Automobile noise cancellation method and device, automobile, electronic equipment and storage medium

Abstract

The application provides a car noise cancellation method and device, a car, electronic equipment and a storage medium, and relates to the technical field of noise cancellation. The method comprises the following steps: determining an input signal based on a vibration signal of a target car and an operating state of the target car; inputting the input signal into a feature extraction layer in a noise cancellation model to obtain a feature tensor output by the feature extraction layer; inputting the feature tensor into a signal generation layer in the noise cancellation model to obtain a noise reduction signal output by the signal generation layer; and outputting the noise reduction signal; wherein the feature extraction layer comprises a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting a nonlinear feature tensor of the input signal; the noise cancellation model is obtained by training based on a sample input signal and a label signal corresponding to the sample input signal, and the sample input signal is determined based on a sample vibration signal and a sample operating state. The application can effectively eliminate nonlinear noise and improve the noise reduction effect.

Description

Technical Field

[0001] This invention relates to the field of noise cancellation technology, and more particularly to a method, apparatus, automobile, electronic device, and storage medium for eliminating automobile noise. Background Technology

[0002] With the rapid development of technology, people have increasingly higher demands for the riding experience in cars. Road noise is a significant factor affecting this experience. Road noise is caused by random broadband vibrations generated by the car tires at the road contact point, which propagate through the structure and into the passenger compartment. Structural propagation refers to the transmission of vibrations through the suspension system to the car body, and then the radiating of sound into the passenger compartment. It is a major source of interior noise below 500 Hz, therefore, it is necessary to eliminate road noise propagated through the structure.

[0003] Currently, active noise control technology can reduce road noise propagating through the structure in automobiles. However, most current active noise control technologies use linear adaptive filtering for noise reduction. Since the hydraulic shock absorbers and rubber bushings in most automobile suspension systems have nonlinear acoustic characteristics, current active noise control technologies cannot reduce the nonlinear noise, resulting in poor noise reduction effects. Summary of the Invention

[0004] This invention provides a method, apparatus, vehicle, electronic device, and storage medium for eliminating automotive noise, thereby overcoming the shortcomings of existing technologies in eliminating nonlinear noise and achieving superior noise reduction effects.

[0005] This invention provides a method for eliminating automobile noise, comprising:

[0006] The input signal is determined based on the vibration signal of the target vehicle and the operating status of the target vehicle;

[0007] The input signal is fed into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer.

[0008] The feature tensor is input into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer;

[0009] Output the noise reduction signal;

[0010] The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal; the noise cancellation model is trained based on the sample input signal and the label signal corresponding to the sample input signal, and the sample input signal is determined based on the sample vibration signal and the sample running state.

[0011] According to a method for eliminating automotive noise provided by the present invention, the step of inputting the feature tensor into a signal generation layer in the noise elimination model to obtain a denoised signal output by the signal generation layer includes:

[0012] The feature tensor is input to the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer. The deconvolution mapping layer includes a deconvolution layer and a nonlinear activation function layer.

[0013] According to a method for eliminating vehicle noise provided by the present invention, the step of inputting the feature tensor into a deconvolution mapping layer in the signal generation layer to obtain a denoised signal output by the deconvolution mapping layer includes:

[0014] The feature tensor is input into the causal convolutional layer in the signal generation layer to obtain the first target feature tensor output by the causal convolutional layer.

[0015] The first target feature tensor and the feature tensor are input into the first feature fusion layer in the signal generation layer to obtain the second target feature tensor output by the first feature fusion layer.

[0016] The second target feature tensor is input into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer.

[0017] According to a vehicle noise cancellation method provided by the present invention, the deconvolution mapping layer further includes a nonlinear threshold layer, which is constructed based on the following nonlinear threshold function:

[0018] f(x) = kx, x > 0;

[0019] f(x) = m(exp(x) - n), x ≤ 0;

[0020] In the formula, x represents the input of the nonlinear threshold layer, f(x) represents the output of the nonlinear threshold layer, k represents a preset first constant, m represents a preset second constant, n represents a preset third constant, and exp() represents an exponential function with the natural constant e as the base.

[0021] According to a vehicle noise cancellation method provided by the present invention, the step of inputting the input signal to a feature extraction layer in a noise cancellation model to obtain a feature tensor output by the feature extraction layer includes:

[0022] The input signal is input to the nonlinear feature extraction layer in the feature extraction layer to obtain the nonlinear feature tensor output by the nonlinear feature extraction layer, wherein the nonlinear feature extraction layer includes the nonlinear activation function layer;

[0023] The input signal is input to the linear feature extraction layer in the feature extraction layer to obtain the linear feature tensor output by the linear feature extraction layer;

[0024] The nonlinear feature tensor and the linear feature tensor are input into the second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.

[0025] According to a vehicle noise cancellation method provided by the present invention, the nonlinear activation function layer is constructed based on the following nonlinear activation function:

[0026] f(x) = sin(x) + cos(x);

[0027] In the formula, x represents the input of the nonlinear activation function layer, and f(x) represents the output of the nonlinear activation function layer.

[0028] According to the present invention, a vehicle noise cancellation method is provided, wherein the tag signal is a noise signal collected based on a microphone inside the vehicle;

[0029] The noise cancellation model is trained in the following manner:

[0030] The sample input signal is input to the model to be trained to obtain the sample denoising signal output by the model to be trained;

[0031] The sample denoised signal is convolved with a preset transmission path to obtain the predicted denoised signal that is transmitted from the sample denoised signal to the microphone.

[0032] Based on the sum of the predicted denoised signal and the label signal, the loss value corresponding to the loss function is determined;

[0033] The model to be trained is trained based on the loss value.

[0034] According to a vehicle noise cancellation method provided by the present invention, determining the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle includes:

[0035] The vibration signal is converted into a first complex spectrum signal, and the operating state is converted into a second complex spectrum signal;

[0036] The input signal is obtained by fusing the first complex spectrum signal and the second complex spectrum signal.

[0037] The present invention also provides an automotive noise cancellation device, comprising:

[0038] The signal determination module is used to determine the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle;

[0039] The feature extraction module is used to input the input signal into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer;

[0040] The signal generation module is used to input the feature tensor into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer.

[0041] The signal output module is used to output the noise reduction signal;

[0042] The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal; the noise cancellation model is trained based on the sample input signal and the label signal corresponding to the sample input signal, and the sample input signal is determined based on the sample vibration signal and the sample running state.

[0043] The present invention also provides an automobile, comprising:

[0044] A vibration sensor, used to acquire vibration signals from the vehicle;

[0045] A loudspeaker, the loudspeaker being used to output a noise-reducing signal;

[0046] A processor for performing any of the above-described automotive noise cancellation methods.

[0047] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the automotive noise cancellation method as described above.

[0048] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vehicle noise cancellation method as described above.

[0049] The present invention provides a vehicle noise cancellation method, apparatus, vehicle, electronic device, and storage medium. Based on the vibration signal and operating state of the target vehicle, the input signal is determined, thus considering not only the vibration signal but also the vehicle's operating state. The input signal is input to the feature extraction layer in the noise cancellation model to obtain the feature tensor output by the feature extraction layer. This extracts features not only from the vibration signal but also from the operating state. The feature tensor is input to the signal generation layer in the noise cancellation model to obtain a more accurate noise reduction signal, thereby improving the noise reduction effect. Simultaneously, the feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal, i.e., it can extract the features of nonlinear noise, thereby providing nonlinear modeling capabilities and effectively eliminating nonlinear noise, ultimately improving the noise reduction effect. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 A schematic flowchart of the vehicle noise elimination method provided by the present invention;

[0052] Figure 2 One of the layout schematic diagrams of the automotive components provided by the present invention;

[0053] Figure 3 The second schematic diagram of the layout of the automotive components provided by the present invention;

[0054] Figure 4 One of the structural schematic diagrams of the noise cancellation model provided by the present invention;

[0055] Figure 5 This is the second schematic diagram of the structure of the noise cancellation model provided by the present invention;

[0056] Figure 6 A schematic diagram of the structure of the automotive noise cancellation device provided by the present invention;

[0057] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0059] With the rapid development of technology, people have increasingly higher demands for the riding experience in cars. Road noise is a significant factor affecting this experience. Road noise is caused by random broadband vibrations generated by the car tires at the road contact point, which propagate through the structure and into the passenger compartment. Structural propagation refers to the transmission of vibrations through the suspension system to the car body, and then the radiating of sound into the passenger compartment. It is a major source of interior noise below 500 Hz, therefore, it is necessary to eliminate road noise propagated through the structure.

[0060] Considering the low-frequency characteristics of road noise propagating through structures, it is difficult to control using passive noise control technologies commonly used in automobiles, such as viscoelastic damping treatments applied to porous materials in the body panels or passenger compartment. These methods are ineffective at low frequencies because the wavelengths of sound and vibration are similar to the material thickness. Therefore, currently, most efforts to reduce road noise propagating through structures in automobiles rely on active noise control technologies, which are more effective at low frequencies. However, most current active noise control technologies employ linear adaptive filtering for noise reduction. Since the hydraulic shock absorbers and rubber bushings in most automotive suspension systems have nonlinear acoustic characteristics, current active noise control technologies cannot reduce the nonlinear noise components, resulting in poor noise reduction performance and limited noise reduction potential.

[0061] To address the above problems, the present invention proposes the following embodiments. Figure 1 This is a schematic flowchart of the vehicle noise cancellation method provided by the present invention, as shown below. Figure 1 As shown, the vehicle noise reduction method includes:

[0062] Step 110: Determine the input signal based on the vibration signal of the target vehicle and the operating status of the target vehicle.

[0063] Here, the target vehicle is the vehicle for which noise reduction is to be performed. The vibration signal can be acquired by a vibration sensor located on the target vehicle.

[0064] In one embodiment, the vibration sensor is an accelerometer to acquire acceleration information of the target vehicle and then generate a vibration signal based on that acceleration information. Further, the vibration sensor is a triaxial accelerometer.

[0065] In one embodiment, such as Figure 2 As shown, the vibration sensor includes four accelerometers, which are respectively located at the suspension system of each tire on the bottom of the target vehicle.

[0066] Here, the operating state may include, but is not limited to, at least one of the following: the target vehicle's speed, the target vehicle's acceleration, and the target vehicle's steering, etc. This operating state can be detected by the vehicle's inherent components, which will not be elaborated on here.

[0067] Here, the input signal is used to characterize the vibration information and operating status information of the target vehicle. In other words, the vibration signal and the operating status of the target vehicle are fused to obtain the input signal. The specific fusion method is not limited here, as long as it is ensured that the input signal can simultaneously characterize the information of both.

[0068] In one embodiment, the vibration signal of the target vehicle is converted into a first complex spectrum signal, and the operating state of the target vehicle is converted into a second complex spectrum signal. The first and second complex spectrum signals are then fused to obtain an input signal, which is also a complex spectrum signal. Based on the above, using the complex spectrum signal as the input to the noise cancellation model results in the output of the noise cancellation model also being a complex spectrum signal. Correspondingly, the output denoised signal needs to be converted into a time-domain signal. Using the complex spectrum signal as the model input can improve the noise cancellation model's feature extraction capability from the input signal, thereby improving the noise reduction effect of the noise cancellation model.

[0069] In another embodiment, the vibration signal of the target vehicle and its operating state are fused to obtain input data; the input data is then converted into a complex spectrum signal (input signal). For example, a Fourier transform is performed on the input data to obtain the input signal; for instance, the sampling rate is 16kHz, and the Fourier transform length is 320 points. Based on the above, using the complex spectrum signal as the input to the noise cancellation model results in the output of the noise cancellation model also being a complex spectrum signal; correspondingly, the output denoised signal needs to be converted into a time-domain signal. Using the complex spectrum signal as the model input can improve the noise cancellation model's feature extraction capability from the input signal, thereby improving the noise reduction effect of the noise cancellation model.

[0070] Step 120: Input the input signal into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer.

[0071] The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal.

[0072] Here, the noise cancellation model is used to generate a corresponding denoised signal based on the input signal, and its feature extraction layer is used to extract features from the input signal. It should be understood that the feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal, i.e., it can extract features of nonlinear noise, thereby providing nonlinear modeling capabilities, effectively eliminating nonlinear noise, and ultimately improving the denoising effect.

[0073] In some embodiments, the input signal is input to a nonlinear feature extraction layer in the feature extraction layer to obtain a feature tensor output by the nonlinear feature extraction layer. The nonlinear feature extraction layer includes a nonlinear activation function layer, and the feature tensor is a nonlinear feature tensor.

[0074] In one embodiment, the nonlinear feature extraction layer further includes a linear layer (fully connected layer). For example, the nonlinear feature extraction layer includes a linear layer, a linear layer and a nonlinear activation function layer connected in sequence, that is, the nonlinear feature extraction layer includes two linear layers and a nonlinear activation function layer connected in sequence. Of course, the number of linear layers is not specifically limited.

[0075] In one embodiment, if the input signal is a complex spectrum signal, the linear layer included in the nonlinear feature extraction layer is a two-dimensional linear layer, so that the output of the two-dimensional linear layer is a feature tensor, rather than a one-dimensional feature vector.

[0076] Step 130: Input the feature tensor into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer.

[0077] Here, the signal generation layer generates the corresponding denoised signal based on the feature tensors extracted by the feature extraction layer. It should be understood that this signal generation layer can be configured according to actual needs.

[0078] In some embodiments, a feature tensor is input to a deconvolution mapping layer in the signal generation layer to obtain a denoised signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer includes a deconvolution layer.

[0079] In some embodiments, a feature tensor is input to a causal convolutional layer in the signal generation layer to obtain a first target feature tensor output by the causal convolutional layer; the first target feature tensor is input to a deconvolutional mapping layer in the signal generation layer to obtain a denoised signal output by the deconvolutional mapping layer, wherein the deconvolutional mapping layer includes a deconvolutional layer. Further, the first target feature tensor and a feature tensor are input to a first feature fusion layer in the signal generation layer to obtain a second target feature tensor output by the first feature fusion layer; the second target feature tensor is input to a deconvolutional mapping layer in the signal generation layer to obtain a denoised signal output by the deconvolutional mapping layer.

[0080] The deconvolutional layer is used to deconvolve the input. The causal convolutional layer can cache the input from previous time steps, thereby modeling the input in time, which can solve the problem of time series inference and thus improve the accuracy of the generated noise-reduced signal.

[0081] Furthermore, the deconvolutional mapping layer also includes a normalization layer, such as an instance normalization layer.

[0082] Furthermore, the deconvolutional mapping layer also includes a nonlinear threshold layer.

[0083] In one embodiment, if the input signal is a complex spectrum signal, the deconvolution layer is a two-dimensional deconvolution layer to perform two-dimensional deconvolution on the input.

[0084] In one embodiment, the deconvolutional mapping layer includes multiple mapping layers, each of which includes a deconvolutional layer. Further, the mapping layer also includes a normalization layer, for example, an instance normalization layer.

[0085] Step 140: Output the noise reduction signal.

[0086] Specifically, noise reduction in the vehicle interior is achieved by outputting a noise reduction signal (control signal) through a speaker. This speaker is located inside the target vehicle. For example, such as... Figure 2 As shown, the target car has 6 speakers inside the passenger compartment.

[0087] In one embodiment, if the input signal is a complex spectrum signal, the noise-reduced signal needs to be converted into a time-domain control signal, and then the control signal is output.

[0088] Furthermore, since the control signal is a digital signal, it needs to be converted into an analog signal by a digital-to-analog converter (DAC) before being output by the in-vehicle speaker.

[0089] Furthermore, considering the different specifications and parameters of each speaker, pre-tuning the speakers can ensure effective noise reduction. When tuning speakers, it's necessary to first determine parameters such as the speaker's frequency response characteristics and sensitivity, and then use appropriate acoustic testing instruments to test and calibrate the speaker. During calibration, indicators such as frequency response flatness, phase response, distortion, and sensitivity can be used to evaluate the speaker's performance and make adjustments to achieve optimal results.

[0090] It should be understood that this noise cancellation model can generate corresponding noise reduction signals (control signals) in real time based on the vibration signals and operating status of the target vehicle, thereby achieving real-time in-vehicle noise reduction.

[0091] The noise cancellation model is trained based on the sample input signal and the corresponding label signal, and the sample input signal is determined based on the sample vibration signal and the sample running state.

[0092] Here, the sample vibration signal can be acquired by a vibration sensor installed in the car.

[0093] In one embodiment, such as Figure 3 As shown, the vibration sensor includes four accelerometers, which are respectively located at the suspension system of each tire on the bottom of the car.

[0094] Here, the sample operating state may include, but is not limited to, at least one of the following: vehicle speed, vehicle acceleration, and vehicle steering, etc. This sample operating state can be detected by the vehicle's inherent components, which will not be elaborated here.

[0095] Here, the sample input signal is used to characterize the vehicle's vibration information and operating status information. In other words, the sample vibration signal and the sample operating status are fused to obtain the sample input signal. The specific fusion method is not limited here, as long as it is ensured that the sample input signal can simultaneously characterize the information of both.

[0096] In one embodiment, the sample vibration signal and the sample running state are fused to obtain sample input data; the sample input data is then converted into a complex spectrum signal (sample input signal). For example, a Fourier transform is performed on the sample input data to obtain the sample input signal.

[0097] In one embodiment, the tag signal can be a noise-reduced signal tag corresponding to the labeled sample input signal.

[0098] In another embodiment, the tag signal is a noise signal collected by a microphone inside the vehicle, that is, the noise signal is collected at the same time as the sample vibration signal and the sample running status.

[0099] In some embodiments, to improve noise reduction performance, the data acquisition process for both the training and application (inference) of the noise cancellation model is based on the same target vehicle. That is, the input signal is fed into the noise cancellation model corresponding to the target vehicle to obtain the noise-reduced signal output by the model, and both the sample vibration signal and the sample operating state are collected based on the target vehicle. In other words, the model is pre-trained for the vehicle type requiring noise reduction to obtain the corresponding noise cancellation model.

[0100] In one embodiment, such as Figure 3As shown, four accelerometers (accelerometer sensors), six speakers, and four microphones are installed on the target vehicle. Specifically, the four microphones are located in the driver's seat, passenger seat, rear left side seat, and rear right side seat. Of course, the number and arrangement of these devices can be adjusted according to different vehicle models. The four accelerometers are used to collect vibration signals and sample vibration signals, the six speakers are used to output noise reduction signals, and the four microphones are used to collect tag signals. These four microphones can be removed during application and only installed during training.

[0101] In one embodiment, the aforementioned sample vibration signal, sample running state, and corresponding label signal can be data collected from the target vehicle under different road surfaces and different speeds. This ensures that the training samples cover different vehicle and road conditions, thereby ensuring the richness of the training samples and improving the training effect of the noise cancellation model.

[0102] In one embodiment, after acquiring the aforementioned sample vibration signals, sample operating states, and corresponding label signals, the acquired training data can be cleaned to remove data that does not meet the requirements. For example, data during vehicle idling can be removed. Specifically, by monitoring parameters such as vehicle speed or engine speed, idling segments in the training data can be identified and removed, retaining only data relevant to actual driving conditions. Within the idling segment, any outliers or unreasonable data points also need to be detected and removed. These outliers may be caused by sensor errors or other problems, which can adversely affect model training. It is crucial to ensure that the cleaned dataset has sufficient balance under different conditions to avoid the model becoming overly reliant on or biased towards data under certain conditions.

[0103] In one embodiment, the audio data captured by the microphone can be framed to obtain the tag signal. Framed audio data is used to divide the continuous audio signal into short time intervals, each time interval being called a frame. This step facilitates more granular processing and analysis of the audio data. First, the duration of each frame and the time interval between adjacent frames need to be selected. Common frame lengths are between 20 and 40 milliseconds, while frame shifts are typically between 10 and 20 milliseconds.

[0104] The vehicle noise cancellation method provided in this invention determines the input signal based on the vibration signal and operating state of the target vehicle. This means it considers not only the vibration signal but also the vehicle's operating state. The input signal is then fed into the feature extraction layer of the noise cancellation model to obtain a feature tensor output by the feature extraction layer. This extracts features not only from the vibration signal but also from the operating state. The feature tensor is then input into the signal generation layer of the noise cancellation model to obtain a more accurate noise reduction signal, thereby improving the noise reduction effect. Simultaneously, the feature extraction layer includes a nonlinear activation function layer, which extracts the nonlinear feature tensor of the input signal, i.e., it can extract features of nonlinear noise, thus providing nonlinear modeling capabilities and effectively eliminating nonlinear noise, ultimately improving the noise reduction effect.

[0105] Based on any of the above embodiments, in this method, step 130 includes:

[0106] The feature tensor is input to the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer. The deconvolution mapping layer includes a deconvolution layer and a nonlinear activation function layer.

[0107] Here, deconvolutional layers are used to deconvolve the input. Nonlinear activation function layers are used to extract nonlinear feature tensors.

[0108] In one embodiment, if the input signal is a complex spectrum signal, the deconvolution layer is a two-dimensional deconvolution layer to perform two-dimensional deconvolution on the input.

[0109] In one embodiment, the deconvolution mapping layer further includes a normalization layer, such as an instance normalization layer.

[0110] In one embodiment, the deconvolution mapping layer further includes a nonlinear threshold layer to further extract features of nonlinear noise, thereby further providing nonlinear modeling capabilities and ultimately further improving the noise reduction effect.

[0111] In one embodiment, the deconvolution mapping layer includes multiple mapping layers, each mapping layer including a deconvolution layer and a nonlinear activation function layer. Further, the mapping layer also includes a normalization layer, for example, an instance normalization layer. For instance, any mapping layer includes a deconvolution layer, an instance normalization layer, and a nonlinear activation function layer connected in sequence.

[0112] To facilitate understanding of the above embodiments, a specific embodiment will be described here. For example... Figure 4 As shown, the deconvolutional mapping layer comprises a nonlinear threshold layer, a mapping layer, another nonlinear threshold layer, another mapping layer, and another nonlinear threshold layer connected in sequence. Each mapping layer comprises a deconvolutional layer, an instance normalization layer, and a nonlinear activation function layer connected in sequence. Based on this, accurate denoising signals can be generated.

[0113] The vehicle noise cancellation method provided in this invention inputs a feature tensor into a deconvolution mapping layer in the signal generation layer to obtain a denoised signal output by the deconvolution mapping layer. The deconvolution mapping layer includes a deconvolution layer and a nonlinear activation function layer, which can further extract the features of nonlinear noise, thereby further providing nonlinear modeling capabilities. In turn, it can generate a denoised signal that eliminates nonlinear noise, which can effectively eliminate nonlinear noise and ultimately further improve the noise reduction effect.

[0114] Based on any of the above embodiments, in this method, the step of inputting the feature tensor into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer includes:

[0115] The feature tensor is input into the causal convolutional layer in the signal generation layer to obtain the first target feature tensor output by the causal convolutional layer.

[0116] The first target feature tensor and the feature tensor are input into the first feature fusion layer in the signal generation layer to obtain the second target feature tensor output by the first feature fusion layer.

[0117] The second target feature tensor is input into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer.

[0118] Here, the causal convolutional layer can cache the input from previous time steps, thereby modeling the input in time, i.e., solving the time series inference problem, thus improving the accuracy of the generated denoised signal, and ultimately further improving the denoising effect. For example, the causal convolutional layer includes multiple nonlinear one-dimensional convolutional layers connected in sequence.

[0119] Here, the first feature fusion layer is used to fuse the first target feature tensor with the feature tensor. That is, one branch performs causal convolution and the other branch performs identity mapping, thereby realizing the residual network structure so as to perform deeper feature modeling, thereby improving the robustness of the noise cancellation model and ultimately improving the noise reduction effect.

[0120] The vehicle noise cancellation method provided in this invention inputs a feature tensor into a causal convolutional layer in the signal generation layer to obtain a first target feature tensor output by the causal convolutional layer. This allows for temporal modeling of the input, enabling time series inference and improving the accuracy of the generated noise-reduced signal, ultimately enhancing the noise reduction effect. The first target feature tensor and the feature tensor are then input into a first feature fusion layer in the signal generation layer to obtain a second target feature tensor output by the first feature fusion layer. This establishes a residual network structure for deeper feature modeling, improving the robustness of the noise cancellation model and ultimately enhancing the noise reduction effect. The second target feature tensor is then input into a deconvolutional mapping layer in the signal generation layer to obtain a noise-reduced signal output by the deconvolutional mapping layer. This deconvolutional mapping layer includes a deconvolutional layer and a nonlinear activation function layer, allowing for further extraction of nonlinear noise features and providing nonlinear modeling capabilities. This enables the generation of a noise-reduced signal that eliminates nonlinear noise, effectively eliminating nonlinear noise and further improving the noise reduction effect.

[0121] Based on any of the above embodiments, in this method, the deconvolution mapping layer further includes a nonlinear threshold layer, which is constructed based on the following nonlinear threshold function:

[0122] f(x) = kx, x > 0;

[0123] f(x) = m(exp(x) - n), x ≤ 0;

[0124] In the formula, x represents the input of the nonlinear threshold layer, f(x) represents the output of the nonlinear threshold layer, k represents a preset first constant, m represents a preset second constant, n represents a preset third constant, and exp() represents an exponential function with the natural constant e as the base.

[0125] For example, if k is 0.5, m is 0.83, and n is 2, then the nonlinear threshold function is as follows:

[0126] f(x) = 0.5x, x > 0;

[0127] f(x)=0.83(exp(x)-2), x≤0.

[0128] The vehicle noise cancellation method provided in this invention uses the aforementioned nonlinear threshold function to perform nonlinear processing on the input when the input is less than or equal to 0, thereby further extracting the features of nonlinear noise and providing nonlinear modeling capabilities to further improve the accuracy of noise reduction signal generation, and ultimately further improve the noise reduction effect.

[0129] Based on any of the above embodiments, in this method, step 120 includes:

[0130] The input signal is input to the nonlinear feature extraction layer in the feature extraction layer to obtain the nonlinear feature tensor output by the nonlinear feature extraction layer, wherein the nonlinear feature extraction layer includes the nonlinear activation function layer;

[0131] The input signal is input to the linear feature extraction layer in the feature extraction layer to obtain the linear feature tensor output by the linear feature extraction layer;

[0132] The nonlinear feature tensor and the linear feature tensor are input into the second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.

[0133] In some embodiments, the nonlinear feature extraction layer further includes a linear layer (fully connected layer). For example, the nonlinear feature extraction layer includes a linear layer, a linear layer and a nonlinear activation function layer connected in sequence, that is, the nonlinear feature extraction layer includes two linear layers and a nonlinear activation function layer connected in sequence. Of course, the number of linear layers is not specifically limited.

[0134] In some embodiments, the linear feature extraction layer includes linear layers to perform linear feature extraction. For example, the linear feature extraction layer includes multiple linear layers connected in sequence; for instance, the linear feature extraction layer includes three linear layers connected in sequence. Of course, the number of linear layers is not specifically limited.

[0135] In one embodiment, if the input signal is a complex spectrum signal, the linear layer included in the nonlinear feature extraction layer is a two-dimensional linear layer, and the linear layer included in the linear feature extraction layer is a two-dimensional linear layer, so that the output of the two-dimensional linear layer is a feature tensor, rather than a one-dimensional feature vector.

[0136] The vehicle noise cancellation method provided in this invention inputs an input signal to a nonlinear feature extraction layer in the feature extraction layer to obtain a nonlinear feature tensor output by the nonlinear feature extraction layer. The input signal is then input to a linear feature extraction layer in the feature extraction layer to obtain a linear feature tensor output by the linear feature extraction layer. Both the nonlinear and linear feature tensors are then input to a second feature fusion layer in the feature extraction layer to obtain a feature tensor output by the second feature fusion layer. This method extracts not only linear features but also nonlinear features, meaning it performs not only linear modeling but also nonlinear modeling, extracting not only the linear noise but also the nonlinear noise. This effectively eliminates both nonlinear and linear noise, thereby improving the noise reduction effect.

[0137] Based on any of the above embodiments, the nonlinear activation function layer is constructed based on the following nonlinear activation function:

[0138] f(x) = sin(x) + cos(x);

[0139] In the formula, x represents the input of the nonlinear activation function layer, and f(x) represents the output of the nonlinear activation function layer.

[0140] The vehicle noise cancellation method provided in this embodiment of the invention can perform nonlinear processing on the input through the aforementioned nonlinear activation function layer, thereby extracting the features of nonlinear noise, thus providing nonlinear modeling capability to effectively eliminate nonlinear noise and ultimately improve the noise reduction effect.

[0141] Based on any of the above embodiments, in this method, the tag signal is a noise signal collected by a microphone inside the vehicle.

[0142] In some embodiments, to improve noise reduction performance, the data acquisition process for both the training and application (inference) of the noise cancellation model is based on the same target vehicle. Therefore, the tag signal is the noise signal acquired from the microphone inside the target vehicle.

[0143] Accordingly, the noise cancellation model is trained in the following manner:

[0144] The sample input signal is input to the model to be trained to obtain the sample denoising signal output by the model to be trained;

[0145] The sample denoised signal is convolved with a preset transmission path to obtain the predicted denoised signal that is transmitted from the sample denoised signal to the microphone.

[0146] Based on the sum of the predicted denoised signal and the label signal, the loss value corresponding to the loss function is determined;

[0147] The model to be trained is trained based on the loss value.

[0148] Since the label signal is a noise signal collected by the microphone inside the car, while the sample noise reduction signal is usually output through the speaker, there will be path loss when the audio output from the speaker reaches the microphone. Therefore, in order to improve the noise reduction effect, that is, to improve the training effect of the noise cancellation model, it is necessary to first perform a convolution operation between the sample noise reduction signal and the preset transmission path to obtain the predicted noise reduction signal transmitted from the sample noise reduction signal to the microphone.

[0149] Here, the preset transmission path is pre-set and can be configured based on the transmission path between the car's speakers and microphone. This preset transmission path characterizes the audio loss rate from the speakers to the microphone. By convolving the sample noise-reduced signal with the preset transmission path, a predicted noise-reduced signal is obtained from the sample noise-reduced signal being transmitted from the speakers to the microphone. This predicted noise-reduced signal is the signal after removing path loss; that is, it serves as the control signal at the microphone, thereby simulating the process of the sample noise-reduced signal propagating from the speakers to the microphone.

[0150] Since the model ultimately outputs an inverted control signal, the loss value corresponding to the loss function is determined based on the sum of the predicted denoised signal and the labeled signal. Furthermore, the loss value corresponding to the loss function is determined based on the square of this sum.

[0151] For example, the model's loss function is shown below:

[0152]

[0153] In the formula, w represents the model parameters, N represents the number of samples in each training round, and y i This represents the label signal corresponding to the i-th sample input signal. This represents the predicted denoising signal corresponding to the i-th sample input signal.

[0154] It should be noted that during training, the backpropagation algorithm can be used to optimize the loss function, continuously updating the model parameters to lower the loss function value. Furthermore, after each round of training, the model is evaluated using a validation set to avoid overfitting. Further, techniques such as cross-validation are used for model selection and parameter tuning to improve the model's accuracy and generalization ability. After multiple rounds of training and parameter tuning, the model with the smallest loss function value on the test set is selected as the final noise reduction model.

[0155] The vehicle noise cancellation method provided in this invention addresses the issue that the label signal is a noise signal collected by a microphone inside the vehicle, and there is a path loss when the audio output from the speaker reaches the microphone. Therefore, the method performs a convolution operation between the sample denoised signal and a preset transmission path to obtain a predicted denoised signal transmitted from the sample denoised signal to the microphone. This improves the training effect of the noise cancellation model and thus enhances the denoising effect. Furthermore, since the model ultimately outputs an inverted control signal, the loss value corresponding to the loss function is determined based on the sum of the predicted denoised signal and the label signal. This ensures the training effect of the model and further improves the denoising effect. In other words, the model training can be effectively completed using the aforementioned loss function.

[0156] Based on any of the above embodiments, in this method, step 110 includes:

[0157] The vibration signal is converted into a first complex spectrum signal, and the operating state is converted into a second complex spectrum signal;

[0158] The input signal is obtained by fusing the first complex spectrum signal and the second complex spectrum signal.

[0159] For example, a Fourier transform is performed on the vibration signal to obtain the first complex spectrum signal; for example, the sampling rate is 16kHz and the length of the Fourier transform is 320 points.

[0160] For example, the value of the running state is used as the real part of the second complex spectrum signal, and 0 is used as the imaginary part of the second complex spectrum signal to obtain the second complex spectrum signal.

[0161] It should be noted that if there are multiple operating states, multiple second complex spectrum signals are obtained by conversion, and the first complex spectrum signal and multiple second complex spectrum signals are fused to obtain the input signal, which is also a complex spectrum signal.

[0162] For example, the fusion method of the first complex spectrum signal and the second complex spectrum signal is to add the real part and the imaginary part respectively. In other words, the first complex spectrum signal is converted into a first matrix, the second complex spectrum signal is converted into a second matrix, the first matrix and the second matrix are fused to obtain a third matrix, and the third matrix is ​​converted into the input signal; for example, if the first matrix is ​​a 10*10 matrix and the second matrix is ​​a 1*10 matrix, then the third matrix is ​​an 11*10 matrix.

[0163] The vehicle noise cancellation method provided in this invention converts vibration signals into a first complex spectrum signal and operating states into a second complex spectrum signal. The first and second complex spectrum signals are then fused to obtain an input signal. Using the complex spectrum signal as model input improves the noise cancellation model's feature extraction capability from the input signal, thereby enhancing the noise reduction effect. Furthermore, fusing the first and second complex spectrum signals to obtain the input signal generates a noise reduction signal based not only on vibration signal information but also on operating state information, resulting in a more accurate noise reduction signal and further improving the noise reduction effect.

[0164] To facilitate understanding of the above embodiments, a specific embodiment will be described here. For example... Figure 5As shown, the noise reduction model includes a linear feature extraction layer, a nonlinear feature extraction layer, a causal convolutional layer, and a deconvolutional mapping layer. Specifically, the input signal is input to the nonlinear feature extraction layer to obtain a nonlinear feature tensor output by the nonlinear feature extraction layer; the input signal is then input to the linear feature extraction layer to obtain a linear feature tensor output by the linear feature extraction layer; the nonlinear feature tensor and the linear feature tensor are fused to obtain a feature tensor; the feature tensor is input to the causal convolutional layer to obtain a first target feature tensor output by the causal convolutional layer; the first target feature tensor is fused with the feature tensor to obtain a second target feature tensor; and the second target feature tensor is input to the deconvolutional mapping layer to obtain the denoised signal output by the deconvolutional mapping layer.

[0165] The automobile noise elimination device provided by the present invention is described below. The automobile noise elimination device described below can be referred to in correspondence with the automobile noise elimination method described above.

[0166] Figure 6 A schematic diagram of the structure of the automotive noise cancellation device provided by the present invention is shown below. Figure 6 As shown, the car noise cancellation device includes:

[0167] The signal determination module 610 is used to determine the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle;

[0168] Feature extraction module 620 is used to input the input signal into the feature extraction layer in the noise cancellation model to obtain the feature tensor output by the feature extraction layer;

[0169] The signal generation module 630 is used to input the feature tensor into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer.

[0170] Signal output module 640 is used to output the noise reduction signal;

[0171] The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal; the noise cancellation model is trained based on the sample input signal and the label signal corresponding to the sample input signal, and the sample input signal is determined based on the sample vibration signal and the sample running state.

[0172] The vehicle noise cancellation device provided in this invention determines the input signal based on the vibration signal and operating state of the target vehicle. This means it considers not only the vibration signal but also the vehicle's operating state. The input signal is then fed into the feature extraction layer of the noise cancellation model to obtain a feature tensor output by the feature extraction layer. This extracts features not only from the vibration signal but also from the operating state. The feature tensor is then input into the signal generation layer of the noise cancellation model to obtain a more accurate noise reduction signal, thereby improving the noise reduction effect. Simultaneously, the feature extraction layer includes a nonlinear activation function layer, which extracts the nonlinear feature tensor of the input signal, i.e., it can extract features of nonlinear noise, thus providing nonlinear modeling capabilities and effectively eliminating nonlinear noise, ultimately improving the noise reduction effect.

[0173] Based on any of the above embodiments, the signal generation module 630 is further configured to:

[0174] The feature tensor is input to the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer. The deconvolution mapping layer includes a deconvolution layer and a nonlinear activation function layer.

[0175] Based on any of the above embodiments, the signal generation module 630 is further configured to:

[0176] The feature tensor is input into the causal convolutional layer in the signal generation layer to obtain the first target feature tensor output by the causal convolutional layer.

[0177] The first target feature tensor and the feature tensor are input into the first feature fusion layer in the signal generation layer to obtain the second target feature tensor output by the first feature fusion layer.

[0178] The second target feature tensor is input into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer.

[0179] Based on any of the above embodiments, the deconvolution mapping layer further includes a nonlinear threshold layer, which is constructed based on the following nonlinear threshold function:

[0180] f(x) = kx, x > 0;

[0181] f(x) = m(exp(x) - n), x ≤ 0;

[0182] In the formula, x represents the input of the nonlinear threshold layer, f(x) represents the output of the nonlinear threshold layer, k represents a preset first constant, m represents a preset second constant, n represents a preset third constant, and exp() represents an exponential function with the natural constant e as the base.

[0183] Based on any of the above embodiments, the feature extraction module 620 is further configured to:

[0184] The input signal is input to the nonlinear feature extraction layer in the feature extraction layer to obtain the nonlinear feature tensor output by the nonlinear feature extraction layer, wherein the nonlinear feature extraction layer includes the nonlinear activation function layer;

[0185] The input signal is input to the linear feature extraction layer in the feature extraction layer to obtain the linear feature tensor output by the linear feature extraction layer;

[0186] The nonlinear feature tensor and the linear feature tensor are input into the second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.

[0187] Based on any of the above embodiments, the nonlinear activation function layer is constructed based on the following nonlinear activation function:

[0188] f(x) = sin(x) + cos(x);

[0189] In the formula, x represents the input of the nonlinear activation function layer, and f(x) represents the output of the nonlinear activation function layer.

[0190] Based on any of the above embodiments, the tag signal is a noise signal collected by a microphone inside the vehicle;

[0191] The device also includes a model training module, which is used for:

[0192] The sample input signal is input to the model to be trained to obtain the sample denoising signal output by the model to be trained;

[0193] The sample denoised signal is convolved with a preset transmission path to obtain the predicted denoised signal that is transmitted from the sample denoised signal to the microphone.

[0194] Based on the sum of the predicted denoised signal and the label signal, the loss value corresponding to the loss function is determined;

[0195] The model to be trained is trained based on the loss value.

[0196] The present invention also provides an automobile, which includes a vibration sensor, a speaker, and a processor. The vibration sensor is used to acquire vibration signals of the automobile; the speaker is used to output noise reduction signals; and the processor is used to execute the automobile noise cancellation method of any of the above embodiments.

[0197] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a vehicle noise cancellation method. This method includes: determining an input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle; inputting the input signal to a feature extraction layer in a noise cancellation model to obtain a feature tensor output by the feature extraction layer; inputting the feature tensor to a signal generation layer in the noise cancellation model to obtain a denoised signal output by the signal generation layer; and outputting the denoised signal. The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal. The noise cancellation model is trained based on a sample input signal and the corresponding label signal, and the sample input signal is determined based on the sample vibration signal and the sample operating state.

[0198] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0199] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the vehicle noise cancellation method provided by the above methods. The method includes: determining an input signal based on a vibration signal of a target vehicle and the operating state of the target vehicle; inputting the input signal to a feature extraction layer in a noise cancellation model to obtain a feature tensor output by the feature extraction layer; inputting the feature tensor to a signal generation layer in the noise cancellation model to obtain a denoised signal output by the signal generation layer; and outputting the denoised signal. The feature extraction layer includes a nonlinear activation function layer for extracting a nonlinear feature tensor of the input signal. The noise cancellation model is trained based on a sample input signal and a corresponding label signal, wherein the sample input signal is determined based on a sample vibration signal and a sample operating state.

[0200] The device embodiments described above are merely illustrative. 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; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0201] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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.

Claims

1. A method for eliminating automobile noise, characterized in that, include: The input signal is determined based on the vibration signal of the target vehicle and the operating status of the target vehicle; The input signal is fed into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer. The feature tensor is input into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer; Output the noise reduction signal; The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal. The noise cancellation model is trained based on the sample input signal and the label signal corresponding to the sample input signal. The sample input signal is determined based on the sample vibration signal and the sample running state.

2. The automotive noise canceling method of claim 1, wherein, The step of inputting the feature tensor into the signal generation layer of the noise cancellation model to obtain the denoised signal output by the signal generation layer includes: The feature tensor is input to the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer. The deconvolution mapping layer includes a deconvolution layer and a nonlinear activation function layer.

3. The vehicle noise elimination method according to claim 2, characterized in that, The step of inputting the feature tensor into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer includes: The feature tensor is input into the causal convolutional layer in the signal generation layer to obtain the first target feature tensor output by the causal convolutional layer. The first target feature tensor and the feature tensor are input into the first feature fusion layer in the signal generation layer to obtain the second target feature tensor output by the first feature fusion layer. The second target feature tensor is input into the deconvolution mapping layer in the signal generation layer to obtain the denoised signal output by the deconvolution mapping layer.

4. The vehicle noise elimination method according to claim 2, characterized in that, The deconvolutional mapping layer further includes a nonlinear threshold layer, which is constructed based on the following nonlinear threshold function: f(x) = kx, x > 0; f(x) = m(exp(x) - n), x ≤ 0; In the formula, x represents the input of the nonlinear threshold layer, f(x) represents the output of the nonlinear threshold layer, k represents a preset first constant, m represents a preset second constant, n represents a preset third constant, and exp() represents an exponential function with the natural constant e as the base.

5. The vehicle noise elimination method according to claim 1, characterized in that, The step of inputting the input signal into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer includes: The input signal is input to the nonlinear feature extraction layer in the feature extraction layer to obtain the nonlinear feature tensor output by the nonlinear feature extraction layer, wherein the nonlinear feature extraction layer includes the nonlinear activation function layer; The input signal is input to the linear feature extraction layer in the feature extraction layer to obtain the linear feature tensor output by the linear feature extraction layer; The nonlinear feature tensor and the linear feature tensor are input into the second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.

6. The automotive noise canceling method of claim 1, 2, or 5, wherein, The nonlinear activation function layer is constructed based on the following nonlinear activation function: f(x) = sin(x) + cos(x); In the formula, x represents the input of the nonlinear activation function layer, and f(x) represents the output of the nonlinear activation function layer.

7. The vehicle noise elimination method according to any one of claims 1 to 5, characterized in that, The tag signal is a noise signal collected by a microphone inside the car. The noise cancellation model is trained in the following manner: The sample input signal is input to the model to be trained to obtain the sample denoising signal output by the model to be trained; The sample denoised signal is convolved with a preset transmission path to obtain the predicted denoised signal that is transmitted from the sample denoised signal to the microphone. Based on the sum of the predicted denoised signal and the label signal, the loss value corresponding to the loss function is determined; The model to be trained is trained based on the loss value.

8. The automotive noise cancellation method of claim 1, wherein, The determination of the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle includes: The vibration signal is converted into a first complex spectrum signal, and the operating state is converted into a second complex spectrum signal; The input signal is obtained by fusing the first complex spectrum signal and the second complex spectrum signal.

9. An automobile noise canceling apparatus characterized by comprising: include: The signal determination module is used to determine the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle; The feature extraction module is used to input the input signal into the feature extraction layer of the noise cancellation model to obtain the feature tensor output by the feature extraction layer; The signal generation module is used to input the feature tensor into the signal generation layer in the noise cancellation model to obtain the noise-reduced signal output by the signal generation layer. The signal output module is used to output the noise reduction signal; The feature extraction layer includes a nonlinear activation function layer, which is used to extract the nonlinear feature tensor of the input signal. The noise cancellation model is trained based on the sample input signal and the corresponding label signal, and the sample input signal is determined based on the sample vibration signal and the sample running state.

10. A car, characterized in that, include: A vibration sensor, used to acquire vibration signals from the vehicle; A loudspeaker, the loudspeaker being used to output a noise-reducing signal; A processor for performing the vehicle noise cancellation method according to any one of claims 1 to 8.

11. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the vehicle noise cancellation method as described in any one of claims 1 to 8.

12. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle noise cancellation method as described in any one of claims 1 to 8.

Images