fast Fourier Transform (FFT). After the FFT, the spectral amplitude of the low-frequency signal from the feedback microphone at each frequency point is obtained. Subsequently, the frequency axis is divided into multiple frequency bands. For each frequency band, the squared values ​​of the spectral amplitude at all frequency points within that band are summed to obtain the energy value of that band. The energy values ​​of all frequency bands are combined to form the frequency band energy distribution.

[0058] For example, assuming that the preset frequency band 1 is 20Hz to 50Hz, frequency band 2 is 50Hz to 100Hz, and frequency band 3 is 100Hz to 200Hz, after the above spectrum analysis and energy accumulation, the frequency band energy vector of a certain short frame is [1200, 800, 300]. After time averaging of all short frames, the final frequency band energy distribution vector is [1150, 780, 310]. This vector is an important component of the sealed state characteristics.

[0059] In this embodiment of the invention, low-frequency energy and multi-band energy distribution of the feedback microphone are extracted by low-pass filtering, the changes in low-frequency sound pressure inside the earcup are quantified, and the energy fluctuation amplitude and frequency band distribution are combined to provide feature input for the sealed state classification agent, thereby improving the response accuracy of the adaptive adjustment of noise reduction parameters.

[0060] S200: Extract low-frequency coherence features based on the feedforward microphone signal and the feedback microphone signal;

[0061] In this embodiment of the invention, the low-frequency coherence characteristic is a signal coherence metric calculated based on the feedforward microphone signal and the feedback microphone signal in the low-frequency range. In the field of signal processing, coherence is a quantitative indicator used to evaluate the degree of correlation between two signals in the frequency domain, and its value ranges from 0 to 1.

[0062] Specifically, after acquiring the feedforward and feedback microphone signals, a joint analysis is performed on these two signals. Low-pass filtering is applied to both signals to extract the low-frequency components. Time alignment and frame segmentation are then performed on the two low-frequency signals, and the coherence values ​​of the feedforward and feedback microphone low-frequency signals are calculated. The coherence values ​​from multiple short frames are then averaged over time to obtain the low-frequency coherence characteristic values.

[0063] Step S200 in the method of this embodiment of the invention includes:

[0064] The feedforward microphone signal and the feedback microphone signal are respectively low-pass filtered to obtain the feedforward microphone low-frequency signal and the feedback microphone low-frequency signal;

[0065] The low-frequency signals from the feedforward microphone and the low-frequency signals from the feedback microphone are aligned in time and divided into multiple short time frames;

[0066] For each short frame, a low-frequency coherence value is calculated based on the low-frequency signal from the feedforward microphone and the low-frequency signal from the feedback microphone;

[0067] The low-frequency coherence characteristics are obtained by averaging the low-frequency coherence values ​​of multiple short frames over time.

[0068] In this embodiment of the invention, firstly, a cutoff frequency is preset, for example, it can be set to 500 Hz. Then, the feedforward microphone signal is input into a low-pass filter, and the feedback microphone signal is input into a low-pass filter. The feedforward microphone signal and the feedback microphone signal are low-pass filtered respectively to obtain the feedforward microphone low-frequency signal and the feedback microphone low-frequency signal.

[0069] Secondly, the low-frequency signals from the feedforward microphone and the low-frequency signals from the feedback microphone are aligned in time.

[0070] Preferably, a cross-correlation method can be used to calculate the cross-correlation function between the low-frequency signal from the feedforward microphone and the low-frequency signal from the feedback microphone. The delay corresponding to the maximum value of the cross-correlation function is then used as the time delay compensation value, and the two signals are synchronized based on this compensation value. After time alignment, the low-frequency signals from the feedforward microphone and the feedback microphone are divided into multiple short-time frames.

[0071] Each short frame contains a preset number of signal sampling points, for example, each short frame can be set to contain 256 sampling points. A preset overlap length can be set between adjacent short frames, for example, 50% overlap of sampling points. After the framing process is completed, each short frame contains a set of time-continuous feedforward microphone low-frequency signal sampling points and feedback microphone low-frequency signal sampling points that are time-aligned with them.

[0072] Next, for each short frame, the low-frequency coherence value is calculated based on the low-frequency signals from the feedforward microphone and the feedback microphone. The low-frequency coherence value is the average of the coherence function values ​​between the frequency components of the two signals within the low-frequency range over the frequency band. The coherence function is a function in the frequency domain that describes the correlation between two signals, and its value ranges from 0 to 1: when the coherence function value is close to 1, it indicates that the two signals have a high linear correlation at that frequency; when the coherence function value is close to 0, it indicates that the two signals lack linear correlation at that frequency.

[0073] Specifically, for each short frame, spectral analysis is performed on the low-frequency signals from the feedforward microphone and the feedback microphone within that frame. For example, a Fast Fourier Transform is used to transform the two signals from the time domain to the frequency domain, obtaining the spectra of the feedforward microphone low-frequency signal and the feedback microphone low-frequency signal. Subsequently, the low-frequency coherence value of that frame is calculated based on the spectra of the two signals.

[0074] Preferably, for multiple frequency points, the coherence function value at each frequency point is calculated, and the coherence function values ​​at each frequency point are averaged within a preset low-frequency range to obtain the low-frequency coherence value of the frame. Low-frequency coherence value = (1 / total number of frequency points in the low-frequency range) × Σ[cross-power spectral density of two signals at the k-th frequency point / (auto-power spectral density of the low-frequency signal of the feedforward microphone × auto-power spectral density of the low-frequency signal of the feedback microphone)].

[0075] Finally, the low-frequency coherence values ​​of multiple short-time frames are averaged over time, that is, the low-frequency coherence values ​​of all short-time frames are summed, and then the sum is divided by the number of short-time frames to obtain the low-frequency coherence characteristics.

[0076] Specifically, when the headphones are well-sealed, ambient noise is primarily collected by the feedforward microphone, processed by the feedforward noise reduction path, and then fed into the inside of the earcups. The coherence between the feedforward signal and the feedback signal is low. However, when the headphones experience a seal leak, ambient noise is directly transmitted into the inside of the earcups through the leakage path. This results in the feedback microphone signal containing more components related to ambient noise, and the coherence between the feedforward signal and the feedback signal increases accordingly. Therefore, low-frequency coherence characteristics can effectively reflect changes in the headphone's seal; the worse the seal, the higher the low-frequency coherence characteristic value.

[0077] In this embodiment of the invention, the feedforward microphone signal and the feedback microphone signal are low-pass filtered respectively, and the low-frequency coherence value is calculated frame by frame and then averaged in the time domain. This quantifies the degree of coherence of external noise coupled into the ear through the leakage path, enhances the ability of the sealing state classification agent to distinguish between slight leakage and obvious leakage, and improves the accuracy and reliability of the adaptive adjustment of noise reduction parameters.

[0078] S300: Acquires motion signals through a motion sensor installed in the Bluetooth headset, and identifies whether the wearer is in a state of instantaneous motion based on the motion signals;

[0079] In this embodiment of the invention, the motion sensor is a sensor device installed inside the Bluetooth headset for detecting the wearer's motion state; the motion signal is an electrical signal output by the motion sensor, including acceleration signal and angular velocity signal; the instantaneous action state refers to the sudden, large-amplitude body or head movement state that occurs in a short period of time.

[0080] Specifically, the motion sensor inside the Bluetooth headset collects acceleration signals in real time, and the gyroscope collects angular velocity signals in real time. Then, the acceleration amplitude is calculated, taking either the combined amplitude of the three-axis angular velocities or the maximum value of each axis's amplitude. The acceleration and angular velocity amplitudes are compared with preset action thresholds to determine whether the wearer is in a state of instantaneous motion.

[0081] Step S300 in the method of this embodiment of the invention includes:

[0082] Acceleration signals are obtained by an accelerometer installed in the Bluetooth headset, and angular velocity signals are obtained by a gyroscope installed in the Bluetooth headset.

[0083] The acceleration amplitude is calculated based on the acceleration signal, and the angular velocity amplitude is calculated based on the angular velocity signal.

[0084] When the acceleration amplitude or the angular velocity amplitude exceeds a preset action threshold within a preset action detection time window, it is determined that the wearer is in a momentary action state.

[0085] When the acceleration amplitude and the angular velocity amplitude do not exceed the preset action threshold within the preset action detection time window, it is determined that the wearer is not in a state of instantaneous action.

[0086] In this embodiment of the invention, firstly, an inertial measurement unit is integrated inside the Bluetooth headset, which includes at least a three-axis accelerometer and a three-axis gyroscope. Then, an acceleration signal is acquired through the accelerometer inside the Bluetooth headset; this signal typically includes gravitational acceleration components and dynamic motion acceleration components. Furthermore, an angular velocity signal is acquired through the gyroscope inside the Bluetooth headset; this signal reflects the rate of rotation of the head around three orthogonal axes: the pitch axis, the roll axis, and the yaw axis.

[0087] Secondly, the acceleration amplitude is calculated based on the acceleration signal, and the angular velocity amplitude is calculated based on the angular velocity signal.

[0088] Specifically, the triaxial components of the acceleration signal are acquired at multiple sampling times within a preset time window. Let t be a sampling time, and the triaxial acceleration components output by the accelerometer be a... x a y and a z These three axial components correspond to three mutually orthogonal spatial directions. Based on these three axial acceleration components, the instantaneous acceleration amplitude A at that moment is calculated using the formula: A = sqrt[a x 2 +a y 2 +a z2Subsequently, all instantaneous acceleration amplitudes within the preset motion detection time window are smoothed. The average or root mean square value of all the instantaneous acceleration amplitudes is calculated to obtain the final acceleration amplitude. Preferably, the acceleration amplitude A... w =sqrt[(1 / N)×ΣA 2 ], where N is the number of instantaneous acceleration amplitudes.

[0089] Similarly, for the angular velocity signal output by the gyroscope, suppose that at a certain sampling time t, the three-axis angular velocity components output by the gyroscope are g x g y and g z These correspond to the rotational angular velocities about the three orthogonal axes, respectively. The instantaneous angular velocity amplitude G at this moment is calculated using the formula: G = sqrt[g x 2 +g y 2 +g z 2 Subsequently, within a preset motion detection time window, all instantaneous angular velocity amplitudes within that time window are smoothed. After smoothing, preferably, the root mean square value of the instantaneous angular velocity amplitudes within that time window is calculated to obtain the final angular velocity amplitude, G. w =sqrt[(1 / N)×ΣG 2 ], where N is the number of instantaneous acceleration amplitudes.

[0090] Finally, preset action thresholds are set, allowing for separate preset action thresholds to be set for either the acceleration amplitude or the angular velocity amplitude. The obtained acceleration amplitude or angular velocity amplitude is then compared with the preset action thresholds.

[0091] Preferably, empirical values ​​are pre-calibrated based on statistical analysis of motion signals from different movement modes, such as walking, running, turning the head, speaking, and chewing, to distinguish between minute physiological vibrations during normal wear and significant movements that may alter the earphone's seal. For example, 5000 motion signals from different movement modes are collected, statistically analyzed, and their mean values ​​are calculated. The mean values ​​of acceleration amplitude or angular velocity amplitude are then used as corresponding preset motion thresholds. For instance, the preset motion threshold for acceleration amplitude can be set to 12 m / s², which is slightly greater than gravitational acceleration to exclude the influence of the gravitational component under static posture, primarily targeting vigorous linear movements such as head shaking and jumping. The preset motion threshold for angular velocity amplitude can be set to 1.5 rad / s to exclude slow head rotations, primarily targeting vigorous rotational movements such as rapid head turning.

[0092] When the acceleration amplitude or angular velocity amplitude exceeds the preset action threshold within the preset action detection time window, it indicates that the wearer's head has undergone a high-intensity linear or rotational movement within that time window, which is sufficient to cause the earphone to shift or loosen in the ear canal. In this state, the microphone signal collected will deviate significantly from the stable wearing state and will not meet the conditions for evaluating changes in the sealing state. Therefore, it is determined that the wearer is in a momentary action state.

[0093] Conversely, when the acceleration amplitude and angular velocity amplitude do not exceed the preset action threshold within the preset action detection time window, it indicates that the wearer's head is in a relatively still or only slightly moving state. In this state, the coupling relationship between the earphone and the ear canal remains stable, and the microphone signal can truly reflect the sealed state. Therefore, the wearer is not in a state of instantaneous action.

[0094] In this embodiment of the invention, acceleration signals are obtained by an accelerometer and angular velocity signals are obtained by a gyroscope. The acceleration amplitude and angular velocity amplitude are calculated and combined with a preset action detection time window and a preset action threshold to determine the instantaneous action state. When motion disturbance occurs, the misjudgment of the sealing state is shielded, avoiding invalid wake-up of the noise reduction parameter adaptive adjustment component due to head shaking or wearing action, thereby improving anti-interference ability and parameter adjustment stability.

[0095] S400: When the sealing state feature and the low-frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state, the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset is activated.

[0096] Instantaneous movements such as chewing, speaking, turning the head, and looking down cause brief relative displacement between the Bluetooth headset earcups and the head, altering the low-frequency energy of the feedback microphone and the low-frequency coherence of the feedforward and feed-forward microphones. For example, when a user is chewing gum, the Bluetooth headset is moved by facial muscles, causing fluctuations in the low-frequency energy of the feedback microphone; when the user suddenly turns their head, the earcups briefly move away from the head, changing the low-frequency coherence.

[0097] However, these changes caused by instantaneous actions are temporary and recoverable, and do not represent a genuine and sustained change in the sealing state of the Bluetooth headset. While the sealing state characteristics and low-frequency coherence characteristics may temporarily meet the preset sealing change conditions during an instantaneous action, this change is likely caused by the instantaneous action itself, rather than by a continuous deterioration of the sealing state. In this case, waking up the noise reduction parameter adaptive adjustment component would not only cause frequent false wake-ups but may also lead the sealing state classification agent to output an incorrect sealing state level based on the disturbed features, thereby matching incorrect noise reduction parameters.

[0098] Therefore, the instantaneous action state must be used as an exclusion condition: only when the sealing state characteristics and low-frequency coherence characteristics meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in an instantaneous action state, can it be confirmed that the sealing state has continuously changed, and the noise reduction parameter adaptive adjustment component can be activated for subsequent classification and parameter adjustment.

[0099] Step S400 in the method of this embodiment of the invention includes:

[0100] The deviation of the sealing state is determined based on the difference between each feature component in the sealing state characteristics and the corresponding feature component in the preset sealing state baseline, wherein the preset sealing state baseline is determined based on the historical sealing state characteristics of the Bluetooth headset in the normal sealing state.

[0101] The coherence deviation is determined based on the difference between the low-frequency coherence characteristics and the preset coherence baseline, wherein the preset coherence baseline is determined based on the historical low-frequency coherence characteristics of the Bluetooth headset under normal sealed conditions.

[0102] When the deviation of the sealing state exceeds the preset sealing deviation threshold for multiple consecutive time windows, and the deviation of the coherence exceeds the preset coherence deviation threshold for multiple consecutive time windows, and the motion signal does not indicate that the wearer is in an instantaneous action state, it is determined that the preset sealing change condition is met.

[0103] After the preset sealing change conditions are met, a preset confirmation timer is started, and the sealing state characteristics and the low-frequency coherence characteristics are monitored within the preset confirmation timer.

[0104] When the sealing state characteristics and the low-frequency coherence characteristics continuously meet the preset sealing change conditions within the preset confirmation time, the noise reduction parameter adaptive adjustment component is activated.

[0105] When the sealing state feature or the low-frequency coherence feature no longer meets the preset sealing change condition within the preset confirmation time, the preset confirmation time is reset and the noise reduction parameter adaptive adjustment component is not activated.

[0106] In this embodiment of the invention, firstly, the deviation of the sealing state is determined based on the difference between each feature component in the sealing state feature and the corresponding feature component in the preset sealing state baseline. The preset sealing state baseline is determined based on the historical sealing state features of the Bluetooth headset in the normal sealing state.

[0107] Specifically, the sealing state characteristics include at least the time average value and energy fluctuation amplitude of the low-frequency energy of the feedback microphone. Therefore, the preset sealing state baseline may include the preset time average value and preset energy fluctuation amplitude of the low-frequency energy of the feedback microphone.

[0108] Preferably, the low-frequency signal from the feedback microphone is divided into multiple consecutive short-time frames according to a preset frame length. For each short-time frame, the sum of the squares of the signal amplitude within that frame is calculated as the short-time energy of that frame, resulting in a short-time energy sequence. Subsequently, the arithmetic mean of the short-time energy sequence is calculated to obtain the time average of the low-frequency energy of the feedback microphone. The energy fluctuation amplitude is used to characterize the drastic degree of low-frequency energy change between adjacent short-time frames. Based on the short-time energy sequence, the average of the absolute values ​​of the energy differences between adjacent short-time frames is calculated to obtain the energy fluctuation amplitude.

[0109] Subsequently, the Euclidean distance or weighted Euclidean distance between each feature component in the current sealing state characteristics and the feature components corresponding to the preset time average value and preset energy fluctuation amplitude of the low-frequency energy of the feedback microphone in the preset sealing state baseline is calculated to obtain the sealing state deviation. Here, w1 and w2 are preset weighting coefficients for the preset time average value and preset energy fluctuation amplitude, respectively, used to adjust the contribution of different feature components to the deviation. Preferably, the sealing state deviation = sqrt[w1 × ((current energy fluctuation amplitude - preset energy fluctuation amplitude) / preset energy fluctuation amplitude)] 2 +w2×((Current time average - Preset time average) / Preset time average) 2 [The preset energy fluctuation amplitude and preset time average value are not 0.]

[0110] Secondly, the coherence deviation is determined based on the difference between the low-frequency coherence characteristics and the preset coherence baseline. The preset coherence baseline is determined based on the historical low-frequency coherence characteristics of the Bluetooth headset under normal sealed conditions. The historical low-frequency coherence characteristics can be obtained by calculating the coherence function of the aligned feedforward microphone low-frequency signal and the feedback microphone low-frequency signal. The value ranges from 0 to 1, and the closer the value is to 1, the stronger the correlation between the two signals.

[0111] Preferably, multiple historical low-frequency coherence features are obtained, and then the mean of the historical low-frequency coherence is calculated through statistical analysis. The mean is then used as a preset coherence baseline.

[0112] Specifically, the coherence deviation is calculated as: current low-frequency coherence characteristic value minus coherence baseline value. When a leak occurs in the sealed state, external noise enters the earcup directly through the leakage path, causing the feedback microphone signal to contain more components related to external environmental noise, thus significantly increasing the coherence between the feedforward microphone signal and the feedback microphone signal in the low-frequency range.

[0113] Furthermore, when the deviation of the sealing state continuously exceeds the preset sealing deviation threshold within multiple consecutive time windows, and the coherence deviation continuously exceeds the preset coherence deviation threshold within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in an instantaneous action state, it is determined that the preset sealing change condition is met.

[0114] The setting of multiple consecutive time windows is used to avoid false triggering caused by instantaneous noise fluctuations or signal anomalies, thereby improving the reliability of the criteria. For example, the number of consecutive judgment time windows is set to M=3, and the length of each time window is T=0.5 seconds. That is, within a continuous time of 1.5 seconds, the deviation of the sealing state and the deviation of coherence calculated by each time window must exceed their respective thresholds before the preset sealing change condition is considered to be met.

[0115] Specifically, the preset sealing deviation threshold and the preset coherence deviation threshold can be obtained from historical data statistical analysis and used to distinguish between normal fluctuations and true changes in sealing status. Preferably, the sample sets of sealing status deviation and coherence deviation under each state are extracted according to the aforementioned method, and outlier removal is performed on each sample set. If the distribution follows a normal distribution, the mean is used to set the threshold, that is, the product of the mean and standard deviation of the deviation samples under good sealing conditions is used as the corresponding threshold.

[0116] Furthermore, after the preset sealing change conditions are met, a preset confirmation timer is started, and the sealing state characteristics and low-frequency coherence characteristics are continuously monitored within the preset confirmation timer. The preset confirmation timer provides an additional confirmation window after the initial determination that the sealing change conditions are met, to prevent false wake-ups caused by a rapid recovery of the state immediately after the conditions are met. Preferably, the preset confirmation timer is set to 2 seconds, that is, the timing starts after the conditions are initially met, and the sealing state characteristics and low-frequency coherence characteristics are continuously monitored.

[0117] Finally, when the sealing state characteristics and low-frequency coherence characteristics continuously meet the preset sealing change conditions within the preset confirmation time, the noise reduction parameter adaptive adjustment component is activated. Preferably, if the sealing state deviation and coherence deviation both continuously exceed the preset sealing deviation threshold at every moment within the 2-second confirmation time, then the sealing state is confirmed to have indeed changed. At this point, the noise reduction parameter adaptive adjustment component is activated to perform subsequent sealing state level judgment and noise reduction parameter adjustment.

[0118] When the sealing condition characteristics or low-frequency coherence characteristics no longer meet the preset sealing change conditions within the preset confirmation time, the preset confirmation time is reset and the noise reduction parameter adaptive adjustment component is not activated.

[0119] Preferably, if the deviation of the sealing state falls below the preset sealing deviation threshold, or the deviation of the coherence falls below the preset coherence deviation threshold, or the motion signal indicates that the wearer is in a momentary motion state, then the preset confirmation timer is reset and the noise reduction parameter adaptive adjustment component is not activated.

[0120] In this embodiment of the invention, the deviation of the sealing state and the deviation of coherence are calculated. Combined with continuous monitoring of multiple consecutive time windows and preset confirmation timing, transient noise and random fluctuations are effectively filtered out. At the same time, motion signals are used to exclude instantaneous action states, avoiding false wake-up of the noise reduction parameter adaptive adjustment component due to non-continuous events. Under the premise of ensuring reliable detection of sealing changes, the misadjustment rate and power consumption are reduced, and the stability of adaptive adjustment is improved.

[0121] S500: The sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state characteristics and the low-frequency coherence characteristics;

[0122] In this embodiment of the invention, the sealing state classification agent is an artificial intelligence model deployed in the noise reduction parameter adaptive adjustment component; the sealing state level is the result of quantitatively classifying the current sealing degree of the Bluetooth headset.

[0123] Specifically, after the noise reduction parameter adaptive adjustment component is activated, the sealing state classification agent receives sealing state features and low-frequency coherence features as input. These features are passed through the agent's input layer, undergo nonlinear transformation and feature extraction in the hidden layer, and finally the output layer outputs the probability values ​​of each sealing state level. The sealing state level with the highest probability is output as the current sealing state level.

[0124] Step S500 in the method of this embodiment of the invention includes:

[0125] The sealing state features and the low-frequency coherence features are input into a pre-trained sealing state classification agent, which outputs the probability of each sealing state level based on the sealing state features and the low-frequency coherence features.

[0126] The sealing condition level with the highest probability is taken as the current sealing condition level, which includes good sealing, slight leakage, and significant leakage.

[0127] The pre-training process of the sealed state classification agent includes:

[0128] Multiple historical samples are acquired, each of which includes historical sealing state features, historical low-frequency coherence features, and corresponding sealing state labels;

[0129] A sealing state classification agent is constructed based on a lightweight multi-classification model. The sealing state classification agent includes an input layer, a hidden layer, and an output layer. The input layer receives sealing state features and low-frequency coherence features, and the output layer outputs the probability of each sealing state level.

[0130] Using the historical sealing state features and the historical low-frequency coherence features as input features, and the sealing state label as the supervision label, the sealing state classification agent is trained in a supervised manner until the verification convergence, thus obtaining the pre-trained sealing state classification agent.

[0131] In this embodiment of the invention, firstly, the sealing state features and low-frequency coherence features are input into a pre-trained sealing state classification agent. The sealing state classification agent is a machine learning model built on a lightweight multi-classification model, which has learned the mapping relationship from the feature space to the sealing state level through a large number of historical samples.

[0132] Subsequently, the sealing state classification agent outputs the probability of each sealing state level based on the sealing state features and low-frequency coherence features. Upon receiving the input sealing state features and low-frequency coherence features, the sealing state classification agent performs a forward propagation calculation and outputs a probability vector. Each element in this probability vector corresponds to a probability value for a sealing state level, and the sum of all probability values ​​equals the value 1.

[0133] Secondly, the maximum value is retrieved from the probability vector, and the sealing status level corresponding to this maximum value is the current sealing status level derived by the model. The sealing status levels include good sealing, minor leakage, and significant leakage.

[0134] Specifically, a good seal indicates that the low-frequency energy of the feedback microphone is within the normal range and the low-frequency coherence is low, indicating that the coupling between the earphone and the ear canal is tight, providing acoustic boundary conditions for the noise reduction algorithm; slight leakage indicates that the low-frequency energy of the feedback microphone has a certain degree of attenuation and the low-frequency coherence is moderate, indicating that there is a small gap between the earphone and the ear canal, resulting in some low-frequency noise leakage; obvious leakage indicates that the low-frequency energy of the feedback microphone is significantly reduced and the low-frequency coherence is high, indicating that there is serious air leakage in the earphone, the acoustic boundary conditions have changed drastically, and the noise reduction performance has dropped significantly.

[0135] The pre-training process of the sealed state classification agent includes:

[0136] First, multiple historical samples are acquired. Each historical sample includes historical sealing state features, historical low-frequency coherence features, and a corresponding sealing state label. The historical samples originate from a pre-constructed training dataset containing a large number of data records collected under different sealing states. Preferably, the aforementioned historical sealing state features and historical low-frequency coherence features are numerical vectors extracted from the feedback microphone signals and feedforward microphone signals of the historical records using the same method as described in the preceding steps. The corresponding sealing state label represents the actual sealing state level of the corresponding data record; this label is determined either manually or using high-precision measurement equipment.

[0137] Preferably, by simulating three states—good sealing, slight leakage, and significant leakage—thousands of feedback microphone signals and feedforward microphone signals are collected for each state, and the corresponding sealing state features and low-frequency coherence features are extracted. Manually labeled sealing state tags are recorded. All collected samples are divided into training and validation sets according to a certain ratio. For example, a total of 3000 samples are collected, with 1000 samples each for the three states: good sealing, slight leakage, and significant leakage. 2400 samples are used as the training set for learning model parameters, and the remaining 600 samples are used as the validation set for evaluating model convergence and adjusting hyperparameters.

[0138] Secondly, a sealing state classification agent is constructed based on a lightweight multi-classification model. The sealing state classification agent includes an input layer, a hidden layer, and an output layer. The input layer receives sealing state features and low-frequency coherence features, and the output layer outputs the probability of each sealing state level.

[0139] Specifically, the input layer receives the aforementioned sealing state features and low-frequency coherence features, with the number of nodes equal to the total dimension of the input features. Assuming the sealing state features include low-frequency energy from the feedback microphone and the frequency band energy distribution across K frequency bands, the sealing state feature dimension is K+2; the low-frequency coherence feature has one component, therefore the input layer has K+3 nodes. The input layer passes the received multi-dimensional feature vector to the next layer as is, without performing nonlinear transformations on the data. The hidden layer extracts high-order combined features from the input features. The hidden layer contains multiple neurons; each neuron performs a weighted summation of the input from the previous layer, adds a bias term, and then outputs the result after mapping by a nonlinear activation function. The number of neurons is set to 32, and a modified linear unit is used as the activation function, which sets all negative inputs to zero while keeping positive values ​​unchanged. The output layer outputs the probability distribution of the current input feature belonging to each sealing state level. It has 3 nodes and uses a flexible maximum function as the activation function, converting the original score of the output layer into a probability distribution.

[0140] For example, suppose the input layer has 7 nodes, the hidden layer has 32 nodes, and the output layer has 3 nodes. Then the trainable parameters of this model include a 7×32 dimension weight matrix from the input layer to the hidden layer, totaling 224 weight parameters, and a corresponding bias vector with a dimension of 32.

[0141] Finally, using historical sealing state features and historical low-frequency coherence features as input features, and sealing state labels as supervision labels, the sealing state classification agent is trained in a supervised manner until the verification convergence, thus obtaining the pre-trained sealing state classification agent.

[0142] Specifically, the training dataset is divided into training and validation sets according to a preset ratio, such as 7:3 or 8:2. The batch size is set to 32, meaning that 32 samples are randomly selected from the training set in each iteration to calculate the loss function and update the parameters. The initial learning rate is 0.001, which is the step size coefficient for parameter updates. The maximum number of training epochs is 100. A multi-class cross-entropy loss function is used as the loss function. Based on the probability vector processed by the flexible maximum function in the model output layer, the predicted probability value of the class corresponding to the true label of the sample is extracted. The natural logarithm of this probability value is taken and then negative, i.e., L = -ln(p), where p is the predicted probability corresponding to the true class. When the predicted probability is closer to 1, the loss value is closer to 0, indicating a more accurate prediction; when the predicted probability is closer to 0, the loss value tends to infinity, indicating a larger prediction bias.

[0143] During training, forward propagation is performed, where the input feature vector is multiplied by the weight matrix from the input layer to the hidden layer and a bias term is added to obtain the weighted input of the hidden layer. This weighted input is then processed by the modified linear unit activation function to obtain the output vector of the hidden layer. The output vector of the hidden layer is then multiplied by the weight matrix from the hidden layer to the output layer and a bias term is added to obtain the original score vector of the output layer. This score vector is then activated by the flexible maximum function to obtain the final predicted probability distribution.

[0144] Then, backpropagation is performed based on the chain rule: first, the partial derivative of the loss function with respect to the input of the output layer is calculated; then, this partial derivative is propagated to the hidden layers to calculate the partial derivatives of the loss function with respect to the hidden layer parameters and the input, and so on, layer by layer, until the input layer. The calculated gradient values ​​are used to indicate how each parameter should be adjusted to reduce the value of the loss function.

[0145] In each iteration step, based on the gradient values ​​calculated by backpropagation, the adaptive moment estimator (IME) synchronously updates all trainable parameters in the model. This update is completed simultaneously within one iteration step. A training epoch is considered complete when all batches of data in the training set have been fed into the model for a single parameter update. After each training epoch, the current model is evaluated using validation set data, and the average loss on the validation set is calculated to monitor the model's convergence and generalization ability. During training, training is stopped when the average loss on the validation set falls below a preset convergence threshold of 0.02, or when the preset maximum number of training epochs (100 epochs) is reached.

[0146] For example, the 2400 training samples are divided into 75 batches with a batch size of 32. During training 30 to 38, if the average loss on the validation set (0.18) exceeds the preset loss convergence threshold of 0.02, training stops and ends. Finally, the model parameters at 28th epoch are saved as the parameters of the pre-trained model, at which point the model has achieved optimal performance on the validation set.

[0147] In this embodiment of the invention, a pre-trained lightweight multi-classification model is used to accurately output the probability of each sealing state level after supervised training, taking sealing state features and low-frequency coherence features as inputs. The model with the highest probability is selected as the current sealing state level, providing a reliable sealing state discrimination result for the noise reduction parameter adjustment module.

[0148] S600: The noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjusts the noise reduction parameter according to the target noise reduction parameter.

[0149] In this embodiment of the invention, the noise reduction parameter adjustment module is a functional module deployed in the noise reduction parameter adaptive adjustment component; the preset noise reduction parameter mapping table is a data table pre-stored in the Bluetooth headset memory; the target noise reduction parameter is a set of noise reduction related parameters that match the current sealing status level; the feedforward noise reduction weight and the feedback noise reduction weight refer to the weight coefficients of the feedforward noise reduction filter and the feedback noise reduction filter in the hybrid noise reduction system, respectively.

[0150] Specifically, after the sealing status classification agent outputs the current sealing status level, the noise reduction parameter adjustment module obtains this level information. Using the current sealing status level as an index, it queries a preset noise reduction parameter mapping table to match the corresponding target noise reduction parameter. Then, it compares the matched target noise reduction parameter with the currently used noise reduction parameter, smoothly transitioning the current noise reduction parameter to the target noise reduction parameter within a preset transition time. Finally, the updated target noise reduction parameter is applied to the noise reduction filter of the Bluetooth headset, enabling the active noise cancellation system to perform noise reduction processing according to the new parameter configuration.

[0151] Step S600 in the method of this embodiment of the invention includes:

[0152] The noise reduction parameter adjustment module obtains the current sealing status level;

[0153] The noise reduction parameter adjustment module matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level. The preset noise reduction parameter mapping table stores the correspondence between the sealing status level and the target noise reduction parameter. The target noise reduction parameter includes the target feedforward noise reduction weight, the target feedback noise reduction weight, the target noise reduction depth, and the target low-frequency compensation curve.

[0154] The noise reduction parameter adjustment module compares the target noise reduction parameter with the current noise reduction parameter, smoothly transitions the current noise reduction parameter to the target noise reduction parameter within a preset transition time, and applies the target noise reduction parameter to the noise reduction filter of the Bluetooth headset.

[0155] In this embodiment of the invention, firstly, the noise reduction parameter adjustment module obtains the current sealing status level output by the sealing status classification agent.

[0156] Secondly, the noise reduction parameter adjustment module matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level. The preset noise reduction parameter mapping table is a data structure that is pre-stored in the non-volatile memory of the Bluetooth headset. It is used to store the correspondence between the sealing status level and the target noise reduction parameter. The target noise reduction parameter includes the target feedforward noise reduction weight, the target feedback noise reduction weight, the target noise reduction depth, and the target low-frequency compensation curve.

[0157] Specifically, the target feedforward noise reduction weight is used to characterize the degree of participation of the feedforward noise reduction path in hybrid noise reduction. The larger the value, the stronger the control effect of the feedforward microphone signal on the noise reduction filter. The target feedback noise reduction weight is used to characterize the degree of participation of the feedback noise reduction path in hybrid noise reduction. The larger the value, the stronger the control effect of the feedback microphone signal on the noise reduction filter. The target noise reduction depth is used to characterize the expected attenuation amplitude of the overall noise reduction system across the entire frequency band. The target low-frequency compensation curve is used to characterize the shape of the response curve of the compensation gain for residual noise after noise reduction as a function of frequency in the low-frequency band.

[0158] Finally, the noise reduction parameter adjustment module compares the target noise reduction parameter with the current noise reduction parameter, and smoothly transitions the current noise reduction parameter to the target noise reduction parameter within a preset transition time. Preferably, the noise reduction parameter adjustment module gradually adjusts the feedforward noise reduction weight, feedback noise reduction weight, and noise reduction depth using linear interpolation or exponential smoothing, thereby applying the target noise reduction parameter to the noise reduction filter of the Bluetooth headset. The preset transition time is a pre-set parameter switching duration.

[0159] Specifically, a preset transition time is set, and the current actual feedback noise reduction weight is obtained. The noise reduction parameter adjustment module gradually increases the feedback noise reduction weight in steps within the preset transition time. For example, assuming the current feedback noise reduction weight is 0.5 and the target feedback noise reduction weight is 0.8, the preset transition time is 500ms, and it updates every 1ms with a step size of (0.8-0.5) / 500=0.0006. Simultaneously, the low-frequency compensation curve also uses frequency domain interpolation to gradually transition from the current compensation curve to the target low-frequency compensation curve within the preset transition time. After the transition is complete, the final target noise reduction parameters are locked and continuously applied to the various filter coefficients of the noise reduction filter until the next change in the sealing state level triggers a new parameter adjustment process.

[0160] In step S600 of the method of this embodiment of the invention, the correspondence between the sealing state level and the target noise reduction parameter in the preset noise reduction parameter mapping table includes:

[0161] When the current sealing status level is good sealing, the target feedback noise reduction weight in the target noise reduction parameters is greater than the target feedforward noise reduction weight, the target noise reduction depth is the first noise reduction depth, and the target low-frequency compensation curve is the first low-frequency compensation curve.

[0162] When the current sealing status level is slight leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the second noise reduction depth, and the target low-frequency compensation curve is the second low-frequency compensation curve.

[0163] When the current sealing status level is obvious leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the third noise reduction depth, and the target low-frequency compensation curve is the third low-frequency compensation curve.

[0164] Wherein, the first noise reduction depth is greater than the second noise reduction depth, the second noise reduction depth is greater than the third noise reduction depth, and the compensation amplitude of the first low-frequency compensation curve, the second low-frequency compensation curve and the third low-frequency compensation curve for the low-frequency signal decreases sequentially.

[0165] In this embodiment of the invention, when the current sealing status level is "good sealing," the sealing status classification agent transmits this sealing status level to the noise reduction parameter adjustment module. The noise reduction parameter adjustment module queries a preset noise reduction parameter mapping table based on the sealing status level to obtain the corresponding target noise reduction parameter.

[0166] Specifically, under good sealing conditions, a relatively complete acoustic seal is formed between the Bluetooth headset and the ear canal. The feedback microphone can accurately collect residual noise signals within the ear canal, and the feedback noise reduction path exhibits high stability and noise reduction potential. Therefore, the target feedback noise reduction weight in the target noise reduction parameters is greater than the target feedforward noise reduction weight, the target noise reduction depth is the first noise reduction depth, and the target low-frequency compensation curve is the first low-frequency compensation curve. Through the above configuration, the noise reduction parameter adjustment module adjusts the current noise reduction parameters to the target noise reduction parameters adapted to the good sealing condition, enabling the Bluetooth headset to output maximum noise reduction performance under optimal sealing conditions.

[0167] For example, assuming the noise reduction parameter adjustment module obtains that the current sealing status level is good, it queries the preset noise reduction parameter mapping table to obtain the target feedforward noise reduction weight as 0.3, the target feedback noise reduction weight as 0.8, the target noise reduction depth as the first noise reduction depth as -35dB, and the target low-frequency compensation curve as the first low-frequency compensation curve at 50Hz compensation gain as +12dB.

[0168] When the current sealing condition level is "slight leakage," it indicates a certain degree of defect in the acoustic seal between the Bluetooth headset and the ear canal. For example, a tiny gap may appear between the earphone's silicone tip and the ear canal wall, allowing some external noise to enter the ear canal through the leakage path. In this "slight leakage" state, the signal collected by the feedback microphone contains noise components introduced by the leakage path, reducing the stability of the feedback noise reduction path. Continuing to use a higher feedback noise reduction weight may cause feedback or deterioration of the noise reduction effect. Therefore, the noise reduction parameter adjustment module, based on the current sealing condition level of "slight leakage," matches the corresponding target noise reduction parameter from the preset noise reduction parameter mapping table.

[0169] Specifically, in the target noise reduction parameters, the target feedforward noise reduction weight is greater than the target feedback noise reduction weight, the target noise reduction depth is the second noise reduction depth, and the target low-frequency compensation curve is the second low-frequency compensation curve. With this configuration, the noise reduction parameter adjustment module prioritizes the stability of the noise reduction system under slight leakage conditions, while maintaining acceptable noise reduction performance.

[0170] When the current sealing level is "significant leakage," it indicates a serious defect in the acoustic seal between the Bluetooth headset and the ear canal. This could be due to incorrect headset wearing, a severely mismatched silicone ear tip size, or significant movement by the wearer causing headset displacement. In this "significant leakage" state, the feedback microphone signal is severely interfered with, and the feedback noise reduction path can barely contribute to noise reduction; it may even produce howling due to positive feedback. Therefore, the noise reduction parameter adjustment module, based on the current "significant leakage" sealing level, matches the corresponding target noise reduction parameter from a preset noise reduction parameter mapping table.

[0171] Specifically, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, and the feedback noise reduction weight is reduced to minimize the negative impact of the feedback noise reduction path on system stability; the target noise reduction depth is the third noise reduction depth, which is less than the second noise reduction depth, indicating that the noise reduction filter can only achieve limited noise reduction effect under obvious leakage conditions; the target low-frequency compensation curve is the third low-frequency compensation curve, and the compensation amplitude of the third low-frequency compensation curve for low-frequency signals is less than that of the second low-frequency compensation curve.

[0172] The first noise reduction depth is greater than the second noise reduction depth, and the second noise reduction depth is greater than the third noise reduction depth. The noise reduction depth is measured by the absolute value of the noise reduction amplitude. The first noise reduction depth is -35dB, the second noise reduction depth is -25dB, and the third noise reduction depth is -12dB. That is, the better the sealing condition, the greater the noise reduction depth, and the worse the sealing condition, the smaller the noise reduction depth, so as to avoid system instability caused by excessive noise reduction under poor sealing conditions.

[0173] Similarly, the compensation amplitudes of the first, second, and third low-frequency compensation curves for low-frequency signals decrease sequentially. That is, the first low-frequency compensation curve has the highest low-frequency compensation amplitude, the second low-frequency compensation curve has a medium low-frequency compensation amplitude, and the third low-frequency compensation curve has the lowest low-frequency compensation amplitude. This ensures that the degree of low-frequency compensation matches the sealing condition and avoids over-compensation of low-frequency noise under leakage conditions, which could lead to audible noise.

[0174] In this embodiment of the invention, the current sealing status level is matched with the target feedforward noise reduction weight, target feedback noise reduction weight, target noise reduction depth and target low-frequency compensation curve in the preset noise reduction parameter mapping table, and smoothly transitions to the noise reduction filter within a preset transition time; and the feedback or feedforward weight is set to take priority for good sealing, slight leakage and obvious leakage respectively, and the noise reduction depth and low-frequency compensation amplitude decrease sequentially as the leakage intensifies, so as to ensure a smooth transition of noise reduction performance and listening experience while adapting to changes in sealing.

[0175] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects:

[0176] In this embodiment, firstly, the low-frequency energy and multi-band energy distribution of the feedback microphone are extracted by low-pass filtering, the low-frequency sound pressure change inside the earmuff is quantified, and the energy fluctuation amplitude and frequency band distribution are combined to provide feature input for the sealed state classification agent, thereby improving the response accuracy of the adaptive adjustment of noise reduction parameters.

[0177] Secondly, the feedforward microphone signal and the feedback microphone signal are low-pass filtered respectively, and the low-frequency coherence value is calculated frame by frame and then averaged in the time domain. This quantifies the degree of coherence of external noise coupled into the ear through the leakage path, enhances the ability of the sealing state classification agent to distinguish between slight leakage and obvious leakage, and improves the accuracy and reliability of the adaptive adjustment of noise reduction parameters.

[0178] Furthermore, by acquiring acceleration signals through an accelerometer and angular velocity signals through a gyroscope, the acceleration amplitude and angular velocity amplitude are calculated and combined with a preset motion detection time window and a preset motion threshold to determine the instantaneous motion state. When motion disturbances occur, the system shields the sealing state from misjudgment, avoids invalid wake-up of the noise reduction parameter adaptive adjustment component due to head shaking or wearing movements, and improves anti-interference capability and parameter adjustment stability.

[0179] Furthermore, the deviation of the sealing state and the deviation of coherence are calculated. Combined with continuous monitoring of multiple time windows and preset confirmation timing, transient noise and random fluctuations are effectively filtered out. At the same time, motion signals are used to exclude instantaneous action states, avoiding false wake-up of the noise reduction parameter adaptive adjustment component due to non-continuous events. Under the premise of ensuring reliable detection of sealing changes, the misadjustment rate and power consumption are reduced, and the stability of adaptive adjustment is improved.

[0180] Furthermore, a pre-trained lightweight multi-classification model is used, with sealing state features and low-frequency coherence features as inputs. After supervised training, the model accurately outputs the probability of each sealing state level and selects the one with the highest probability as the current sealing state level, providing a reliable sealing state discrimination result for the noise reduction parameter adjustment module.

[0181] Ultimately, the current sealing status level is matched with the target feedforward noise reduction weight, target feedback noise reduction weight, target noise reduction depth, and target low-frequency compensation curve in the preset noise reduction parameter mapping table, and smoothly transitions to the noise reduction filter within a preset transition time; and feedback or feedforward weights are set to prioritize good sealing, slight leakage, and obvious leakage respectively, and the noise reduction depth and low-frequency compensation amplitude decrease sequentially as leakage intensifies, so as to ensure a smooth transition in noise reduction performance and listening experience while adapting to changes in sealing.

[0182] Example 2, as Figure 3 As shown, based on the same inventive concept as the Bluetooth headset noise reduction parameter adaptive adjustment method provided in Embodiment 1, this embodiment of the invention also provides a Bluetooth headset noise reduction parameter adaptive adjustment system, the system comprising:

[0183] The sealing state feature extraction module 11 is used to acquire the feedback microphone signal and the feedforward microphone signal of the Bluetooth headset, and extract the sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone.

[0184] The low-frequency coherence feature extraction module 12 is used to extract low-frequency coherence features based on the feedforward microphone signal and the feedback microphone signal.

[0185] The instantaneous motion state recognition module 13 is used to acquire motion signals through a motion sensor installed in the Bluetooth headset, and to identify whether the wearer is in an instantaneous motion state based on the motion signals.

[0186] The adjustment component wake-up module 14 is used to wake up the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset when the sealing state feature and the low frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state.

[0187] The sealing status level output module 15 is used by the sealing status classification agent in the noise reduction parameter adaptive adjustment component to output the current sealing status level according to the sealing status characteristics and the low-frequency coherence characteristics.

[0188] The noise reduction parameter adjustment module 16 is used by the noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component to match the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjust the noise reduction parameter according to the target noise reduction parameter.

[0189] In one embodiment, the sealing state feature extraction module 11 is used for:

[0190] Feedback microphone signals are obtained through a feedback microphone located inside the earcups of the Bluetooth headset, and feedforward microphone signals are obtained through a feedforward microphone located outside the earcups of the Bluetooth headset.

[0191] The feedback microphone signal is low-pass filtered to obtain the low-frequency feedback microphone signal;

[0192] Based on the low-frequency signal from the feedback microphone, extract the low-frequency energy of the feedback microphone and the frequency band energy distribution in multiple frequency bands;

[0193] The low-frequency energy of the feedback microphone and the frequency band energy distribution in the multiple frequency bands are used as the characteristics of the sealed state.

[0194] Specifically, based on the low-frequency signal from the feedback microphone, the low-frequency energy of the feedback microphone and the frequency band energy distribution across multiple frequency bands are extracted, including:

[0195] The low-frequency signal of the feedback microphone is divided into multiple short frames, and the energy of the low-frequency signal of the feedback microphone in each short frame is calculated to obtain the low-frequency energy sequence of the feedback microphone.

[0196] Based on the low-frequency energy sequence of the feedback microphone, calculate the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude between adjacent short frames, and use the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude as the low-frequency energy of the feedback microphone.

[0197] Spectral analysis is performed on the low-frequency signal of the feedback microphone to obtain the frequency band energy distribution of the low-frequency signal of the feedback microphone in multiple frequency bands, and the frequency band energy distribution in the multiple frequency bands is taken as the frequency band energy distribution.

[0198] In one embodiment, the low-frequency coherence feature extraction module 12 is used for:

[0199] The feedforward microphone signal and the feedback microphone signal are respectively low-pass filtered to obtain the feedforward microphone low-frequency signal and the feedback microphone low-frequency signal;

[0200] The low-frequency signals from the feedforward microphone and the low-frequency signals from the feedback microphone are aligned in time and divided into multiple short time frames;

[0201] For each short frame, a low-frequency coherence value is calculated based on the low-frequency signal from the feedforward microphone and the low-frequency signal from the feedback microphone;

[0202] The low-frequency coherence characteristics are obtained by averaging the low-frequency coherence values ​​of multiple short frames over time.

[0203] In one embodiment, the instantaneous action state recognition module 13 is used for:

[0204] Acceleration signals are obtained by an accelerometer installed in the Bluetooth headset, and angular velocity signals are obtained by a gyroscope installed in the Bluetooth headset.

[0205] The acceleration amplitude is calculated based on the acceleration signal, and the angular velocity amplitude is calculated based on the angular velocity signal.

[0206] When the acceleration amplitude or the angular velocity amplitude exceeds a preset action threshold within a preset action detection time window, it is determined that the wearer is in a momentary action state.

[0207] When the acceleration amplitude and the angular velocity amplitude do not exceed the preset action threshold within the preset action detection time window, it is determined that the wearer is not in a state of instantaneous action.

[0208] In one embodiment, the regulating component wake-up module 14 is used to:

[0209] The deviation of the sealing state is determined based on the difference between each feature component in the sealing state characteristics and the corresponding feature component in the preset sealing state baseline, wherein the preset sealing state baseline is determined based on the historical sealing state characteristics of the Bluetooth headset in the normal sealing state.

[0210] The coherence deviation is determined based on the difference between the low-frequency coherence characteristics and the preset coherence baseline, wherein the preset coherence baseline is determined based on the historical low-frequency coherence characteristics of the Bluetooth headset under normal sealed conditions.

[0211] When the deviation of the sealing state exceeds the preset sealing deviation threshold for multiple consecutive time windows, and the deviation of the coherence exceeds the preset coherence deviation threshold for multiple consecutive time windows, and the motion signal does not indicate that the wearer is in an instantaneous action state, it is determined that the preset sealing change condition is met.

[0212] After the preset sealing change conditions are met, a preset confirmation timer is started, and the sealing state characteristics and the low-frequency coherence characteristics are monitored within the preset confirmation timer.

[0213] When the sealing state characteristics and the low-frequency coherence characteristics continuously meet the preset sealing change conditions within the preset confirmation time, the noise reduction parameter adaptive adjustment component is activated.

[0214] When the sealing state feature or the low-frequency coherence feature no longer meets the preset sealing change condition within the preset confirmation time, the preset confirmation time is reset and the noise reduction parameter adaptive adjustment component is not activated.

[0215] In one embodiment, the sealing status level output module 15 is used for:

[0216] The sealing state features and the low-frequency coherence features are input into a pre-trained sealing state classification agent, which outputs the probability of each sealing state level based on the sealing state features and the low-frequency coherence features.

[0217] The sealing condition level with the highest probability is taken as the current sealing condition level, which includes good sealing, slight leakage, and significant leakage.

[0218] The pre-training process of the sealed state classification agent includes:

[0219] Multiple historical samples are acquired, each of which includes historical sealing state features, historical low-frequency coherence features, and corresponding sealing state labels;

[0220] A sealing state classification agent is constructed based on a lightweight multi-classification model. The sealing state classification agent includes an input layer, a hidden layer, and an output layer. The input layer receives sealing state features and low-frequency coherence features, and the output layer outputs the probability of each sealing state level.

[0221] Using the historical sealing state features and the historical low-frequency coherence features as input features, and the sealing state label as the supervision label, the sealing state classification agent is trained in a supervised manner until the verification convergence, thus obtaining the pre-trained sealing state classification agent.

[0222] In one embodiment, the noise reduction parameter adjustment module 16 is used for:

[0223] The noise reduction parameter adjustment module obtains the current sealing status level;

[0224] The noise reduction parameter adjustment module matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level. The preset noise reduction parameter mapping table stores the correspondence between the sealing status level and the target noise reduction parameter. The target noise reduction parameter includes the target feedforward noise reduction weight, the target feedback noise reduction weight, the target noise reduction depth, and the target low-frequency compensation curve.

[0225] The noise reduction parameter adjustment module compares the target noise reduction parameter with the current noise reduction parameter, smoothly transitions the current noise reduction parameter to the target noise reduction parameter within a preset transition time, and applies the target noise reduction parameter to the noise reduction filter of the Bluetooth headset.

[0226] The correspondence between the sealing status level and the target noise reduction parameter in the preset noise reduction parameter mapping table includes:

[0227] When the current sealing status level is good sealing, the target feedback noise reduction weight in the target noise reduction parameters is greater than the target feedforward noise reduction weight, the target noise reduction depth is the first noise reduction depth, and the target low-frequency compensation curve is the first low-frequency compensation curve.

[0228] When the current sealing status level is slight leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the second noise reduction depth, and the target low-frequency compensation curve is the second low-frequency compensation curve.

[0229] When the current sealing status level is obvious leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the third noise reduction depth, and the target low-frequency compensation curve is the third low-frequency compensation curve.

[0230] Wherein, the first noise reduction depth is greater than the second noise reduction depth, the second noise reduction depth is greater than the third noise reduction depth, and the compensation amplitude of the first low-frequency compensation curve, the second low-frequency compensation curve and the third low-frequency compensation curve for the low-frequency signal decreases sequentially.

[0231] Compared to existing technologies, this invention first acquires the feedback microphone signal and feedforward microphone signal from the Bluetooth headset, extracts sealing state features including at least the low-frequency energy of the feedback microphone, and extracts low-frequency coherence features. Secondly, by setting a motion sensor within the Bluetooth headset to acquire motion signals and identify whether it is in a transient motion state, it can shield against misjudgments of the sealing state when motion disturbances occur. Thirdly, when the sealing state features and low-frequency coherence features meet preset sealing change conditions within multiple consecutive time windows, and the device is in a transient motion state, the noise reduction parameter adaptive adjustment component is activated to ensure the reliability of sealing change detection. Furthermore, the sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state features and low-frequency coherence features, providing a reliable basis for noise reduction parameters. Finally, based on the current sealing state level, the target noise reduction parameter is matched from a preset noise reduction parameter mapping table, and the current noise reduction parameter is smoothly transitioned to the target noise reduction parameter before being applied to the noise reduction filter of the Bluetooth headset, ensuring a smooth transition in noise reduction performance and listening experience while adapting to sealing changes.

Claims

1. A method for adaptively adjusting noise reduction parameters of Bluetooth headphones, characterized in that, The method includes: Acquire the feedback microphone signal and feedforward microphone signal of the Bluetooth headset, and extract the sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone; Based on the feedforward microphone signal and the feedback microphone signal, low-frequency coherence features are extracted; Motion signals are acquired by a motion sensor installed in the Bluetooth headset, and the wearer is identified as being in a state of instantaneous motion based on the motion signals. When the sealing state feature and the low-frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state, the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset is activated. The sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state characteristics and the low-frequency coherence characteristics. The noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjusts the noise reduction parameter according to the target noise reduction parameter.

2. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, Acquire the feedback microphone signal and feedforward microphone signal from the Bluetooth headset, and extract the sealing state features based on the feedback microphone signal, including: Feedback microphone signals are obtained through a feedback microphone located inside the earcups of the Bluetooth headset, and feedforward microphone signals are obtained through a feedforward microphone located outside the earcups of the Bluetooth headset. The feedback microphone signal is low-pass filtered to obtain the low-frequency feedback microphone signal; Based on the low-frequency signal from the feedback microphone, extract the low-frequency energy of the feedback microphone and the frequency band energy distribution in multiple frequency bands; The low-frequency energy of the feedback microphone and the frequency band energy distribution in the multiple frequency bands are used as the characteristics of the sealed state.

3. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 2, characterized in that, Based on the low-frequency signal from the feedback microphone, extract the low-frequency energy of the feedback microphone and the frequency band energy distribution across multiple frequency bands, including: The low-frequency signal of the feedback microphone is divided into multiple short frames, and the energy of the low-frequency signal of the feedback microphone in each short frame is calculated to obtain the low-frequency energy sequence of the feedback microphone. Based on the low-frequency energy sequence of the feedback microphone, calculate the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude between adjacent short frames, and use the time average value of the low-frequency energy of the feedback microphone and the energy fluctuation amplitude as the low-frequency energy of the feedback microphone. Spectral analysis is performed on the low-frequency signal of the feedback microphone to obtain the frequency band energy distribution of the low-frequency signal of the feedback microphone in multiple frequency bands, and the frequency band energy distribution in the multiple frequency bands is taken as the frequency band energy distribution.

4. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, Based on the feedforward microphone signal and the feedback microphone signal, low-frequency coherence features are extracted, including: The feedforward microphone signal and the feedback microphone signal are respectively low-pass filtered to obtain the feedforward microphone low-frequency signal and the feedback microphone low-frequency signal; The low-frequency signals from the feedforward microphone and the low-frequency signals from the feedback microphone are aligned in time and divided into multiple short time frames; For each short frame, a low-frequency coherence value is calculated based on the low-frequency signal from the feedforward microphone and the low-frequency signal from the feedback microphone; The low-frequency coherence characteristics are obtained by averaging the low-frequency coherence values ​​of multiple short frames over time.

5. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, The system acquires motion signals using a motion sensor embedded in the Bluetooth headset and identifies whether the wearer is in a state of momentary motion based on the motion signals, including: Acceleration signals are obtained by an accelerometer installed in the Bluetooth headset, and angular velocity signals are obtained by a gyroscope installed in the Bluetooth headset. The acceleration amplitude is calculated based on the acceleration signal, and the angular velocity amplitude is calculated based on the angular velocity signal. When the acceleration amplitude or the angular velocity amplitude exceeds a preset action threshold within a preset action detection time window, it is determined that the wearer is in a momentary action state. When the acceleration amplitude and the angular velocity amplitude do not exceed the preset action threshold within the preset action detection time window, it is determined that the wearer is not in a state of instantaneous action.

6. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, When the sealing state characteristic and the low-frequency coherence characteristic meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state, the noise cancellation parameter adaptive adjustment component deployed in the Bluetooth headset is activated, including: The deviation of the sealing state is determined based on the difference between each feature component in the sealing state characteristics and the corresponding feature component in the preset sealing state baseline, wherein the preset sealing state baseline is determined based on the historical sealing state characteristics of the Bluetooth headset in the normal sealing state. The coherence deviation is determined based on the difference between the low-frequency coherence characteristics and the preset coherence baseline, wherein the preset coherence baseline is determined based on the historical low-frequency coherence characteristics of the Bluetooth headset under normal sealed conditions. When the deviation of the sealing state exceeds the preset sealing deviation threshold for multiple consecutive time windows, and the deviation of the coherence exceeds the preset coherence deviation threshold for multiple consecutive time windows, and the motion signal does not indicate that the wearer is in an instantaneous action state, it is determined that the preset sealing change condition is met. After the preset sealing change conditions are met, a preset confirmation timer is started, and the sealing state characteristics and the low-frequency coherence characteristics are monitored within the preset confirmation timer. When the sealing state characteristics and the low-frequency coherence characteristics continuously meet the preset sealing change conditions within the preset confirmation time, the noise reduction parameter adaptive adjustment component is activated. When the sealing state feature or the low-frequency coherence feature no longer meets the preset sealing change condition within the preset confirmation time, the preset confirmation time is reset and the noise reduction parameter adaptive adjustment component is not activated.

7. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, The sealing state classification agent in the noise reduction parameter adaptive adjustment component outputs the current sealing state level based on the sealing state characteristics and the low-frequency coherence characteristics, including: The sealing state features and the low-frequency coherence features are input into a pre-trained sealing state classification agent, which outputs the probability of each sealing state level based on the sealing state features and the low-frequency coherence features. The sealing condition level with the highest probability is taken as the current sealing condition level, which includes good sealing, slight leakage, and significant leakage. The pre-training process of the sealed state classification agent includes: Multiple historical samples are acquired, each of which includes historical sealing state features, historical low-frequency coherence features, and corresponding sealing state labels; A sealing state classification agent is constructed based on a lightweight multi-classification model. The sealing state classification agent includes an input layer, a hidden layer, and an output layer. The input layer receives sealing state features and low-frequency coherence features, and the output layer outputs the probability of each sealing state level. Using the historical sealing state features and the historical low-frequency coherence features as input features, and the sealing state label as the supervision label, the sealing state classification agent is trained in a supervised manner until the verification convergence, thus obtaining the pre-trained sealing state classification agent.

8. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 1, characterized in that, The noise reduction parameter adjustment module in the adaptive noise reduction parameter adjustment component matches a target noise reduction parameter from a preset noise reduction parameter mapping table according to the current sealing status level, and adjusts the noise reduction parameter according to the target noise reduction parameter, including: The noise reduction parameter adjustment module obtains the current sealing status level; The noise reduction parameter adjustment module matches the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level. The preset noise reduction parameter mapping table stores the correspondence between the sealing status level and the target noise reduction parameter. The target noise reduction parameter includes the target feedforward noise reduction weight, the target feedback noise reduction weight, the target noise reduction depth, and the target low-frequency compensation curve. The noise reduction parameter adjustment module compares the target noise reduction parameter with the current noise reduction parameter, smoothly transitions the current noise reduction parameter to the target noise reduction parameter within a preset transition time, and applies the target noise reduction parameter to the noise reduction filter of the Bluetooth headset.

9. The method for adaptive adjustment of noise reduction parameters in Bluetooth headphones according to claim 8, characterized in that, The correspondence between the sealing status level and the target noise reduction parameter in the preset noise reduction parameter mapping table includes: When the current sealing status level is good sealing, the target feedback noise reduction weight in the target noise reduction parameters is greater than the target feedforward noise reduction weight, the target noise reduction depth is the first noise reduction depth, and the target low-frequency compensation curve is the first low-frequency compensation curve. When the current sealing status level is slight leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the second noise reduction depth, and the target low-frequency compensation curve is the second low-frequency compensation curve. When the current sealing status level is obvious leakage, the target feedforward noise reduction weight in the target noise reduction parameters is greater than the target feedback noise reduction weight, the target noise reduction depth is the third noise reduction depth, and the target low-frequency compensation curve is the third low-frequency compensation curve. Wherein, the first noise reduction depth is greater than the second noise reduction depth, the second noise reduction depth is greater than the third noise reduction depth, and the compensation amplitude of the first low-frequency compensation curve, the second low-frequency compensation curve and the third low-frequency compensation curve for the low-frequency signal decreases sequentially.

10. A noise reduction parameter adaptive adjustment system for Bluetooth headphones, characterized in that, A method for adaptively adjusting noise reduction parameters of a Bluetooth headset according to any one of claims 1 to 9, the system comprising: A sealing state feature extraction module is used to acquire the feedback microphone signal and the feedforward microphone signal of the Bluetooth headset, and extract sealing state features based on the feedback microphone signal, wherein the sealing state features include at least the low-frequency energy of the feedback microphone; A low-frequency coherence feature extraction module is used to extract low-frequency coherence features based on the feedforward microphone signal and the feedback microphone signal; The instantaneous motion state recognition module is used to acquire motion signals through a motion sensor installed in the Bluetooth headset, and to identify whether the wearer is in an instantaneous motion state based on the motion signals. The adjustment component wake-up module is used to wake up the noise reduction parameter adaptive adjustment component deployed in the Bluetooth headset when the sealing state feature and the low frequency coherence feature meet the preset sealing change conditions within multiple consecutive time windows, and the motion signal does not indicate that the wearer is in a momentary action state. A sealing status level output module is used by the sealing status classification agent in the noise reduction parameter adaptive adjustment component to output the current sealing status level based on the sealing status characteristics and the low-frequency coherence characteristics. The noise reduction parameter adjustment module is used by the noise reduction parameter adjustment module in the noise reduction parameter adaptive adjustment component to match the target noise reduction parameter from the preset noise reduction parameter mapping table according to the current sealing status level, and adjust the noise reduction parameter according to the target noise reduction parameter.