A method and system for individualized noise reduction of a bluetooth headset

Abstract

The application is suitable for the technical field of audio processing, and provides a method and system for individualized noise reduction of a Bluetooth earphone, the method comprising: acquiring an original time-domain noise signal outside the earphone collected by an external microphone and an error signal inside the earphone collected by an internal microphone, performing frame preprocessing on the original time-domain noise signal to obtain continuous signal frames; decomposing each signal frame into noise components with different time-frequency characteristics and time-domain signals thereof; generating respective corresponding anti-noise signals based on a plurality of noise components and the error signal; calculating a mixing weight corresponding to each noise component according to the error signal and the time-domain signal, and weighting and combining the respective corresponding anti-noise signals of the plurality of noise components into a reverse sound wave signal. Through fine processing of the external and internal noise signals, the effectiveness and adaptability of the noise reduction technology can be significantly improved.

Description

Technical Field

[0001] This invention belongs to the technical field of audio processing, and particularly relates to a method and system for personalized noise reduction of Bluetooth headphones. Background Technology

[0002] With the rapid development of wireless technology, Bluetooth headphones have become widely popular due to their convenience and high sound quality. However, in daily use, environmental noise interference remains one of the main factors affecting the user's listening experience. Especially in noisy environments, traditional noise cancellation technologies struggle to effectively eliminate various types of noise, causing discomfort for users when using Bluetooth headphones. Therefore, developing a personalized noise cancellation solution has become an urgent technical problem to be solved.

[0003] Existing noise reduction technologies are mainly divided into two types: active noise control (ANC) and passive noise isolation. Active noise control technology uses microphones to collect ambient noise and generates sound wave signals that are opposite to the ambient noise to cancel it out. However, existing technologies have limitations when dealing with complex noisy environments, especially in their ability to identify and process transient impulse noise and steady-state background noise. In addition, existing active noise cancellation solutions often use a uniform processing strategy, which is difficult to adapt to different users' listening habits and environmental changes, resulting in unsatisfactory noise reduction effects. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method and system for individualized noise reduction in Bluetooth headsets, in order to solve the technical problem that existing active noise reduction solutions often adopt a uniform processing strategy, which is difficult to adapt to different users' listening habits and environmental changes, resulting in unsatisfactory noise reduction effects.

[0005] A first aspect of this invention provides a method for individualized noise reduction in a Bluetooth headset, the method comprising:

[0006] Step 101: Acquire the original temporal noise signal from the outside of the earphone collected by the external microphone and the error signal from the inside of the earphone collected by the internal microphone. Perform frame-by-frame preprocessing on the original temporal noise signal to obtain continuous signal frames. The error signal is the residual noise that actually enters the ear after noise reduction.

[0007] Step 102: Decompose each signal frame into noise components and their time-domain signals with different time-frequency characteristics; wherein, the noise components include steady-state floor noise components and transient impulse noise components;

[0008] Step 103: Generate corresponding anti-noise signals based on the multiple noise components and the error signal;

[0009] Step 104: Calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic signal.

[0010] Further, the step of decomposing each signal frame into noise components with different time-frequency characteristics and their time-domain signals includes:

[0011] Multi-layer wavelet scattering transform is performed on the signal frame to generate a high-dimensional scattering feature vector that includes information on energy distribution and envelope changes at different time scales;

[0012] The high-dimensional scattering feature vector is reduced in dimensionality by a preset feature encoder to obtain a low-dimensional context feature vector.

[0013] Obtain the preset steady-state basis linear projection matrix and transient impact linear projection matrix;

[0014] The context feature vector is input into the steady-state basis linear projection matrix to obtain the time-frequency mask of the steady-state basis noise component;

[0015] The context feature vector is input into the transient impact linear projection matrix to obtain the time-frequency mask of the transient impact noise component;

[0016] The time-frequency representation of the signal frame is weighted and separated using the time-frequency mask to reconstruct the time-domain signal corresponding to each noise component.

[0017] Furthermore, the step of performing multi-layer wavelet scattering transform on the signal frame to generate a high-dimensional scattering feature vector including energy distribution and envelope variation information at different time scales includes:

[0018] The signal frame is convolved and moduloed using the first set of complex wavelet filters to obtain the first layer of modulo signal.

[0019] The first layer modulus signal is low-pass filtered and downsampled to obtain the first layer scattering coefficients that characterize the steady-state spectrum of the signal.

[0020] The first layer modulus signal is convolved and moduloed using the second set of complex wavelet filters to obtain the second layer modulus signal;

[0021] The second-layer modulus signal is low-pass filtered and downsampled to obtain the second-layer scattering coefficients that characterize the signal energy envelope variation mode.

[0022] The scattering coefficients of the first layer and the scattering coefficients of the second layer are concatenated to form the high-dimensional scattering feature vector.

[0023] Further, the step of using the time-frequency mask to perform weighted separation of the time-frequency representation of the signal frame to reconstruct the time-domain signal corresponding to each noise component includes:

[0024] Calculate the short-time Fourier transform of the signal frame to obtain a complex time-frequency representation;

[0025] The complex time-frequency representation is multiplied point by point with the time-frequency mask corresponding to each noise component to obtain the separated time-frequency representation corresponding to each noise component;

[0026] The inverse short-time Fourier transform is performed on each of the separated time-frequency representations, and the continuous time-domain signal is synthesized by the overlapping addition method to obtain the time-domain signals corresponding to the steady-state basis noise component and the transient impulse noise component.

[0027] Furthermore, the step of generating corresponding anti-noise signals based on multiple noise components and the error signal includes:

[0028] Calculate the time-frequency representation of the error signal;

[0029] Calculate a first correlation measure between the time-frequency representation of the error signal and the time-frequency mask corresponding to the steady-state basis noise component;

[0030] Based on the first correlation metric, adjust the adaptive step size of the adaptive filter path;

[0031] The time-frequency mask of the steady-state basis noise component is input into the adaptive filter path for processing to obtain the first anti-noise signal;

[0032] Real-time monitoring of the time-domain energy of transient impact noise components;

[0033] When the temporal energy exceeds a preset energy value, noise segments of historical samples with a first preset duration before the current transient event start point, noise segments of samples that have occurred with a second preset duration since the current moment, and the context feature vector extracted from the time-frequency mask of the transient impact noise component are input into a pre-trained causal convolutional neural network; wherein, the current transient event start point refers to the moment when the temporal energy exceeds the preset energy value;

[0034] The causal convolutional neural network is used to predict a second noise-resistant signal for a third preset duration in the future.

[0035] Further, the step of adjusting the adaptive step size of the adaptive filter path based on the first correlation metric includes:

[0036] Obtain the high correlation threshold, low correlation threshold, maximum allowable step size, and minimum allowable step size;

[0037] When the first correlation metric is greater than or equal to the high correlation threshold, the maximum allowable step size is used as the adaptive step size;

[0038] When the first correlation metric is less than or equal to the low correlation threshold, the minimum allowable step size is used as the adaptive step size;

[0039] When the first correlation metric is between the high correlation threshold and the low correlation threshold, the adaptive step size is... .

[0040] Further, the step of calculating the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and weighting and synthesizing the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic signal includes:

[0041] Calculate the second correlation measure between the error signal and the time-domain signals of each noise component;

[0042] Normalize each of the second correlation measures to obtain the initial weighting factor corresponding to each noise component;

[0043] The initial weighting factors are subjected to time-domain smoothing filtering to obtain the final mixed weights;

[0044] Based on the aforementioned hybrid weights, the anti-noise signals corresponding to each of the multiple noise components are weighted and synthesized into an anti-phase acoustic signal.

[0045] A second aspect of the present invention provides a personalized noise cancellation system for Bluetooth headsets, comprising:

[0046] The acquisition unit is used to acquire the original time-domain noise signal outside the earphone collected by the external microphone and the error signal inside the earphone collected by the internal microphone, and to perform frame-by-frame preprocessing on the original time-domain noise signal to obtain continuous signal frames.

[0047] A decomposition unit is used to decompose each of the signal frames into noise components and their time-domain signals with different time-frequency characteristics; wherein the noise components include steady-state floor noise components and transient impulse noise components;

[0048] The generation unit is configured to generate corresponding anti-noise signals based on the multiple noise components and the error signal;

[0049] The synthesis unit is used to calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and to weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic wave signal; wherein the anti-phase acoustic wave signal is used to drive the loudspeaker for noise reduction processing.

[0050] A third aspect of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the personalized noise reduction method for Bluetooth headsets described in the first aspect.

[0051] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the personalized noise reduction method for Bluetooth headsets described in the first aspect.

[0052] The beneficial effects of this invention compared to existing technologies are as follows: By acquiring the original temporal noise signal collected by an external microphone and the error signal collected by an internal microphone, and performing frame-by-frame preprocessing, the noise signal can be effectively decomposed into steady-state floor noise components and transient impulse noise components. This precise component identification enables subsequent noise reduction processing to be more effective against different types of noise, improving the targeting and effectiveness of noise reduction. By generating corresponding anti-noise signals based on multiple noise components and error signals, the noise reduction strategy can be dynamically adjusted according to the user's specific environment and listening needs. This personalized noise reduction method can provide each user with a personalized noise reduction experience, meeting the noise reduction needs in different scenarios. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A schematic flowchart of a method for individualized noise reduction in a Bluetooth headset provided by the present invention is shown.

[0055] Figure 2 A schematic diagram of a personalized noise reduction system for a Bluetooth headset according to an embodiment of the present invention is shown.

[0056] Figure 3 A schematic diagram of a terminal device provided in an embodiment of the present invention is shown. Detailed Implementation

[0057] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0058] This invention provides a method and system for individualized noise cancellation in Bluetooth headsets, addressing the technical problem that existing active noise cancellation solutions often employ a uniform processing strategy, making it difficult to adapt to different users' listening habits and environmental changes, resulting in unsatisfactory noise cancellation effects.

[0059] First, this invention provides a method for individualized noise reduction in Bluetooth headsets. Please see below. Figure 1 , Figure 1 A schematic flowchart of a method for personalized noise reduction in a Bluetooth headset, provided by the present invention, is shown. Figure 1 As shown, the personalized noise cancellation method for this Bluetooth headset may include the following steps:

[0060] Step 101: Acquire the original time-domain noise signal from the outside of the earphone collected by the external microphone and the error signal from the inside of the earphone collected by the internal microphone. Perform frame-by-frame preprocessing on the original time-domain noise signal to obtain continuous signal frames.

[0061] An external microphone collects the raw time-domain noise signal, which is the actual noise in the user's environment. An internal microphone collects the error signal, which is the residual noise that actually enters the ear after noise reduction. This is the key feedback of the system, used to evaluate the noise reduction effect and guide adjustments. The continuous raw noise signal is segmented into short time frames to facilitate subsequent time-frequency analysis and real-time processing.

[0062] Step 102: Decompose each signal frame into noise components and their time-domain signals with different time-frequency characteristics; wherein, the noise components include steady-state floor noise components and transient impulse noise components;

[0063] Each frame of signal is analyzed and decomposed into sub-components with different time-frequency characteristics. Noise components include, but are not limited to, steady-state background noise and transient impulse noise. Steady-state background noise refers to noise with relatively stable energy and frequency distribution that persists continuously (such as air conditioner noise or engine hum). This type of noise is highly regular and easily suppressed by traditional ANC. Transient impulse noise refers to sudden, brief, high-energy noise (such as knocking sounds, car horns, or sudden human voices). This type of noise changes drastically, and traditional ANC is prone to insufficient suppression or residual noise due to algorithm delays.

[0064] Specifically, step 102 includes steps 1021 to 1026:

[0065] Step 1021: Perform multi-layer wavelet scattering transform on the signal frame to generate a high-dimensional scattering feature vector that includes information on energy distribution and envelope changes at different time scales;

[0066] Wavelet scattering transform is a mathematical tool used in deep learning for signal analysis. It combines the multi-resolution analysis of wavelet transform with the hierarchical structure of deep networks. Through multi-layer operations, it can simultaneously capture short-term burst details (corresponding to transient noise) and long-term stable patterns (corresponding to steady-state noise) in noisy signals. Wavelet scattering transform is stable to small deformations of the signal (such as time shifts), making the extracted features insensitive to subtle changes in noise and focusing more on its essential time-frequency structure, making it very suitable for noise classification and separation.

[0067] High-dimensional scattering eigenvectors are highly condensed mathematical representations of the energy distribution and envelope shape of the original signal across multiple time-frequency scales, containing all the information needed for decomposition.

[0068] Specifically, step 1021 includes steps A1 to A5:

[0069] Step A1: Convolve the signal frame using the first set of complex wavelet filters and perform modulo operation to obtain the first layer of modulo signal;

[0070] The first set of complex wavelet filters is a set of basis functions covering different frequencies and exhibiting good time-frequency locality. Complex wavelets can provide both amplitude and phase information simultaneously. Convolving the signal frame with each wavelet filter essentially calculates the signal's response intensity in different frequency bands. This is equivalent to a high-resolution time-frequency analysis. The modulus of the complex convolution result is then taken, discarding phase information and retaining energy information. This step is crucial; it makes the features insensitive to small time shifts (phase changes) in the signal, thereby enhancing the stability of the features. The output is called the first-layer modulus signal.

[0071] In wavelet scattering networks, complex Morlet wavelets (also known as Gabor wavelets) are used. They are modulation forms of complex sine waves within a Gaussian window, exhibiting excellent time-frequency localization characteristics. The first set of complex wavelet filters (extracting the steady-state basis) covers the main frequency bands of the audio range to capture the steady-state spectral basis of noise. This set of filters requires good frequency resolution to analyze stable harmonic components.

[0072] Example design parameters (for audio signals with a 16kHz sampling rate): Center frequencies between 50Hz and 8000Hz (Nyquist frequency), evenly spaced on a logarithmic scale. For example, select 32 frequency bands. Specific frequency points could be:

[0073] [50,70,98,137,192,269,376,527,738,1033,1446,2024,2833,3965,5550,7760]Hz (and their corresponding mirror high-frequency components, a total of 32).

[0074] Logarithmic scales align with the characteristics of human hearing (Mel scales) and can better capture perceptual information in speech and music.

[0075] The quality factor Q is set to a relatively high value, for example, Q=8. Q = center frequency / bandwidth. A high Q value means a narrow bandwidth, resulting in better frequency selectivity.

[0076] The scaling parameter σ is determined by the center frequency ξ and the Q value. σ = Q / center frequency. For a wavelet with a center frequency of 200 Hz, σ ≈ 8 / (2π * 200) ≈ 6.4 ms. Its time-domain support is approximately 3σ ≈ 19 ms.

[0077] There are 32 filters, which act like a set of narrowband bandpass filters, covering the main frequency range audible to the human ear. When a signal passes through them, each filter outputs energy for its corresponding frequency band. After time averaging (low-pass filtering) the modulus, the average energy of the signal in that frequency band, i.e., the steady-state spectrum, is obtained.

[0078] Step A2: Perform low-pass filtering and downsampling on the first layer modulus signal to obtain the first layer scattering coefficients that characterize the steady-state spectrum substrate of the signal;

[0079] The first-layer modulus signal is low-pass filtered. This is equivalent to smoothing the time-varying energy envelope, filtering out rapid fluctuations and retaining only the slowly changing trend. Since the effective bandwidth of the signal is reduced after low-pass filtering, downsampling can be performed to reduce the amount of data without losing information.

[0080] The first layer of scattering coefficients represents the result of time-localized energy extraction and smoothing of the original signal frame. They mainly characterize the relatively stable and highly periodic spectral components in the signal, that is, the time-frequency energy distribution corresponding to the steady-state floor noise.

[0081] Step A3: Convolve the first layer modulus signal using the second set of complex wavelet filters and perform modulo operation to obtain the second layer modulus signal;

[0082] The second group of complex wavelet filters has a different scale than the first group. This operation is repeated for the first-layer modulus signal. This is equivalent to analyzing the changing patterns of the signal's energy envelope. It can capture how energy fluctuates over time—for example, a sudden shock would produce a sharp envelope spike.

[0083] The second set of complex wavelet filters (capturing transient interactions) analyzes the envelope changes of the first-layer output U_1. This set of filters requires higher time resolution to capture rapid fluctuations in the envelope (such as shocks and transients).

[0084] Design parameter example: The center frequency (ξ / 2π) covers a low frequency range, such as 0.5Hz to 200Hz. This range corresponds to the frequency of envelope variation (i.e., the frequency of amplitude modulation). It is also distributed on a logarithmic scale, for example, choosing 16 frequency bands.

[0085] Specific frequency points can be: [0.5, 1.1, 2.4, 5.3, 11.7, 25.7, 56.5, 124, 200] Hz, etc.

[0086] The quality factor Q is set to a relatively low constant, such as Q=1. A low Q value means a wide bandwidth, resulting in better temporal resolution to locate rapid changes in the envelope.

[0087] The scaling parameter σ is given by σ = Q / ξ. For a wavelet with a center frequency of 10 Hz, σ ≈ 1 / (2π*10) ≈ 16 ms. Its time domain support is approximately 3σ ≈ 48 ms. This is wider in time than the first-level wavelet (approximately 19 ms when analyzing 200 Hz), but because it analyzes the envelope (which changes more slowly), the relative resolution is sufficient.

[0088] Sixteen filters are used to analyze the rhythm or fluctuation pattern of the signal energy envelope. A transient impact (such as a knocking sound) will produce a synchronous, rapid rise and fall across the first-layer envelope U_1 of all frequency bands. The second-layer wavelet will strongly respond to this broadband, synchronous envelope transient. Conversely, the envelope of a stationary noise or tone signal changes slowly, and the second-layer response is weaker.

[0089] Step A4: Perform low-pass filtering and downsampling on the second-layer modulus signal to obtain the second-layer scattering coefficients that characterize the signal energy envelope variation mode;

[0090] Similarly, the second-layer modulus signal is smoothed and downsampled to extract its slowly changing trend.

[0091] The second-layer scattering coefficient characterizes the morphological changes of the first-layer energy envelope itself. It excels at capturing transient, non-stationary events, such as the onset and cessation of impact noise and sudden energy changes. This directly corresponds to the time-frequency characteristics of transient impact noise (especially the variation pattern on the time axis).

[0092] Step A5: Concatenate the first layer scattering coefficients with the second layer scattering coefficients to form the high-dimensional scattering feature vector.

[0093] The scattering coefficients of the first layer (steady-state spectral basis) and the scattering coefficients of the second layer (energy envelope variation) are concatenated along the vector dimension to form a high-dimensional scattering feature vector. Because it comes from two layers and multiple filters, the dimension of this vector is much higher than that of the original signal frame, containing rich, structured information. This vector simultaneously encodes both the steady-state and transient components of the signal. This provides all the information needed for subsequent feature encoders and projection matrices to distinguish between steady-state and transient noise.

[0094] In the embodiments corresponding to steps A1 to A5, multi-level wavelet scattering transform and complex wavelet filters are used to extract features from the signal at multiple levels, aiming to generate a high-dimensional scattering feature vector. This vector not only contains the energy distribution information of the signal at different time scales, but also reflects the dynamic changes of the signal, providing crucial basic data for subsequent noise component separation and processing.

[0095] Step 1022: The high-dimensional scattering feature vector is reduced in dimensionality using a preset feature encoder to obtain a low-dimensional context feature vector;

[0096] The preset feature encoder is a trained neural network model (such as an autoencoder, PCA transform, or other supervised learning model).

[0097] High-dimensional feature vectors are compressed into low-dimensional context feature vectors. This removes redundant information, retains the most essential and discriminative features, and improves the efficiency and robustness of subsequent processing. This low-dimensional vector can be understood as a "feature fingerprint" or context code of the current noisy frame, summarizing the global characteristics of the noise in that frame.

[0098] Step 1023: Obtain the preset steady-state base linear projection matrix and transient impact linear projection matrix;

[0099] These two projection matrices (steady-state basis, transient shock) are pre-trained using a large amount of labeled noisy data. They are essentially two classifiers or separators.

[0100] Steady-state basis linear projection matrix learning maps contextual features to the frequency band dominated by steady-state noise. Steady-state noise has a narrow-band harmonic structure (such as engine noise), a long-term stable spectral envelope, and a low rate of change over time.

[0101] The steady-state basis linear projection matrix can be trained using data such as car engine noise, air conditioner fan noise, and transformer humming.

[0102] Matrix structure example (simplified):

[0103] Assuming dimension d = 4 (actually 128), number of frequency points F = 5 (actually 129), and frequencies corresponding to [low frequency, low-mid frequency, mid frequency, mid-high frequency, high frequency]:

[0104] Steady-state basis linear projection matrix = .

[0105] Transient impact linear projection matrix learning maps contextual features to the frequency band where the transient impact event occurs. The transient impact linear projection matrix can be trained using data such as knocking sounds, collision sounds, and popping sounds.

[0106] Assuming dimension d = 4 (actually 128), number of frequency points F = 5 (actually 129), and frequencies corresponding to [low frequency, low-mid frequency, mid frequency, mid-high frequency, high frequency]:

[0107] Transient impact linear projection matrix = .

[0108] Step 1024: Input the context feature vector into the steady-state basis linear projection matrix to obtain the time-frequency mask of the steady-state basis noise components;

[0109] Step 1025: Input the context feature vector into the transient impact linear projection matrix to obtain the time-frequency mask of the transient impact noise component;

[0110] Perform matrix multiplication operations between the context feature vectors obtained in the previous step and these two matrices respectively.

[0111] The steady-state basis linear projection matrix is ​​trained to output a high response value when the input features contain steady-state noise characteristics.

[0112] The transient impact linear projection matrix is ​​trained to output a high response value when the input features contain transient noise characteristics.

[0113] Two time-frequency masks are obtained. Each mask is a matrix with the same dimension as the time-frequency representation of the signal frame, and the value of each point is between 0 and 1, representing the probability or intensity of that time-frequency unit belonging to the corresponding noise component.

[0114] Step 1026: Use the time-frequency mask to perform weighted separation on the time-frequency representation of the signal frame to reconstruct the time-domain signal corresponding to each noise component.

[0115] Time-frequency representation refers to the spectrum obtained by performing a short-time Fourier transform on the original signal frame.

[0116] By multiplying (weighting) the time-frequency mask of the steady-state noise by the original time-frequency representation, the time-frequency energy belonging to the steady-state component can be filtered out, thus obtaining the time-frequency representation of the steady-state noise. Similarly, by multiplying the time-frequency mask of the transient impulse noise by the original time-frequency representation, the time-frequency representation of the transient noise can be obtained.

[0117] By performing an inverse short-time Fourier transform on the separated time-frequency representations, the time-domain signals of the steady-state basis noise component and the transient impulse noise component can be reconstructed for use in subsequent steps.

[0118] In the embodiments corresponding to steps 1021 to 1026, multi-level signal processing and feature extraction techniques are used to efficiently and accurately separate and reconstruct the steady-state and transient noise components in the Bluetooth headset, providing a data foundation for subsequent noise reduction signal generation and noise reduction processing.

[0119] Specifically, step 1026 includes steps B1 to B3:

[0120] Step B1: Calculate the short-time Fourier transform of the signal frame to obtain the complex time-frequency representation;

[0121] Applying a short-time Fourier transform to each frame of the signal yields a complex time-frequency representation (a complex matrix). Each complex element in this matrix contains both the energy (amplitude) and phase information of that time-frequency unit (at a specific moment and frequency). This is the ideal domain for refined signal processing.

[0122] Step B2: Multiply the complex time-frequency representation point by point with the time-frequency mask corresponding to each noise component to obtain the separated time-frequency representation corresponding to each noise component;

[0123] Two matrices are used for steady-state floor noise and transient impulse noise. Each mask matrix has the same dimension as the complex time-frequency representation, and the value of each element is between 0 and 1, representing the probability or energy proportion of that time-frequency unit belonging to the corresponding noise component.

[0124] For a steady-state mask, the steady-state separated time-frequency representation = complex time-frequency representation ⊙ steady-state time-frequency mask;

[0125] For transient masks, transient separated time-frequency representation = complex time-frequency representation ⊙ transient time-frequency mask;

[0126] Time-frequency cells with a mask value close to 1 retain almost all of their energy in the corresponding component. Time-frequency cells with a mask value close to 0 have their energy significantly suppressed. This is equivalent to drawing an energy distribution map for each of the two noise components in the time-frequency domain based on the learned features, and extracting their respective components from the mixed signal accordingly.

[0127] The two separated complex time-frequency representations mainly contain the energy and phase information of steady-state noise and transient noise, respectively.

[0128] Step B3: Perform inverse short-time Fourier transform on each of the separated time-frequency representations, and use the overlapping addition method to synthesize a continuous time-domain signal to obtain the time-domain signals corresponding to the steady-state basis noise component and the transient impulse noise component, respectively.

[0129] Perform an inverse short-time Fourier transform on the complex time-frequency representation of each component to obtain a series of time-domain signal frames (each component corresponds to a set of frames). However, these frames are independent and have undergone windowing processing, and cannot be directly spliced ​​together.

[0130] In STFT analysis, to reduce spectral leakage, the signal is usually windowed (e.g., Hamming window), and there is overlap between frames. Directly splicing the inverse-transformed frames will cause distortion at the seams.

[0131] The time-domain frames after inverse transformation are aligned according to their original time positions. The sample values ​​of the overlapping portions are added together. Because the window function used in the analysis meets specific conditions (such as the overlap-addition reconstruction condition), this operation can perfectly eliminate the windowing effect and synthesize a continuous, smooth, and distortion-free complete time-domain signal.

[0132] In the embodiments corresponding to steps B1 to B3, the combination of short-time Fourier transform and time-frequency masking achieves effective separation and reconstruction of different noise components in the signal. The time-domain signal of each noise component can provide important information for subsequent denoising processing or signal analysis. This method not only improves the ability to identify noise components but also ensures the quality and accuracy of the reconstructed signal.

[0133] Step 103: Generate corresponding anti-noise signals based on the multiple noise components and the error signal;

[0134] For each noise component obtained from the decomposition (steady-state and transient), a corresponding anti-noise signal is generated by combining the error signal (i.e., the feedback of the current noise reduction effect).

[0135] This is equivalent to setting up a parallel, dedicated ANC controller for each type of noise. Each controller independently calculates and generates the optimal anti-phase sound wave based on the characteristics of the noise component it is responsible for (such as frequency and dynamic range) and the current noise reduction error. This is more targeted than a single controller handling mixed noise.

[0136] Specifically, step 103 includes steps 1031 to 1037:

[0137] Step 1031: Calculate the time-frequency representation of the error signal;

[0138] The residual noise (error signal) collected by the internal microphone is converted to the time-frequency domain so that it can be correlated with the steady-state noise mask, which is also characterized in the time-frequency domain.

[0139] Step 1032: Calculate the first correlation measure between the time-frequency representation of the error signal and the time-frequency mask corresponding to the steady-state basis noise component;

[0140] The correlation (such as cross-correlation coefficient, coherence, etc.) between the time-frequency representation of the error signal and the time-frequency mask of the steady-state basis noise is analyzed. This correlation metric directly reflects the current adaptive filter's suppression effect on steady-state noise. A high correlation indicates that the error still contains a large number of components related to steady-state noise, meaning the suppression effect is poor; a low correlation indicates a good suppression effect. The calculation process of the correlation metric is a traditional technique and will not be elaborated here.

[0141] Step 1033: Based on the first correlation metric, adjust the adaptive step size of the adaptive filter path;

[0142] When the first correlation metric is high (poor suppression), the adaptive step size is increased to update the filter coefficients more quickly, accelerating convergence to track noise changes. When the first correlation metric is low (good suppression), the adaptive step size is decreased to make the filter coefficient updates smoother, improving steady-state performance and reducing overshoot risk. This achieves a dynamic optimal trade-off between convergence speed and steady-state error, avoiding the contradiction caused by the fixed step size in the traditional FxLMS algorithm (large step size leads to fast convergence but large steady-state error, while small step size leads to small steady-state error but slow convergence).

[0143] Specifically, step 1033 includes steps C1 to C4:

[0144] Step C1: Obtain the high correlation threshold, low correlation threshold, maximum allowable step size, and minimum allowable step size;

[0145] The high correlation threshold is a key parameter. When the first correlation metric is greater than or equal to this threshold, the system determines that the current suppression effect on steady-state noise is very poor, and the error signal still contains a large number of components related to steady-state noise.

[0146] The low correlation threshold is another key parameter. When the first correlation metric is less than or equal to this threshold, the system determines that the suppression effect is very good, and the steady-state noise has been sufficiently canceled.

[0147] The maximum allowable step size is the upper limit that the adaptive filter's step size can be set to. The larger the step size, the faster the filter coefficients are updated and the faster the convergence speed, but it may introduce instability and a larger steady-state error.

[0148] The minimum allowable step size is the lower limit of the allowable step size setting for an adaptive filter. The smaller the step size, the more stable the filter and the smaller the steady-state error, but the slower the convergence and noise tracking speed.

[0149] These parameters define a clear operating range and safety boundary for the entire control logic, making the algorithm behavior predictable and tunable.

[0150] The high correlation threshold can be set to 0.8. The high correlation threshold means that when the correlation between the error signal and a certain noise component is greater than 0.8, this noise component will be considered very important to the noise reduction process and should be given priority.

[0151] The low correlation threshold can be set to 0.3. The low correlation threshold means that when the correlation between the error signal and a certain noise component is less than 0.3, the noise component will be considered to have a limited contribution to the noise reduction process and can be ignored or have its weight reduced.

[0152] The maximum allowable step size can be set to 0.1. This value limits the maximum change during weight updates, ensuring that each weight adjustment does not exceed 0.1. This helps maintain system stability and avoids unnecessary fluctuations caused by overly rapid adjustments.

[0153] The minimum allowable step size can be set to 0.01. This value sets the minimum acceptable change during weight updates, ensuring the system can still effectively respond to changes even with minor adjustments. This helps maintain the system's sensitivity to environmental changes under small variations.

[0154] Step C2: When the first correlation metric is greater than or equal to the high correlation threshold, the maximum allowable step size is used as the adaptive step size;

[0155] When the system determines that the suppression effect is inadequate, it decisively adopts the maximum step size, allowing the filter to update its coefficients at the fastest speed, aggressively tracking changes in external steady-state noise, and striving to rapidly reduce errors. This is an emergency acceleration mode.

[0156] Step C3: When the first correlation metric is less than or equal to the low correlation threshold, the minimum allowable step size is used as the adaptive step size;

[0157] When the system determines that the suppression effect is good, it switches to the minimum step size, making the filter coefficients update very slowly. At this time, the system is in fine-tuning mode, which aims to minimize steady-state error, improve the noise reduction depth, and maintain high system stability.

[0158] Step C4: When the first correlation metric is between the high correlation threshold and the low correlation threshold, the adaptive step size = .

[0159] The step size is calculated using a linear interpolation formula. This achieves a smooth and continuous transition of the step size between the maximum and minimum values.

[0160] As the correlation metric increases from a low threshold to a high threshold (the suppression effect worsens), the step size increases linearly from the minimum to the maximum value.

[0161] This smooth transition avoids abrupt changes in step size at the threshold, thus preventing filter oscillation or performance instability that may be caused by sudden changes in step size, and ensuring the stability of the system's dynamic response.

[0162] In the embodiments corresponding to steps C1 to C4, the adaptive step size adjustment mechanism provides an effective strategy for the filter's adaptability under different noise conditions. This flexible step size adjustment can improve the performance of Bluetooth headphones in the individualized noise cancellation process, enabling them to quickly adapt to environmental changes and optimize noise cancellation effects.

[0163] Step 1034: Input the time-frequency mask of the steady-state basis noise component into the adaptive filter path for processing to obtain the first anti-noise signal;

[0164] The time-frequency mask of the steady-state floor noise component is input into the adaptive filter path with dynamically adjusted adaptive step size (such as the FxLMS algorithm framework) and outputs the first anti-noise signal, which is specifically used to cancel the steady-state floor noise.

[0165] Step 1035: Real-time monitoring of the time-domain energy of transient impact noise components;

[0166] Directly monitoring the energy of transient impulse noise components in the time domain is a direct and rapid method for detecting sudden events.

[0167] When the energy exceeds a preset value, it marks the beginning of a transient event. Once the event is detected, three key pieces of information are immediately extracted: ① Historical samples (first preset duration), capturing background noise and possible precursors before the event. ② Occurring samples (second preset duration from the current moment), capturing waveform features of the initial stage of the event. ③ Context feature vector (extracted from the time-frequency mask), high-dimensional features extracted from the generated transient mask, characterizing the time-frequency properties of the event.

[0168] Step 1036: When the temporal energy exceeds a preset energy value, input the noise segments of the first preset duration historical samples before the current transient event start point, the noise segments of the second preset duration samples that have occurred since the current moment, and the context feature vector extracted from the time-frequency mask of the transient impact noise component into a pre-trained causal convolutional neural network; wherein, the current transient event start point refers to the moment when the temporal energy exceeds the preset energy value;

[0169] Step 1037: Predict the second noise-resistant signal for the third preset duration in the future using the causal convolutional neural network.

[0170] The three segments of information extracted above are combined into an input tensor and fed into a pre-trained causal convolutional neural network. The causal convolutional neural network is a traditional network structure, which will not be elaborated upon here.

[0171] The causal structure of CNNs ensures that their output (prediction) depends only on current and past inputs, and not on future information. This is a necessary physical condition for denoising systems that need to generate noise-resistant signals in real time.

[0172] The CNN model has been trained offline using a large amount of noisy data. Its goal is to learn the mapping relationship from historical / current segments of transient noise to the optimal noise-resistant signal in the future.

[0173] The second noise immunity signal is used to cancel out transient impact noise for a future third preset duration.

[0174] In the embodiments corresponding to steps 1031 to 1037, the combination of adaptive filters and causal convolutional neural networks effectively achieves separate processing of steady-state noise and transient noise. This method not only improves the adaptability to noise but also enhances the system's real-time processing capabilities, providing strong support for personalized noise reduction in Bluetooth headsets.

[0175] Step 104: Calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic wave signal; wherein, the anti-phase acoustic wave signal is used to drive the loudspeaker for noise reduction processing.

[0176] Based on the error signal and the time-domain signals of each component obtained from the decomposition of the original noise, the mixing weight corresponding to each noise component is dynamically calculated. Weight calculation is the core of individualization and adaptation.

[0177] If the error signal shows a large residual amount of a certain type of noise (such as transient impact), the system may automatically increase the weight of the anti-noise signal corresponding to that type of noise to strengthen its suppression.

[0178] At the same time, the system will also refer to the real-time intensity of the time-domain signals of each component (e.g., a sudden increase in transient noise) and adjust the weights in advance or synchronously to achieve a fast response.

[0179] The noise-resistant signals generated by each parallel channel are multiplied by their respective dynamically calculated weights, then summed to form a unified, inverted acoustic signal. This synthesized signal drives the headphone speaker to emit sound waves, which cancel out the original noise within the ear canal, thus achieving noise reduction.

[0180] Specifically, step 104 includes steps 1041 to 1044:

[0181] Step 1041: Calculate the second correlation measure between the error signal and the time-domain signals of each noise component;

[0182] The error signal comes from the internal microphone and represents the noise that remains after the current noise reduction (i.e., the part where the noise reduction is imperfect). The time-domain signals of each noise component are the steady-state floor noise time-domain signal and the transient impulse noise time-domain signal obtained by decomposition.

[0183] Calculate the second correlation metric (such as cross-correlation coefficient or coherence function) between the error signal and the time-domain signal of each noise component. The calculation process for the correlation metric is a traditional technique and is not limited here.

[0184] If the error signal is highly correlated with the steady-state components, it indicates that the current steady-state denoising path (adaptive filter) is ineffective and needs to be strengthened. If the error signal is highly correlated with the transient components, it indicates that the current transient denoising path (CNN predictor) is ineffective and needs to be strengthened.

[0185] Step 1042: Normalize each of the second correlation measures to obtain the initial weighting factors corresponding to each noise component;

[0186] Normalize the calculated two (or more) secondary correlation measures. For example, calculate the proportion of each correlation measure in the sum. Steady-state initial weight factor = steady-state correlation / (steady-state correlation + transient correlation). Transient initial weight factor = transient correlation / (steady-state correlation + transient correlation).

[0187] After normalization, the sum of all weighting factors is 1 (or 100%), which ensures that the total energy of the synthesized antiphase sound wave is controllable, avoiding output saturation or system instability that may result from unconstrained weights. The two weighting factors are inversely related. The noise component that is more relevant to the current error will receive a higher weight in its corresponding denoising path, thus playing a more dominant role in the final antiphase sound wave.

[0188] Step 1043: Perform time-domain smoothing filtering on the initial weight factors to obtain the final mixed weights;

[0189] The initial weighting factors calculated above are then smoothed in the time domain (e.g., using a first-order low-pass filter or moving average).

[0190] The correlation measure and initial weighting factors may fluctuate drastically with rapid changes in the input signal. If these drastically fluctuating weights are used directly to synthesize an out-of-phase sound wave, it will result in discontinuous output sound, potentially producing audible modulation noise, breathing effects, or frequent jumps in the system's operating point, affecting overall stability.

[0191] After smoothing filtering, the final hybrid weights are obtained. These weights change gradually, making the system's switching between different noise reduction modes (steady-state vs. transient) smooth, natural, and imperceptible, greatly improving the user experience.

[0192] Step 1044: Based on the hybrid weights, the anti-noise signals corresponding to each of the multiple noise components are weighted and synthesized into an anti-phase acoustic signal.

[0193] Using the smoothed final mixing weights, the two generated anti-noise signals (first anti-noise signal and second anti-noise signal) are weighted and summed. The inverted acoustic signal = (steady-state mixing weight * first anti-noise signal) + (transient mixing weight * second anti-noise signal).

[0194] The synthesized signal is the final out-of-phase sound wave that drives the loudspeaker to actively cancel out noise.

[0195] In the embodiments corresponding to steps 1041 to 1044, the hybrid weight calculation and noise-resistant signal synthesis process provides a noise reduction strategy for Bluetooth headsets based on dynamic adjustment of error signals and noise components. This method not only considers the correlation of various noise components, but also optimizes the generation of anti-phase sound wave signals through smoothing and weighting, thereby improving the noise reduction performance of the headset in complex noise environments.

[0196] In the embodiments corresponding to steps 101 to 104, by acquiring the original temporal noise signal collected by the external microphone and the error signal collected by the internal microphone, and performing frame-by-frame preprocessing, the noise signal can be effectively decomposed into steady-state floor noise components and transient impulse noise components. This accurate component identification enables subsequent noise reduction processing to be more effective against different types of noise, improving the targeting and effectiveness of noise reduction. By generating corresponding anti-noise signals based on multiple noise components and error signals, the noise reduction strategy can be dynamically adjusted according to the user's specific environment and listening needs. This personalized noise reduction method can provide each user with a personalized noise reduction experience and meet the noise reduction needs in different scenarios.

[0197] like Figure 2 This invention provides a personalized noise reduction system for Bluetooth headsets; please refer to [link / reference]. Figure 2 , Figure 2 A schematic diagram of a personalized noise reduction system for Bluetooth headsets provided by the present invention is shown, as follows: Figure 2 The system shown is a personalized noise cancellation system for Bluetooth headphones, comprising:

[0198] The acquisition unit 21 is used to acquire the original time-domain noise signal outside the earphone collected by the external microphone and the error signal inside the earphone collected by the internal microphone, and to perform frame-by-frame preprocessing on the original time-domain noise signal to obtain continuous signal frames.

[0199] The decomposition unit 22 is used to decompose each of the signal frames into noise components and their time-domain signals with different time-frequency characteristics; wherein, the noise components include steady-state floor noise components and transient impulse noise components;

[0200] The generation unit 23 is used to generate corresponding anti-noise signals based on the multiple noise components and the error signal;

[0201] The synthesis unit 24 is used to calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and to weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic wave signal; wherein the anti-phase acoustic wave signal is used to drive the loudspeaker for noise reduction processing.

[0202] This invention provides a personalized noise reduction system for Bluetooth headsets. By acquiring the raw time-domain noise signal collected by an external microphone and the error signal collected by an internal microphone, and performing frame-by-frame preprocessing, the noise signal can be effectively decomposed into steady-state floor noise components and transient impulse noise components. This precise component identification enables subsequent noise reduction processing to more effectively target different types of noise, improving the targeting and effectiveness of noise reduction. By generating corresponding anti-noise signals based on multiple noise components and error signals, the noise reduction strategy can be dynamically adjusted according to the user's specific environment and listening needs. This personalized noise reduction method can provide each user with a personalized noise reduction experience, meeting the noise reduction needs in different scenarios.

[0203] Figure 3 This is a schematic diagram of a terminal device provided in an embodiment of the present invention. Figure 3 As shown, a terminal device 3 in this embodiment includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30, such as a program for personalized noise reduction of a Bluetooth headset. When the processor 30 executes the computer program 32, it implements the steps in the various embodiments of the method for personalized noise reduction of a Bluetooth headset described above, for example... Figure 1 Steps 101 to 104 are shown. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each unit in the above-described device embodiments, for example... Figure 2 The function of the unit shown.

[0204] For example, the computer program 32 can be divided into one or more units, which are stored in the memory 31 and executed by the processor 30 to complete the present invention. The one or more units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 32 in the terminal device 3. For example, the specific functions of each unit of the computer program 32 can be divided as follows:

[0205] The acquisition unit is used to acquire the original time-domain noise signal outside the earphone collected by the external microphone and the error signal inside the earphone collected by the internal microphone, and to perform frame-by-frame preprocessing on the original time-domain noise signal to obtain continuous signal frames.

[0206] A decomposition unit is used to decompose each of the signal frames into noise components and their time-domain signals with different time-frequency characteristics; wherein the noise components include steady-state floor noise components and transient impulse noise components;

[0207] The generation unit is configured to generate corresponding anti-noise signals based on the multiple noise components and the error signal;

[0208] The synthesis unit is used to calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and to weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic wave signal; wherein the anti-phase acoustic wave signal is used to drive the loudspeaker for noise reduction processing.

[0209] The terminal device includes, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of a terminal device 3 and does not constitute a limitation on a terminal device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0210] The processor 30 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0211] The memory 31 can be an internal storage unit of the terminal device 3, such as a hard disk or memory of the terminal device 3. The memory 31 can also be an external storage device of the terminal device 3, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device 3. Furthermore, the memory 31 can include both internal and external storage units of the terminal device 3. The memory 31 is used to store the computer program and other programs and data required by the roaming control device. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0212] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0213] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0214] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0215] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0216] This invention provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to implement the steps described in the above-described method embodiments.

[0217] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0218] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0219] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0220] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0221] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units.

[0222] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0223] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0224] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0225] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0226] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0227] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for individualized noise reduction in Bluetooth headsets, characterized in that, The individualized noise reduction method for the Bluetooth headset includes: The system acquires the original temporal noise signal from the outside of the earphone via an external microphone and the error signal from the inside of the earphone via an internal microphone. The original temporal noise signal is preprocessed by frame segmentation to obtain continuous signal frames. The error signal is the residual noise that actually enters the ear after noise reduction. Each signal frame is decomposed into noise components and their time-domain signals with different time-frequency characteristics; wherein, the noise components include steady-state floor noise components and transient impulse noise components; Based on the multiple noise components and the error signal, a corresponding anti-noise signal is generated; The mixing weights corresponding to each noise component are calculated based on the error signal and the time domain signal, and the anti-noise signals corresponding to each of the multiple noise components are weighted and synthesized into an anti-phase acoustic wave signal. The step of generating corresponding anti-noise signals based on multiple noise components and the error signal includes: Calculate the time-frequency representation of the error signal; Calculate a first correlation metric between the time-frequency representation of the error signal and the time-frequency mask corresponding to the steady-state basis noise component; the first correlation metric is the suppression effect of the current adaptive filter on steady-state noise; Based on the first correlation metric, adjust the adaptive step size of the adaptive filter path; The time-frequency mask of the steady-state basis noise component is input into the adaptive filter path for processing to obtain the first anti-noise signal; Real-time monitoring of the time-domain energy of transient impact noise components; When the temporal energy exceeds a preset energy value, noise segments of historical samples with a first preset duration before the current transient event start point, noise segments of samples that have occurred with a second preset duration since the current moment, and the context feature vector extracted from the time-frequency mask of the transient impact noise component are input into a pre-trained causal convolutional neural network; wherein, the current transient event start point refers to the moment when the temporal energy exceeds the preset energy value; The second anti-noise signal with a third preset duration is predicted using the causal convolutional neural network. The step of calculating the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and weighting and synthesizing the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic signal includes: Calculate the second correlation measure between the error signal and the time-domain signals of each noise component; Normalize each of the second correlation measures to obtain the initial weighting factor corresponding to each noise component; The initial weighting factors are subjected to time-domain smoothing filtering to obtain the final mixed weights; Based on the aforementioned hybrid weights, the anti-noise signals corresponding to each of the multiple noise components are weighted and synthesized into an anti-phase acoustic signal; wherein, the anti-phase acoustic signal is used to drive the loudspeaker for noise reduction processing.

2. The method for individualized noise reduction of Bluetooth headsets as described in claim 1, characterized in that, The step of decomposing each signal frame into noise components and their time-domain signals with different time-frequency characteristics includes: Multi-layer wavelet scattering transform is performed on the signal frame to generate a high-dimensional scattering feature vector that includes information on energy distribution and envelope changes at different time scales; The high-dimensional scattering feature vector is reduced in dimensionality by a preset feature encoder to obtain a low-dimensional context feature vector. Obtain the preset steady-state basis linear projection matrix and transient impact linear projection matrix; The context feature vector is input into the steady-state basis linear projection matrix to obtain the time-frequency mask of the steady-state basis noise component; The context feature vector is input into the transient impact linear projection matrix to obtain the time-frequency mask of the transient impact noise component; The time-frequency representation of the signal frame is weighted and separated using the time-frequency mask to reconstruct the time-domain signal corresponding to each noise component.

3. The method for individualized noise reduction of Bluetooth headsets as described in claim 2, characterized in that, The step of performing multi-layer wavelet scattering transform on the signal frame to generate a high-dimensional scattering feature vector including energy distribution and envelope variation information at different time scales includes: The signal frame is convolved and moduloed using the first set of complex wavelet filters to obtain the first layer of modulo signal. The first layer modulus signal is low-pass filtered and downsampled to obtain the first layer scattering coefficients that characterize the steady-state spectrum of the signal. The first layer modulus signal is convolved and moduloed using the second set of complex wavelet filters to obtain the second layer modulus signal; The second-layer modulus signal is low-pass filtered and downsampled to obtain the second-layer scattering coefficients that characterize the signal energy envelope variation mode. The scattering coefficients of the first layer and the scattering coefficients of the second layer are concatenated to form the high-dimensional scattering feature vector.

4. The method for individualized noise reduction of Bluetooth headsets as described in claim 2, characterized in that, The step of using the time-frequency mask to perform weighted separation of the time-frequency representation of the signal frame to reconstruct the time-domain signal corresponding to each noise component includes: Calculate the short-time Fourier transform of the signal frame to obtain a complex time-frequency representation; The complex time-frequency representation is multiplied point by point with the time-frequency mask corresponding to each noise component to obtain the separated time-frequency representation corresponding to each noise component; The inverse short-time Fourier transform is performed on each of the separated time-frequency representations, and the continuous time-domain signal is synthesized by the overlapping addition method to obtain the time-domain signals corresponding to the steady-state basis noise component and the transient impulse noise component.

5. The method for individualized noise reduction of Bluetooth headsets as described in claim 1, characterized in that, The step of adjusting the adaptive step size of the adaptive filter path based on the first correlation metric includes: Obtain the high correlation threshold, low correlation threshold, maximum allowable step size, and minimum allowable step size; When the first correlation metric is greater than or equal to the high correlation threshold, the maximum allowable step size is used as the adaptive step size; When the first correlation metric is less than or equal to the low correlation threshold, the minimum allowable step size is used as the adaptive step size; When the first correlation metric is between the high correlation threshold and the low correlation threshold, the adaptive step size is... .

6. A personalized noise cancellation system for Bluetooth headsets, characterized in that, The personalized noise cancellation system for the Bluetooth headset includes: The acquisition unit is used to acquire the original temporal noise signal outside the earphone collected by the external microphone and the error signal inside the earphone collected by the internal microphone, and to perform frame-by-frame preprocessing on the original temporal noise signal to obtain continuous signal frames; the error signal is the residual noise that actually enters the ear after noise reduction. A decomposition unit is used to decompose each of the signal frames into noise components and their time-domain signals with different time-frequency characteristics; wherein the noise components include steady-state floor noise components and transient impulse noise components; The generation unit is configured to generate corresponding anti-noise signals based on the multiple noise components and the error signal; The synthesis unit is used to calculate the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and to weight and synthesize the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic signal. The step of generating corresponding anti-noise signals based on multiple noise components and the error signal includes: Calculate the time-frequency representation of the error signal; Calculate a first correlation metric between the time-frequency representation of the error signal and the time-frequency mask corresponding to the steady-state basis noise component; the first correlation metric is the suppression effect of the current adaptive filter on steady-state noise; Based on the first correlation metric, adjust the adaptive step size of the adaptive filter path; The time-frequency mask of the steady-state basis noise component is input into the adaptive filter path for processing to obtain the first anti-noise signal; Real-time monitoring of the time-domain energy of transient impact noise components; When the temporal energy exceeds a preset energy value, noise segments of historical samples with a first preset duration before the current transient event start point, noise segments of samples that have occurred with a second preset duration since the current moment, and the context feature vector extracted from the time-frequency mask of the transient impact noise component are input into a pre-trained causal convolutional neural network; wherein, the current transient event start point refers to the moment when the temporal energy exceeds the preset energy value; The second anti-noise signal with a third preset duration is predicted using the causal convolutional neural network. The step of calculating the mixing weights corresponding to each noise component based on the error signal and the time-domain signal, and weighting and synthesizing the anti-noise signals corresponding to each of the multiple noise components into an anti-phase acoustic signal includes: Calculate the second correlation measure between the error signal and the time-domain signals of each noise component; Normalize each of the second correlation measures to obtain the initial weighting factor corresponding to each noise component; The initial weighting factors are subjected to time-domain smoothing filtering to obtain the final mixed weights; Based on the hybrid weights, the anti-noise signals corresponding to each of the multiple noise components are weighted and synthesized into an anti-phase acoustic signal; wherein, the anti-phase acoustic signal is used to drive the loudspeaker for noise reduction processing; wherein, the anti-phase acoustic signal is used to drive the loudspeaker for noise reduction processing.

7. A terminal device, characterized in that, The terminal device includes: a memory, a processor, and a Bluetooth headset personalized noise reduction program stored in the memory and executable on the processor, the Bluetooth headset personalized noise reduction program being configured to implement the steps of the Bluetooth headset personalized noise reduction method as claimed in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the personalized noise reduction method for the Bluetooth headset as described in any one of claims 1 to 6.

Images