Noise elimination method and device, electronic equipment and vehicle

Abstract

The invention provides a noise elimination method and device, electronic equipment and a vehicle, and the method comprises the steps: collecting time series data of environment noise in the vehicle, analyzing the time series data, and determining a main noise type and noise intensity spatial distribution; activating a target noise reduction channel of a target partition in the vehicle through noise intensity spatial distribution, and determining an original reference signal of a reference microphone in the target noise reduction channel through a main noise type; performing secondary path estimation and forward path calculation of a preset adaptive filter on the original reference signal to obtain a first noise reduction driving signal for the loudspeaker; acquiring a historical reference signal of the reference microphone and a training noise signal of the error microphone, and inputting the historical reference signal and the training noise signal into a preset depth active noise control network to output a second noise reduction driving signal; and adaptive weighting is performed on the first noise reduction driving signal and the second noise reduction driving signal to obtain a target noise reduction driving signal, and elimination of the vehicle environment noise is completed through the target noise reduction driving signal, so that the user experience is improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a noise cancellation method, apparatus, electronic device, and vehicle. Background Technology

[0002] With the increasing demands for driving comfort in automobiles, in-vehicle noise management has become an important research topic in automotive design. Existing in-vehicle noise management technologies mainly include passive noise reduction measures, represented by sound insulation materials and structural optimization, and active noise reduction technologies based on adaptive filtering algorithms. Passive noise reduction technologies attenuate noise propagation through physical blocking and absorption, while active noise reduction technologies achieve noise cancellation by generating anti-phase sound waves with the opposite phase to the noise.

[0003] However, the above solutions have obvious drawbacks: on the one hand, passive noise reduction measures have limited effectiveness against low-frequency noise and are difficult to cope with complex and ever-changing noise environments; on the other hand, active noise reduction technology performs well under low-frequency stable noise, but its noise reduction effect drops significantly when there are high frequencies, sudden noises, or rapid environmental changes. Furthermore, it is not adaptable enough to handle nonlinear and non-stationary noise, and its separation and cancellation effect on multi-source composite noise is poor, making it difficult to meet the increasingly demanding requirements for automotive driving comfort. Summary of the Invention

[0004] In view of this, this application aims to propose a noise cancellation method, device, electronic device, and vehicle to solve the problems that current noise reduction technologies suffer from a significant decrease in noise reduction effect under high frequency, sudden, or rapidly changing environmental conditions, insufficient adaptability in handling nonlinear and non-stationary noise, and poor separation and cancellation effect on multi-source composite noise, making it difficult to meet the increasingly demanding requirements for automotive driving comfort. The specific technical solution is as follows: According to a first aspect of this application, a noise cancellation method is provided, the method comprising: Time-series data of ambient noise in the vehicle were collected using several microphones; Analyze the time-series data to determine the dominant noise type and spatial distribution of noise intensity in the environment; By using the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated, and the original reference signal of the reference microphone in the target noise reduction channel is determined by the main noise type. Secondary path estimation and forward path calculation of a preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel; The historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel are obtained, and the historical reference signal and the training noise signal are input into a preset depth active noise control network to output a second noise reduction driving signal. The first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal, and the vehicle environmental noise is eliminated by the target noise reduction driving signal.

[0005] Optionally, after acquiring time-series data of ambient noise in the vehicle via several microphones, the process includes: The time-series data are subjected to bandpass filtering and wavelet transform to obtain wavelet coefficients; The dominant noise type of the environmental noise is identified using the time-series data; If the main noise type is a low-frequency stable main noise type, then the wavelet coefficients are processed using a soft thresholding algorithm to obtain the first selected wavelet coefficients. If the main noise type is a high-frequency main noise type or a burst main noise type, then the wavelet coefficients are processed using a hard threshold algorithm to obtain the second selected wavelet coefficients. Perform inverse wavelet transform on the first or second selected wavelet coefficients to obtain denoised time series data; The denoised time series data is normalized and subjected to adaptive gain control to obtain preprocessed time series data.

[0006] Optionally, analyzing the time-series data to determine the dominant noise type and spatial distribution of noise intensity in the environment includes: The preprocessed time-series data is divided into continuous time-domain frame data; Perform a discrete Fourier transform on the time-domain frame data to obtain the frequency domain complex spectrum corresponding to the time-domain frame data; The Mel energy spectrum of the ambient noise, the time difference between the arrival of the ambient noise at any two microphones, and the spatial spectral power between the ambient noise and each microphone are determined by the frequency domain complex spectrum. The Mel frequency cepstral coefficients of the environmental noise are determined using the Mel energy spectrum. The dominant noise type and spatial distribution of noise intensity of the environmental noise are determined by the Mel energy spectrum, the Mel frequency cepstral coefficients, the time difference, and the spatial spectral power.

[0007] Optionally, activating the target noise reduction channel for the target zone in the vehicle based on the spatial distribution of noise intensity further includes: The noise intensity of several zones of the vehicle is determined by the spatial distribution of the noise intensity. Target zones with noise intensity greater than the intensity threshold are selected, and the target noise reduction channel of the target zone in the vehicle is activated.

[0008] Optionally, determining the original reference signal of the reference microphone within the target noise reduction channel based on the main noise type includes: Correlation analysis was performed on several reference microphones in the target noise reduction channel based on the main noise type to obtain the correlation coefficient; The correlation coefficient is used to assign a weight to each reference microphone in the target partition, wherein the larger the correlation coefficient, the larger the weight is assigned. The timing data acquired by the reference microphone is weighted by weighting to obtain the original reference signal of the reference microphone in the target noise reduction channel.

[0009] Optionally, after performing secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the speaker in the target noise reduction channel, the method further includes: The first noise reduction driving signal drives the loudspeaker in the noise reduction channel to generate a first reverse sound wave. The current residual noise signal received by the error microphone in the target noise reduction channel is identified. The current residual noise signal is generated based on the linear superposition of ambient noise and the first reverse sound wave in the sound field. The weights of the preset adaptive filter are adjusted based on the current residual noise signal.

[0010] Optionally, after adaptively weighting the first noise reduction driving signal and the second noise reduction driving signal to obtain the target noise reduction driving signal, the method further includes: The target noise reduction driving signal drives the speaker in the noise reduction channel to generate a second reverse sound wave. The actual residual noise signal received by the error microphone in the target noise reduction channel is identified. The actual residual noise signal is generated based on the linear superposition of environmental noise and the second reverse sound wave in the sound field. Calculate the power of the actual residual noise signal; If the power of the actual residual noise signal is greater than the power threshold, the weights for adaptive weighting are adjusted.

[0011] According to a second aspect of this application, a noise cancellation device is provided, the device comprising: The first acquisition module is used to acquire time-series data of environmental noise in the vehicle through several microphones; The first determining module is used to analyze the time-series data to determine the main noise type and spatial distribution of noise intensity of the environmental noise. The second determining module is used to activate the target noise reduction channel of the target partition in the vehicle through the spatial distribution of noise intensity, and to determine the original reference signal of the reference microphone in the target noise reduction channel through the main noise type. The calculation module is used to perform secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel. The input / output module is used to acquire the historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel, and input the historical reference signal and the training noise signal into a preset deep active noise control network, and output a second noise reduction driving signal. The weighting module is used to adaptively weight the first noise reduction driving signal and the second noise reduction driving signal to obtain the target noise reduction driving signal, and to eliminate the vehicle's environmental noise through the target noise reduction driving signal.

[0012] According to another aspect of this application, an electronic device is also provided, comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the noise cancellation method described above.

[0013] According to another aspect of this application, a vehicle is also provided, including the aforementioned noise cancellation device.

[0014] The noise cancellation method provided in this application collects time-series data of environmental noise in a vehicle using several microphones; analyzes the time-series data to determine the dominant noise type and spatial distribution of noise intensity; identifying the dominant noise type facilitates the separation of multi-source composite noise, providing a basis for subsequently increasing the weight of microphones that require special attention, thus achieving precise removal of the dominant noise; determining the spatial distribution of noise intensity provides a basis for subsequent zone activation; based on the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated; based on the dominant noise type, the original reference signal of the reference microphone in the target noise reduction channel is determined; secondary path estimation and forward path calculation of the preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the speaker in the noise reduction channel, realizing real-time tracking and cancellation of environmental noise, ensuring timely response; and simultaneously acquiring the historical reference signal of the reference microphone and the error signal in the target noise reduction channel. The training noise signal from the microphone is compared with the historical reference signal and the training noise signal, and then input into a preset deep active noise control network to output a second noise reduction driving signal. The preset deep active noise control network can learn and predict difficult-to-handle nonlinear and non-stationary noise. By analyzing historical signals, it captures the changing trend of noise and realizes feedforward predictive noise reduction to quickly respond to high-frequency, sudden or rapidly changing noise and reduce convergence delay. Finally, the first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal. The target noise reduction driving signal is used to eliminate the vehicle's environmental noise. At this time, the adaptive weighting can dynamically adjust the contribution weight of the two paths according to the real-time noise scene, ensuring that the "current optimal" control signal can be output under any operating condition, thereby comprehensively improving the efficiency, fault tolerance and robustness of the noise reduction process and improving the user's driving experience.

[0015] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of the steps of a noise cancellation method provided in this application; Figure 2 yes Figure 1 The flowchart shown is a step 102 of a noise cancellation method provided in this application; Figure 3 This is a schematic diagram of the structure of a noise cancellation device provided in this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to facilitate a better understanding of the application. However, the technical solutions claimed in this application can be implemented even without these technical details and with various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0018] Current in-vehicle noise management technologies include passive noise reduction measures, represented by sound insulation materials and structural optimization, and active noise reduction technologies based on signal processing algorithms such as adaptive filtering. Passive noise reduction measures struggle to cope with complex and variable noise sources, while active noise reduction technologies have limitations in handling nonlinear, non-stationary noise, and multi-source composite noise, resulting in limited noise reduction effectiveness and difficulty in adapting to complex driving environments. Based on these problems, this application proposes a noise elimination method. (Refer to...) Figure 1 The diagram illustrates a flowchart of a noise cancellation method provided in this application, the method comprising: Step 101: Collect time-series data of ambient noise in the vehicle using several microphones.

[0019] This application deploys multiple microphones and speakers within the vehicle. The microphone deployment locations are determined based on the vehicle's primary noise sources. These primary noise sources may include at least one of the following: range extender, tires, seats, curtains, engine compartment, sunshades, doors, and chassis. Microphones are deployed within a certain range of these primary noise sources (this range can be adjusted as needed, and this application does not specifically limit it), with at least one microphone deployed at each primary noise source. When collecting ambient noise within the vehicle, microphones in different deployment locations will collect the ambient noise from the nearest corresponding noise source; that is, the range at which microphones collect ambient noise is limited. For example, if the passenger side experiences the sound of a seat moving, a microphone within a certain range of the passenger side seat will collect this noise, while microphones in other locations (such as those near the range extender) will not.

[0020] When ambient noise is collected via a microphone, the microphone generates a continuously changing analog voltage signal. Since computers cannot directly process such continuous signals, this application employs synchronous sampling technology to convert it into digital form. All sampling channels share a unified sampling clock, ensuring that the sampling timestamps of each channel are perfectly aligned and avoiding signal time base drift. The sampling frequency can be set to 16kHz~48kHz, with 16-24 bit precision. Each channel is sampled and quantized independently according to this clock, resulting in discrete sampled data. Subsequently, the continuous sampled data stream of each channel is divided into continuous multi-frame data, each frame containing a fixed number of sampling points. These frames, arranged in chronological order, together constitute the timing data of each microphone channel. This application assigns synchronization frame numbers to these timing data. Where m is the sampling channel number and n is the frame number within the sampling channel. For example, if there are 2 sampling channels, and each channel has 4 frames, then the frame number of sampling channel 1 is... The frame number of sampling channel 2 is Each sampling point corresponds to a unique frame number n, and the frame number increments uniformly across all channels to facilitate subsequent temporal alignment and spatial analysis of multi-channel signals.

[0021] After obtaining the time series data, this application first preprocesses it, including the following steps: Bandpass filtering and wavelet transform are performed on the time series data to obtain wavelet coefficients; Identify the main noise type in the environment using time-series data; If the main noise type is a low-frequency stable main noise type, then the wavelet coefficients are processed using a soft thresholding algorithm to obtain the first selected wavelet coefficients. If the main noise type is high-frequency main noise or burst main noise, then the wavelet coefficients are processed using the hard threshold algorithm to obtain the second selection wavelet coefficients. Perform inverse wavelet transform on the first or second selected wavelet coefficients to obtain denoised time series data. The denoised time series data is normalized and adaptively gain controlled to obtain preprocessed time series data.

[0022] When performing bandpass filtering, a finite impulse response (FIR) filter or an infinite impulse response (IIR) filter can be selected. When setting the filter parameters, the bandwidth can be set to 100Hz~4kHz, and this bandwidth can be adjusted appropriately according to the noise situation. FIR filters can be designed using window functions (such as Hamming windows or Hanning windows), with orders typically ranging from 40 to 200. IIR filters can be Butterworth, Chebyshev, etc., with orders typically ranging from 4 to 8. The processing formula (1) for bandpass filtering is: (1) in, It is the data after bandpass filtering at time n. These are the bandpass filter coefficients, where K is the length / order of the bandpass filter, k is the tap index of the bandpass filter, corresponding to a time delay, and nk refers to the time step k sampling points backward from time n, often called "time nk". This is the sampled data of the m-th channel at time nk. Because during bandpass filtering, a data sliding window of length K is set at the current time n, this window covers K consecutive input samples [x] from the past n-(K-1) times to the current time n. m [n-(K-1)], ..., x m [n-1], x m [n]], It is to weight all input samples within a data sliding window (corresponding to...) The result of summing.

[0023] When performing wavelet denoising, common wavelet basis functions such as Daubechies (db4, db8), Symlet, and Coiflet can be selected. The choice depends on the data. For example, if the data is smooth and complex, the db8 function can be selected. The formula (2) for wavelet denoising is: (2) in, It is the data after bandpass filtering at time n. It is a threshold parameter. Yes Perform multi-scale wavelet transform to obtain wavelet coefficients. The wavelet coefficients are selected by applying a threshold function to the wavelet coefficients. = ,at this time, This involves selecting wavelet coefficients. It's important to note that Threshold(·, λ) here doesn't refer to a single, fixed function, but rather a family of functions. This family includes different specific threshold functions, typically soft-thresholding and hard-thresholding functions. In actual calculations, a specific function can be chosen for substitution. The function type should be selected based on the type of dominant noise. For example, if the dominant noise is low-frequency stable noise, a soft-thresholding function is preferred. If the dominant noise is high-frequency or bursty, a hard-thresholding function is preferred. This involves performing an inverse wavelet transform on the selected wavelet coefficients. , This is denoised time-series data, in which multi-scale wavelet transform W(·) and inverse wavelet transform are performed. When (·), the core tools (basis functions) used internally can be selected from common wavelet basis functions.

[0024] Formula (3) for normalizing denoised time series data is: (3) in, Yes The normalization process yields the normalized time series data for the m-th channel at time n. It is denoised time series data, and max(.) means to take the maximum value.

[0025] The formula (4) for adaptive gain control is:

[0026] (4) in, Yes The timing data of the m-th channel at time n obtained by adaptive gain control is obtained because the preprocessing of the timing data of the ambient noise in the vehicle collected by several microphones has been completed at this time. It refers to the preprocessed time series data of the m-th channel at time n. It is the normalized time series data of the m-th channel at time n. It is the gain control factor for the m-th channel. It is the expected root mean square magnitude of the time series data. It is the actual root mean square magnitude of the time series data.

[0027] The above steps combine adaptive threshold selection with wavelet denoising to achieve precise and dynamic noise reduction for different noise characteristics. Furthermore, by identifying the main noise type (stable low-frequency, high-frequency, or bursty) and automatically matching the optimal threshold algorithm (soft thresholding for smooth suppression of low-frequency noise, and hard thresholding for sharp removal of high-frequency and bursty noise), the useful components of the original signal can be preserved to the greatest extent while effectively removing noise. Finally, high-quality, amplitude-consistent preprocessed time-series data is output through normalization and gain control.

[0028] Step 102: Analyze the time series data to determine the main noise type and spatial distribution of noise intensity in the environment.

[0029] After obtaining the preprocessed time-series data, this application needs to convert it into frequency domain data through Discrete Fourier Transform (DFT). Since DFT can only process data blocks of finite length at a time, the preprocessed time-series data must first be divided into continuous time-domain frame data, and then subjected to DFT to obtain the frequency domain complex spectrum corresponding to the time-domain frame data. The formula (5) for the transformation is: (5) in, It is the m-th channel, the... In the time-domain frame data of a frame, the sampled value (amplitude) at the i-th time point, where L is the frame length of the Discrete Fourier Transform, i.e., the total number of sampled points contained in a frame. is the complex basis function of the discrete Fourier transform, j is the imaginary unit, k refers to the k-th frequency point, and i refers to the i-th time point. It is the m-th channel, the... The time-domain frame data of a frame, after undergoing a discrete Fourier transform, is represented by its complex frequency domain value at the k-th frequency point. By iterating through all possible frequency indices (typically from 0 to L-1) with k, the complex frequency spectrum corresponding to the time-domain frame data is obtained. Here, the m-th channel, the... The frequency domain complex spectrum corresponding to the time-domain frame data of the frame is:

[0030] The Mel energy spectrum of environmental noise is obtained by processing the frequency domain complex spectrum using the Mel filter bank. The processing formula (6) is as follows: (6) in, These are the parameters of the Mel filter. It is the m-th channel, the... The complex spectrum in the frequency domain corresponding to the time-domain frame data of the frame. It is the m-th channel, the... The frame represents the Mel energy output of the b-th Mel filter, where L is the frame length of the Discrete Fourier Transform. The Mel energy spectrum is obtained by integrating the Mel energies output from all Mel filters in the Mel filter bank. The Mel filter bank maps spectral energy to the Mel frequency scale by simulating the human ear's frequency response. The Mel frequency scale simulates the frequency response of the human ear, with high resolution at low frequencies and sparseness at high frequencies, better reflecting human auditory perception. For speech signals, it can better capture features such as timbre, vowel, and consonant variations. As input for deep learning / features in traditional speech recognition, Mel energy can more accurately reflect useful speech information.

[0031] Taking the logarithm of the Mel energy output of all Mel filters in the Mel filter bank and performing a discrete cosine transform yields the Mel frequency cepstral coefficients, processed by formula (7): (7) in, It is the m-th channel, the... Frame, c-th MFCC coefficient, where c is the index of the Mel-Frequency Cepstral Coefficients (MFCCs), B is the total number of filters in the Mel filter bank, and b is the b-th Mel filter in the Mel filter bank. It is the m-th channel, the... The frame represents the Mel energy output of the b-th Mel filter. MFCC coefficients are commonly used as inputs for speech recognition and deep learning.

[0032] This application also performs cross-correlation calculations based on the frequency domain complex spectrum. In this case, the frequency domain complex spectrum corresponding to the time-domain frame data of the i-th channel is used... This means that the frequency domain complex spectrum corresponding to the time-domain frame data of the j-th channel is represented by... The time difference between the arrival of ambient noise at the microphones of the two channels is calculated using formula (8): (8) in, It is the complex spectrum in the frequency domain acquired by the microphone of the i-th channel. It is the complex spectrum in the frequency domain acquired by the microphone of the j-th channel. It is the conjugate of the complex spectrum in the frequency domain acquired by the microphone of the i-th channel. It is the product of the amplitudes of the complex spectrum in the frequency domain of the two channels (normalized to eliminate the influence of amplitude). It is a time-domain cross-correlation function, representing the time difference between the arrival of ambient noise at the two microphones. It is the inverse Fourier transform.

[0033] In addition, this application also calculates the spatial spectral power between the ambient noise and each microphone, using formula (9): (9) in, It is the steering vector of angle θ (which describes the phase difference caused by the path difference when a plane wave from the θ direction arrives at each element of the microphone array). It is the noise subspace matrix. It is the noise subspace matrix The conjugate transpose of . It is the spatial spectral power of the ambient noise and the microphone at an azimuth angle θ, where θ is the assumed angle at which the ambient noise reaches the microphone. It is the conjugate transpose of the guiding vector a(θ).

[0034] The time-frequency features of multiple channels (Mel energy spectrum, Mel frequency cepstral coefficients, time difference, and spatial spectral power) are concatenated along their feature dimensions and normalized to a unified range to form the input tensor. Normalization can employ methods such as mean-variance standardization and Min-Max scaling to ensure a more balanced distribution of features with different dimensions and distributions during model training. The resulting input tensor for one frame can be represented as:

[0035] The input tensor is fed into a deep learning noise recognition and spatial localization network. This network is a multi-task, multimodal deep learning network, internally configured with channel-dimensional convolution (Spatial CNN) to capture spatial correlations; temporal modeling (Bi-LSTM / Transformer) to model short-term dynamics; and an attention mechanism / spatial mask to improve partition sensitivity. The output is the probability of noise type. Spatial distribution heat map Based on noise type probability The probability of different identified dominant noise types can be determined, and based on this, the dominant noise type of the environment (the noise type with a probability greater than a certain value) can be determined. The spatial distribution of noise intensity is determined through a spatial distribution heatmap, thus achieving spatial noise localization. According to the above, step 102 specifically includes the following sub-steps, such as... Figure 2 As shown: Sub-step 1021: The preprocessed time-series data is divided into continuous time-domain frame data.

[0036] Sub-step 1022: Perform a discrete Fourier transform on the time-domain frame data to obtain the frequency domain complex spectrum corresponding to the time-domain frame data.

[0037] Sub-step 1023: Determine the Mel energy spectrum of the ambient noise, the time difference between the ambient noise and any two microphones, and the spatial spectral power between the ambient noise and each microphone by using the frequency domain complex spectrum.

[0038] Sub-step 1024: Determine the Mel frequency cepstral coefficients of the environmental noise using Mel energy spectrum.

[0039] Sub-step 1025 determines the main noise type and spatial distribution of noise intensity of the environment noise by using Mel energy spectrum, Mel frequency cepstral coefficient, time difference and spatial spectral power.

[0040] The above steps, through joint analysis of the multidimensional frequency domain and spatial characteristics of the signal, achieve a comprehensive and detailed characterization of environmental noise. Not only can the spectral characteristics and type of noise be accurately identified using Mel energy spectrum and cepstral coefficients, but the spatial distribution and intensity of noise can also be precisely located by combining time difference and spatial spectral power. This integrates noise "attribute identification" and "spatial location," providing a solid data foundation for subsequent directional noise reduction and spatial sound field control.

[0041] Furthermore, this application will also train a deep learning noise recognition and spatial localization network. During training, based on the output... and And human-labeled , To determine the error between the two, a loss function is set. Based on the loss function, the model parameters of the deep learning noise recognition and spatial localization network are adjusted to reduce the error.

[0042] Step 103: Activate the target noise reduction channel of the target partition in the vehicle by means of the spatial distribution of noise intensity, and determine the original reference signal of the reference microphone in the target noise reduction channel by means of the main noise type.

[0043] This application pre-divides the vehicle's space into several zones (such as the driver's side zone, passenger side zone, left rear zone, right rear zone, and exterior zone), and identifies the noise intensity of each zone in the spatial distribution of noise intensity. Each zone corresponds to a set of microphones and speakers. This application determines the zone to be activated based on the noise intensity of each zone. When the noise intensity of a zone is greater than an intensity threshold (this value is set according to requirements or experimental results), the zone is identified as the target zone, and the target noise reduction channel of the target zone is activated. When the noise intensity of a zone is less than or equal to the intensity threshold, the noise reduction channel of that zone is not activated. This setting can focus on eliminating noise in the target zone, saving resources and reducing interference to non-target zones. Therefore, step 103, "activating the target noise reduction channel of the target zone in the vehicle through the spatial distribution of noise intensity," specifically includes the following sub-steps: Step 1031: Determine the noise intensity of several zones of the vehicle by analyzing the spatial distribution of noise intensity.

[0044] Step 1032: Filter out target partitions with noise intensity greater than the intensity threshold, and activate the target noise reduction channel of the target partition in the vehicle.

[0045] The above steps combine spatial noise localization with zoned control to achieve precise and efficient active noise cancellation. It can intelligently identify specific areas with excessive noise intensity (such as near tires, engine compartment, etc.) and activate noise cancellation only for these "target zones," thereby avoiding ineffective energy loss throughout the vehicle and saving resources while improving noise cancellation effectiveness.

[0046] After activating the target noise reduction channel of the target partition, this application first performs a correlation analysis on the reference microphones in the target noise reduction channel according to the main noise type, selects the reference microphones with high correlation, and assigns them a larger weight, so that the original reference signal of the reference microphone is more focused on the main noise. Among them, the formula (10) for obtaining the original reference signal is: (10) in, It is the weight of the m-th channel reference microphone. This is the preprocessed time series data of the m-th channel at time n. The preprocessing operations include the aforementioned bandpass filtering, wavelet transform, normalization, and adaptive gain control. is the original reference signal of the reference microphone within the target noise reduction channel, m is the m-th channel, and R is the set of reference microphones within the m-th channel. The reference microphone is the microphone within the target noise reduction channel used to receive ambient noise. Therefore, step 103, "determining the original reference signal of the reference microphone within the target noise reduction channel based on the main noise type," specifically includes the following sub-steps: Step 1033: Perform correlation analysis on several reference microphones in the target noise reduction channel based on the main noise type to obtain the correlation coefficient.

[0047] Step 1034: Assign weights to each reference microphone in the target partition based on the correlation coefficient, where the larger the correlation coefficient, the greater the weight.

[0048] Step 1035: Weight the timing data collected by the reference microphone to obtain the original reference signal of the reference microphone in the target noise reduction channel.

[0049] For example, suppose the target zone is the driver's zone, and there are four reference microphones in this zone: Ref1, Ref2, Ref3, and Ref4. Ref1 is located near the engine compartment, Ref2 is located inside the left front wheel arch, Ref3 is located in the middle of the chassis, and Ref4 is located inside the A-pillar. The current main noise type is low-frequency harmonic noise from the engine. Correlation calculations are performed for the typical frequency band of engine noise (e.g., 50-200Hz). The calculation results are: Ref1 has the highest correlation coefficient (0.95), Ref2 has a slightly lower correlation coefficient (0.85), Ref3 has the weakest correlation (0.60), and Ref4 has the lowest correlation (0.30). Based on the principle that "the higher the correlation coefficient, the higher the weight," weights are assigned to each reference microphone, which can be set to 0.35, 0.31, 0.22, and 0.12. The time-series data of the reference microphones are weighted using these weights to obtain the original reference signals of the reference microphones within the target noise reduction channel.

[0050] The above steps achieve adaptive optimization of the reference signal through correlation weight allocation guided by noise type. It can automatically filter and highlight the reference microphone data most closely related to the main noise, and suppress the interference of unrelated or weakly correlated channels, thereby providing a highly pure and target-clear original reference signal for subsequent noise reduction algorithms, significantly improving the accuracy and efficiency of active noise reduction.

[0051] Step 104: Perform secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the speaker in the noise reduction channel.

[0052] This application uses secondary channel modeling-linear convolution modeling to perform secondary path estimation on the original reference signal to obtain the reference estimated signal. The calculation formula (11) is as follows: (11) in, It is the secondary path impulse response (reflecting the transmission characteristics from the speaker to the error signal microphone within the target noise reduction channel; the error microphone is a microphone within the target noise reduction channel used to monitor the final noise reduction effect). It is the reference estimation signal after secondary path filtering. This is the original reference signal from the reference microphone within the target noise reduction channel. (Through...) Convolution compensation is applied to compensate for the transmission effect from the loudspeaker to the error microphone, ensuring the stability and convergence of the adaptive filter.

[0053] Then, the forward path is calculated on the reference estimated signal using a preset adaptive filter to obtain the first noise reduction driving signal for the speaker in the noise reduction channel. Because the forward path calculation is the process of weighted summation of the reference estimated signal at the current time and the reference estimated signal at a historical time with the weights of the preset adaptive filter, when performing the forward path calculation on the reference estimated signal at the current time, it is necessary to obtain the reference estimated signals at the current time (time n) and the historical time (which corresponds to the reference estimated signals k sampling points prior to the current time n), and then calculate the first noise reduction driving signal at time n. Therefore, the calculation formula (12) is: (12) in, It is the reference estimation signal after secondary path filtering, at time n The value at time k, Here, L1 represents the weights of the preset adaptive filter, n represents the current time, k is the tap index of the preset filter, corresponding to a time delay, nk refers to the time step k sampling points backward from time n, often called "time nk", and L1 is the length of the preset adaptive filter, i.e., the filter order. It is the first noise reduction drive signal output by the speaker in the target noise reduction channel at time n.

[0054] It should be noted that the speaker receives This electrical signal will be converted into a corresponding reverse acoustic wave signal. At this time, the first reverse acoustic wave carried by the reverse acoustic wave signal will propagate in the air and interfere with and cancel out the original ambient noise sound wave in the target area.

[0055] This application can adjust the weights of a preset adaptive filter based on the first noise reduction driving signal. During adjustment, it is necessary to first determine the noise reduction effect of the first inverse acoustic wave generated by the current first noise reduction driving signal, and then adjust based on the noise reduction effect. The noise reduction effect is determined based on the current residual noise signal received by the error microphone. The formula (13) for obtaining the current residual noise signal is: (13) in, It is the first noise reduction drive signal of the speaker in the target noise reduction channel at time n. It is the instantaneous value of the original primary noise signal that the error microphone will receive at time n, assuming the speaker is completely inactive (this value cannot be measured). It is the current residual noise signal received by the error microphone. Equation (13) describes a physical superposition process, when When the value is large, the adjustment is made by adjusting the weights of the preset adaptive filter. ,when The value is small enough that the adjustment is now complete. A threshold can be set when... If the value is greater than the threshold, it indicates that the noise reduction effect on environmental noise is not good. In this case, it is determined that the weight needs to be adjusted. The specific threshold value is not specifically limited in this application.

[0056] Among them, when adjusting the weights, the weight adjustment formula (14) for the preset adaptive filter is: (14) in, It is the i-th coefficient value of the preset adaptive filter at time n, where i is the tap index of the preset adaptive filter, corresponding to a time delay, and ni refers to the i sampling points backward from time n, often called "time ni". It is the step size coefficient. It is the reference estimation signal after secondary path filtering, at time n The value at time i, It is the i-th coefficient value of the preset adaptive filter at time n+1. It is the current residual noise signal received by the error microphone.

[0057] Here, because the adaptive filter is designed so that the i-th coefficient specifically processes the signal delayed by i sampling points, 'i' in the formula is both an index and a time point. The adaptive filter is dynamically adjusted using the reference signal to generate inverse sound waves for active noise cancellation; combined with the spatial activation matrix, it effectively utilizes the speaker and microphone signals from different spatial partitions, forming spatial adaptive noise reduction. The provided spatial information enables dynamic adjustment of the reference signal and partition activation.

[0058] Based on the above, the steps for adjusting the weights of the preset adaptive filter according to the first noise reduction driving signal include: The first noise reduction drive signal drives the speaker in the noise reduction channel to generate a first reverse sound wave. The current residual noise signal received by the error microphone in the target noise reduction channel is identified. The current residual noise signal is generated based on the linear superposition of ambient noise and the first reverse sound wave in the sound field. The weights of the preset adaptive filter are adjusted based on the current residual noise signal.

[0059] The above steps capture the residual result of the superposition of the original noise and the reverse sound wave in real time through the error microphone, and dynamically adjust the weight of the filter based on this. This allows the generated reverse sound wave to continuously track and accurately match changes in environmental noise, thereby achieving efficient and adaptive active cancellation of non-steady-state noise and significantly improving the accuracy and environmental adaptability of the noise reduction effect.

[0060] It should be noted that the above The calculation process is based on the traditional Filtered-x Least Mean Square (FxLMS) path algorithm.

[0061] Step 105: Obtain the historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel, and input the historical reference signal and the training noise signal into the preset depth active noise control network to output the second noise reduction driving signal.

[0062] This application also establishes a preset deep active noise control network. This network can output a second noise reduction driving signal for the loudspeaker in the target noise reduction channel based on the historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel. The historical reference signal of the reference microphone is obtained by processing the time-series signal of the environmental noise collected by the reference microphone in the target noise reduction channel at several historical moments over a period of time (including several historical moments). The processing method is referred to formulas (1) to (4) and formula (10). The training noise signal of the error microphone is pure noise recorded by the error microphone in the target noise reduction channel in a controlled and ideal experimental environment during the training phase of the preset deep active noise control network, after only the main noise source is turned on and all other sound sources are turned off. At this time, the training noise signal is not contaminated by any anti-phase sound waves. The backbone architecture of the preset deep active noise control network can adopt at least one of the following: Temporal Convolutional Network (TNC), Bidirectional Long Short-Term Memory (BiLSTM), or Transformer Attention Network. After training, the second noise reduction driving signal can be output through the pre-trained active noise control network with preset depth.

[0063]

[0064] in, It is the training noise signal of length N from time n-N+1 to time n of the error microphone. It is a historical reference signal of length N from time n-N+1 to time n, based on the reference microphone. It is the second noise reduction driving signal at time n. () is the preset depth active noise control network after training.

[0065] Step 106: Adaptively weight the first noise reduction drive signal and the second noise reduction drive signal to obtain the target noise reduction drive signal, and complete the elimination of vehicle environmental noise through the target noise reduction drive signal.

[0066] After obtaining the first noise reduction driving signal and the second noise reduction driving signal, this application adaptively weights them and outputs the target noise reduction driving signal for the loudspeaker in the target noise reduction channel. Formula (15) is as follows: (15) in, It is an adaptive weighted weight. It is the target noise reduction driving signal at time n. It is the second noise reduction driving signal at time n. This is the first noise reduction drive signal for the loudspeaker in the target noise reduction channel at time n. =1 indicates that the acquisition of the target noise reduction driving signal completely relies on the traditional FxLMS path algorithm. =0 indicates that the acquisition of the target noise reduction driving signal is completely trusted by the preset depth active noise control network. ∈ (0, 1) adopts a mixed mode, where the contributions of the two are weighted and merged. Adjustments can be made based on feedback and performance indicators. During adjustment, the noise reduction effect of the second inverse sound wave generated by the speaker in the target noise reduction channel based on the target noise reduction drive signal is first determined, and adjustments are made based on this effect. When determining the noise reduction effect, the actual residual noise signal received by the error microphone is first acquired, and then the power of the actual residual noise signal is calculated. If the power is large, exceeding the power threshold, it indicates that the noise reduction capability of the traditional FxLMS path is insufficient to cope with the current noise environment. The noise may become non-stationary, complex, or undergo abrupt changes; in this case, it is necessary to reduce... This allows a pre-defined deep active noise control network to utilize its predictive capabilities and attempt to capture new noise patterns. The steps involved include: The target noise reduction drive signal drives the speaker in the noise reduction channel to generate a second reverse sound wave; The actual residual noise signal received by the error microphone in the target noise reduction channel is identified. The actual residual noise signal is generated based on the linear superposition of environmental noise and the second reverse sound wave in the sound field. Calculate the power of the actual residual noise signal; If the power of the actual residual noise signal is greater than the power threshold, the weights for adaptive weighting are adjusted.

[0067] The above steps monitor the residual noise energy after noise reduction in real time and automatically trigger weight adjustment only when the noise reduction effect is not up to standard. This avoids the waste of resources caused by continuous computation and can dynamically correct the sound wave cancellation deviation, ensuring that the noise reduction effect is always stable within the expected threshold, thus achieving a balance between high efficiency, energy saving and robustness.

[0068] It should be understood that, in addition to adjusting the weights of the adaptive weighting, the step size coefficients of the weights of the preset adaptive filter can also be adjusted.

[0069] The noise cancellation method provided in this application collects time-series data of environmental noise in a vehicle using several microphones; analyzes the time-series data to determine the dominant noise type and spatial distribution of noise intensity; identifying the dominant noise type facilitates the separation of multi-source composite noise, providing a basis for subsequently increasing the weight of microphones that require special attention, thus achieving precise removal of the dominant noise; determining the spatial distribution of noise intensity provides a basis for subsequent zone activation; based on the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated; based on the dominant noise type, the original reference signal of the reference microphone in the target noise reduction channel is determined; secondary path estimation and forward path calculation of the preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the speaker in the noise reduction channel, realizing real-time tracking and cancellation of environmental noise, ensuring timely response; and simultaneously acquiring the historical reference signal of the reference microphone and the error signal in the target noise reduction channel. The training noise signal from the microphone is compared with the historical reference signal and the training noise signal, and then input into a preset deep active noise control network to output a second noise reduction driving signal. The preset deep active noise control network can learn and predict difficult-to-handle nonlinear and non-stationary noise. By analyzing historical signals, it captures the changing trend of noise and realizes feedforward predictive noise reduction to quickly respond to high-frequency, sudden or rapidly changing noise and reduce convergence delay. Finally, the first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal. The target noise reduction driving signal is used to eliminate the vehicle's environmental noise. At this time, the adaptive weighting can dynamically adjust the contribution weight of the two paths according to the real-time noise scene, ensuring that the "current optimal" control signal can be output under any operating condition, thereby comprehensively improving the efficiency, fault tolerance and robustness of the noise reduction process and improving the user's driving experience.

[0070] Reference Figure 3 The diagram shows a structural schematic of a noise cancellation device provided in this application, the device comprising: The first acquisition module 201 is used to acquire time-series data of environmental noise in the vehicle through several microphones.

[0071] The first determining module 202 is used to analyze time-series data to determine the main noise type and spatial distribution of noise intensity in the environment.

[0072] The second determining module 203 is used to activate the target noise reduction channel of the target partition in the vehicle by means of the spatial distribution of noise intensity, and to determine the original reference signal of the reference microphone in the target noise reduction channel by means of the main noise type.

[0073] The calculation module 204 is used to perform secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel.

[0074] The input / output module 205 is used to acquire the historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel, and input the historical reference signal and the training noise signal into a preset depth active noise control network, and output a second noise reduction driving signal.

[0075] The weighting module 206 is used to adaptively weight the first noise reduction driving signal and the second noise reduction driving signal to obtain the target noise reduction driving signal, and to eliminate the environmental noise of the vehicle through the target noise reduction driving signal.

[0076] Optionally, the noise cancellation device may also include: The filtering and transformation submodule is used to perform bandpass filtering and wavelet transform on time-series data to obtain wavelet coefficients.

[0077] The identification submodule is used to identify the main noise type of the environment through time-series data.

[0078] The first screening submodule is used to process the wavelet coefficients using a soft thresholding algorithm if the main noise type is a low-frequency stable main noise type, so as to obtain the first screening wavelet coefficients.

[0079] The second filtering submodule is used to process the wavelet coefficients using a hard thresholding algorithm if the main noise type is a high-frequency main noise type or a burst main noise type, so as to obtain the second filtering wavelet coefficients.

[0080] The inverse wavelet transform submodule is used to perform inverse wavelet transform on the first or second selected wavelet coefficients to obtain denoised time series data.

[0081] The normalization and gain submodule is used to normalize the denoised time series data and perform adaptive gain control to obtain preprocessed time series data.

[0082] Optionally, the first determining module 202 specifically includes: The segmentation submodule is used to segment the preprocessed time-series data into continuous time-domain frame data.

[0083] The Fourier transform submodule is used to perform discrete Fourier transform on time-domain frame data to obtain the frequency domain complex spectrum corresponding to the time-domain frame data.

[0084] The first determination submodule is used to determine the Mel energy spectrum of the ambient noise, the time difference between the arrival of the ambient noise at any two microphones, and the spatial spectral power between the ambient noise and each microphone through the frequency domain complex spectrum.

[0085] The second determination submodule is used to determine the Mel frequency cepstral coefficients of environmental noise using Mel energy spectrum.

[0086] The third determination submodule is used to determine the main noise type and spatial distribution of noise intensity in the environment through Mel energy spectrum, Mel frequency cepstral coefficients, time difference and spatial spectral power.

[0087] Optionally, the second determining module 203 specifically includes: The fourth determination submodule is used to determine the noise intensity of several zones of the vehicle by analyzing the spatial distribution of noise intensity.

[0088] The activation submodule is used to filter out target partitions with noise intensity greater than the intensity threshold and activate the target noise reduction channel of the target partition in the vehicle.

[0089] The correlation analysis submodule is used to perform correlation analysis on several reference microphones in the target noise reduction channel based on the main noise type, and obtain the correlation coefficient.

[0090] The weight setting submodule is used to set weights for each reference microphone in the target partition based on the correlation coefficient. The larger the correlation coefficient, the greater the weight.

[0091] The weighting submodule is used to weight the timing data acquired by the reference microphone to obtain the original reference signal of the reference microphone in the target noise reduction channel.

[0092] Optionally, the noise cancellation device may also include: The first sound wave generation module is used to drive the speaker in the noise reduction channel to generate a first reverse sound wave through the first noise reduction drive signal.

[0093] The first identification module is used to identify the current residual noise signal received by the error microphone in the target noise reduction channel. The current residual noise signal is generated by the linear superposition of environmental noise and the first reverse sound wave in the sound field.

[0094] The first weight adjustment module is used to adjust the weights of the preset adaptive filter based on the current residual noise signal.

[0095] The second sound wave generation module is used to drive the speaker in the noise reduction channel to generate a second reverse sound wave through the target noise reduction drive signal.

[0096] The second identification module is used to identify the actual residual noise signal received by the error microphone in the target noise reduction channel. The actual residual noise signal is generated by the linear superposition of environmental noise and the second reverse sound wave in the sound field.

[0097] The power calculation module is used to calculate the power of the actual residual noise signal.

[0098] The second weight adjustment module is used to adjust the weights for adaptive weighting if the power of the actual residual noise signal is greater than the power threshold.

[0099] The noise cancellation method provided in this application collects time-series data of environmental noise in a vehicle using several microphones; analyzes the time-series data to determine the dominant noise type and spatial distribution of noise intensity; identifying the dominant noise type facilitates the separation of multi-source composite noise, providing a basis for subsequently increasing the weight of microphones that require special attention, thus achieving precise removal of the dominant noise; determining the spatial distribution of noise intensity provides a basis for subsequent zone activation; based on the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated; based on the dominant noise type, the original reference signal of the reference microphone in the target noise reduction channel is determined; secondary path estimation and forward path calculation of the preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the speaker in the noise reduction channel, realizing real-time tracking and cancellation of environmental noise, ensuring timely response; and simultaneously acquiring the historical reference signal of the reference microphone and the error signal in the target noise reduction channel. The training noise signal from the microphone is compared with the historical reference signal and the training noise signal, and then input into a preset deep active noise control network to output a second noise reduction driving signal. The preset deep active noise control network can learn and predict difficult-to-handle nonlinear and non-stationary noise. By analyzing historical signals, it captures the changing trend of noise and realizes feedforward predictive noise reduction to quickly respond to high-frequency, sudden or rapidly changing noise and reduce convergence delay. Finally, the first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal. The target noise reduction driving signal is used to eliminate the vehicle's environmental noise. At this time, the adaptive weighting can dynamically adjust the contribution weight of the two paths according to the real-time noise scene, ensuring that the "current optimal" control signal can be output under any operating condition, thereby comprehensively improving the efficiency, fault tolerance and robustness of the noise reduction process and improving the user's driving experience.

[0100] Reference Figure 4 This application also provides an electronic device, such as Figure 4 As shown, it includes a processor 301, a communication interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304. Processor 301, memory 303 for storing processor-executable instructions; The processor 301 is configured to execute the instructions to implement the noise cancellation method described above: Time-series data of ambient noise in the vehicle were collected using several microphones; Analyze the time-series data to determine the dominant noise type and spatial distribution of noise intensity in the environment; By using the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated, and the original reference signal of the reference microphone in the target noise reduction channel is determined by the main noise type. Secondary path estimation and forward path calculation of a preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel; The historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel are obtained, and the historical reference signal and the training noise signal are input into a preset depth active noise control network to output a second noise reduction driving signal. The first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal, and the vehicle environmental noise is eliminated by the target noise reduction driving signal.

[0101] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0102] The communication interface is used for communication between the aforementioned terminal and other devices.

[0103] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0104] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be 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, or discrete hardware components.

[0105] In another embodiment provided in this application, a vehicle is also provided, which may specifically include the above-described noise cancellation device.

[0106] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0108] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0109] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A noise cancellation method, characterized in that, The method includes: Time-series data of ambient noise in the vehicle were collected using several microphones; Analyze the time-series data to determine the dominant noise type and spatial distribution of noise intensity in the environment; By using the spatial distribution of noise intensity, the target noise reduction channel of the target zone in the vehicle is activated, and the original reference signal of the reference microphone in the target noise reduction channel is determined by the main noise type. Secondary path estimation and forward path calculation of a preset adaptive filter are performed on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel; The historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel are obtained, and the historical reference signal and the training noise signal are input into a preset depth active noise control network to output a second noise reduction driving signal. The first noise reduction driving signal and the second noise reduction driving signal are adaptively weighted to obtain the target noise reduction driving signal, and the vehicle environmental noise is eliminated by the target noise reduction driving signal.

2. The method according to claim 1, characterized in that, After acquiring time-series data of ambient noise in the vehicle via several microphones, the process includes: The time-series data are subjected to bandpass filtering and wavelet transform to obtain wavelet coefficients; The dominant noise type of the environmental noise is identified using the time-series data; If the main noise type is a low-frequency stable main noise type, then the wavelet coefficients are processed using a soft thresholding algorithm to obtain the first selected wavelet coefficients. If the main noise type is a high-frequency main noise type or a burst main noise type, then the wavelet coefficients are processed using a hard threshold algorithm to obtain the second selected wavelet coefficients. Perform inverse wavelet transform on the first or second selected wavelet coefficients to obtain denoised time series data; The denoised time series data is normalized and subjected to adaptive gain control to obtain preprocessed time series data.

3. The method according to claim 2, characterized in that, The analysis of the time-series data to determine the dominant noise type and spatial distribution of noise intensity in the environment includes: The preprocessed time-series data is divided into continuous time-domain frame data; Perform a discrete Fourier transform on the time-domain frame data to obtain the frequency domain complex spectrum corresponding to the time-domain frame data; The Mel energy spectrum of the ambient noise, the time difference between the arrival of the ambient noise at any two microphones, and the spatial spectral power between the ambient noise and each microphone are determined by the frequency domain complex spectrum. The Mel frequency cepstral coefficients of the environmental noise are determined using the Mel energy spectrum. The dominant noise type and spatial distribution of noise intensity of the environmental noise are determined by the Mel energy spectrum, the Mel frequency cepstral coefficients, the time difference, and the spatial spectral power.

4. The method according to claim 1, characterized in that, The step of activating the target noise reduction channel of the target zone in the vehicle through the spatial distribution of noise intensity further includes: The noise intensity of several zones of the vehicle is determined by the spatial distribution of the noise intensity. Target zones with noise intensity greater than the intensity threshold are selected, and the target noise reduction channel of the target zone in the vehicle is activated.

5. The method according to claim 1, characterized in that, Determining the original reference signal of the reference microphone within the target noise reduction channel based on the main noise type includes: Correlation analysis was performed on several reference microphones in the target noise reduction channel based on the main noise type to obtain the correlation coefficient; The correlation coefficient is used to assign a weight to each reference microphone in the target partition, wherein the larger the correlation coefficient, the larger the weight is assigned. The timing data acquired by the reference microphone is weighted by weighting to obtain the original reference signal of the reference microphone in the target noise reduction channel.

6. The method according to claim 1, characterized in that, After performing secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the speaker in the target noise reduction channel, the method further includes: The first noise reduction driving signal drives the loudspeaker in the noise reduction channel to generate a first reverse sound wave. The current residual noise signal received by the error microphone in the target noise reduction channel is identified. The current residual noise signal is generated based on the linear superposition of ambient noise and the first reverse sound wave in the sound field. The weights of the preset adaptive filter are adjusted based on the current residual noise signal.

7. The method according to claim 1, characterized in that, After adaptively weighting the first noise reduction driving signal and the second noise reduction driving signal to obtain the target noise reduction driving signal, the method further includes: The target noise reduction driving signal drives the speaker in the noise reduction channel to generate a second reverse sound wave. The actual residual noise signal received by the error microphone in the target noise reduction channel is identified. The actual residual noise signal is generated based on the linear superposition of environmental noise and the second reverse sound wave in the sound field. Calculate the power of the actual residual noise signal; If the power of the actual residual noise signal is greater than the power threshold, the weights for adaptive weighting are adjusted.

8. A noise cancellation device, characterized in that, The device includes: The first acquisition module is used to acquire time-series data of environmental noise in the vehicle through several microphones; The first determining module is used to analyze the time-series data to determine the main noise type and spatial distribution of noise intensity of the environmental noise. The second determining module is used to activate the target noise reduction channel of the target partition in the vehicle through the spatial distribution of noise intensity, and to determine the original reference signal of the reference microphone in the target noise reduction channel through the main noise type. The calculation module is used to perform secondary path estimation and forward path calculation of the preset adaptive filter on the original reference signal to obtain the first noise reduction drive signal for the loudspeaker in the target noise reduction channel. The input / output module is used to acquire the historical reference signal of the reference microphone and the training noise signal of the error microphone in the target noise reduction channel, and input the historical reference signal and the training noise signal into a preset deep active noise control network, and output a second noise reduction driving signal. The weighting module is used to adaptively weight the first noise reduction driving signal and the second noise reduction driving signal to obtain the target noise reduction driving signal, and to eliminate the vehicle's environmental noise through the target noise reduction driving signal.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the instructions to implement the noise cancellation method as described in any one of claims 1 to 7.

10. A vehicle, characterized in that, include: The noise cancellation device according to claim 8.

Images