CN117765912A - Automobile noise elimination method and device, automobile, electronic equipment and storage medium - Google Patents
Automobile noise elimination method and device, automobile, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN117765912A CN117765912A CN202311629578.9A CN202311629578A CN117765912A CN 117765912 A CN117765912 A CN 117765912A CN 202311629578 A CN202311629578 A CN 202311629578A CN 117765912 A CN117765912 A CN 117765912A
- Authority
- CN
- China
- Prior art keywords
- signal
- layer
- noise
- nonlinear
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 73
- 230000008030 elimination Effects 0.000 title claims abstract description 64
- 238000003379 elimination reaction Methods 0.000 title claims abstract description 64
- 230000009467 reduction Effects 0.000 claims abstract description 136
- 238000000605 extraction Methods 0.000 claims abstract description 113
- 230000004913 activation Effects 0.000 claims abstract description 62
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 claims abstract description 55
- 230000006870 function Effects 0.000 claims description 82
- 238000013507 mapping Methods 0.000 claims description 62
- 238000001228 spectrum Methods 0.000 claims description 47
- 230000004927 fusion Effects 0.000 claims description 28
- 238000012549 training Methods 0.000 claims description 23
- 230000001364 causal effect Effects 0.000 claims description 19
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 abstract description 35
- 238000010606 normalization Methods 0.000 description 14
- 239000011159 matrix material Substances 0.000 description 12
- 230000008569 process Effects 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 10
- 230000001133 acceleration Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 239000000725 suspension Substances 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 3
- 230000001902 propagating effect Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 230000001131 transforming effect Effects 0.000 description 3
- 239000006096 absorbing agent Substances 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000013019 agitation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000009432 framing Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000035939 shock Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000037433 frameshift Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000011148 porous material Substances 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000011282 treatment Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
Abstract
The invention provides an automobile noise elimination method and device, an automobile, electronic equipment and a storage medium, and relates to the technical field of noise elimination. The method comprises the following steps: determining an input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle; inputting an input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer; inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer; outputting a noise reduction signal; the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state. The invention can effectively eliminate nonlinear noise and improve the noise reduction effect.
Description
Technical Field
The present invention relates to the field of noise cancellation technology, and in particular, to an automobile noise cancellation method, an apparatus, an automobile, an electronic device, and a storage medium.
Background
With rapid development of technology, people have increasingly high requirements on riding experience of automobiles. Important factors that affect the ride experience include road agitation. The road noise is random broadband vibration generated by the automobile tire at the road contact point, and is transmitted into the carriage through the structure and air. Structural propagation refers to the propagation of vibrations through the suspension system to the vehicle body and then the emission of sound into the cabin, which is a major source of noise inside the vehicle below 500 hz, and thus the elimination of road noise propagating through the structure is required.
At present, road noise propagated through structures in automobiles can be reduced through active noise control technology. However, most of the current active noise control technologies adopt a linear adaptive filtering mode to reduce noise, and as most of the hydraulic shock absorbers and rubber bushings in the suspension systems of automobiles have nonlinear acoustic characteristics, the current active noise control technologies cannot reduce noise of nonlinear parts, so that the noise reduction effect is poor.
Disclosure of Invention
The invention provides an automobile noise elimination method, an automobile noise elimination device, an automobile, electronic equipment and a storage medium, which are used for solving the defect that nonlinear noise cannot be eliminated in the prior art and realizing a better noise reduction effect.
The invention provides an automobile noise elimination method, which comprises the following steps:
determining an input signal based on a vibration signal of a target vehicle and a running state of the target vehicle;
inputting the input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer;
inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer;
outputting the noise reduction signal;
the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
According to the method for eliminating the automobile noise provided by the invention, the characteristic tensor is input into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer, and the method comprises the following steps:
and inputting the characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer comprises a deconvolution layer and a nonlinear activation function layer.
According to the method for eliminating the automobile noise provided by the invention, the characteristic tensor is input to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, and the method comprises the following steps:
inputting the characteristic tensor to a causal convolution layer in the signal generation layer to obtain a first target characteristic tensor output by the causal convolution layer;
inputting the first target feature tensor and the feature tensor 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;
and inputting the second target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer.
According to the method for eliminating the automobile noise, the deconvolution mapping layer further comprises a nonlinear threshold layer, and the nonlinear threshold layer is constructed based on the following nonlinear threshold function:
f(x)＝kx，x>0；
f(x)＝m(exp(x)-n)，x≤0；
wherein 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, exp () represents an exponential function based on a natural constant e.
According to the method for eliminating the automobile noise provided by the invention, the input signal is input to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer, and the method comprises the following steps:
inputting the 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, wherein the nonlinear feature extraction layer comprises the nonlinear activation function layer;
inputting the input signal to a linear feature extraction layer in the feature extraction layers to obtain a linear feature tensor output by the linear feature extraction layer;
and inputting the nonlinear feature tensor and the linear feature tensor into a second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.
According to the method for eliminating the automobile noise, the nonlinear activation function layer is constructed based on the following nonlinear activation functions:
f(x)＝sin(x)+cos(x)；
where x represents the input of the nonlinear activation function layer and f (x) represents the output of the nonlinear activation function layer.
According to the method for eliminating the noise of the automobile, the tag signal is a noise signal acquired based on a microphone in the automobile;
The noise cancellation model is trained based on the following:
inputting the sample input signal to a model to be trained to obtain a sample noise reduction signal output by the model to be trained;
performing convolution operation on the sample noise reduction signal and a preset transmission path to obtain a prediction noise reduction signal transmitted to the microphone by the sample noise reduction signal;
determining a loss value corresponding to a loss function based on the sum of the predicted noise reduction signal and the tag signal;
and training the model to be trained based on the loss value.
According to the method for eliminating the noise of the automobile, the method for determining the input signal based on the vibration signal of the target automobile and the running state of the target automobile comprises the following steps:
converting the vibration signal into a first complex spectrum signal and converting the operating state into a second complex spectrum signal;
and fusing the first complex spectrum signal and the second complex spectrum signal to obtain an input signal.
The invention also provides an automobile noise elimination device, which comprises:
the signal determining module is used for determining an input signal based on a vibration signal of a target automobile and the running state of the target automobile;
The feature extraction module is used for inputting the input signal to a feature extraction layer in the noise elimination model to obtain a feature tensor output by the feature extraction layer;
the signal generation module is used for inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer;
the signal output module is used for outputting the noise reduction signal;
the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
The present invention also provides an automobile comprising:
the vibration sensor is used for acquiring a vibration signal of the automobile;
a speaker for outputting a noise reduction signal;
a processor for performing the method of car noise cancellation as described in any one of the above.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of car noise cancellation as described in any of the above when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method of noise cancellation for a vehicle as described in any of the above.
According to the automobile noise elimination method, the device, the automobile, the electronic equipment and the storage medium, based on the vibration signal of the target automobile and the running state of the target automobile, the input signal is determined, so that not only the vibration signal is considered, but also the running state of the automobile is considered, the input signal is input to the feature extraction layer in the noise elimination model to obtain the feature tensor output by the feature extraction layer, so that the features of the vibration signal are extracted, the features of the running state are extracted, the feature tensor is input to the signal generation layer in the noise elimination model, and further a more accurate noise reduction signal is obtained, and the noise reduction effect is improved; meanwhile, the feature extraction layer comprises a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting nonlinear feature tensors of input signals, namely, the features of nonlinear noise can be extracted, so that nonlinear modeling capacity is provided, the nonlinear noise can be effectively eliminated, and finally, the noise reduction effect is improved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an automobile noise elimination method provided by the invention;
FIG. 2 is a schematic layout of an automotive device according to the present invention;
FIG. 3 is a second schematic layout of the automotive device according to the present invention;
FIG. 4 is a schematic diagram of a noise cancellation model according to the present invention;
FIG. 5 is a second schematic diagram of a noise cancellation model according to the present invention;
FIG. 6 is a schematic diagram of a noise cancellation device for an automobile according to the present invention;
fig. 7 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
With rapid development of technology, people have increasingly high requirements on riding experience of automobiles. Important factors that affect the ride experience include road agitation. The road noise is random broadband vibration generated by the automobile tire at the road contact point, and is transmitted into the carriage through the structure and air. Structural propagation refers to the propagation of vibrations through the suspension system to the vehicle body and then the emission of sound into the cabin, which is a major source of noise inside the vehicle below 500 hz, and thus the elimination of road noise propagating through the structure is required.
Considering the low frequency characteristics of the road noise propagating through the structure, making it difficult to control by passive noise control techniques commonly used in automobiles, such as viscoelastic vibration damping treatments applied to porous materials in body panels or cabins, these methods are less effective at low frequencies because the wavelengths of sound and vibration are similar to the thickness of the material. Therefore, at present, the road noise propagated through the structure in the automobile is mostly reduced by the active noise control technology, and the effect is better at low frequency. However, most of the current active noise control technologies adopt a linear adaptive filtering mode to reduce noise, and as most of the hydraulic shock absorbers and rubber bushings in the suspension systems of automobiles have nonlinear acoustic characteristics, the current active noise control technologies cannot reduce the noise of nonlinear parts, so that the noise reduction effect is poor, namely, the noise reduction effect and the noise reduction space are limited.
In view of the above problems, the present invention proposes the following embodiments. Fig. 1 is a schematic flow chart of an automobile noise elimination method provided by the invention, and as shown in fig. 1, the automobile noise elimination method comprises the following steps:
step 110, determining an input signal based on a vibration signal of a target vehicle and an operating state of the target vehicle.
Here, the target automobile is an automobile to be noise reduced currently. The vibration signal can be acquired by a vibration sensor, and the vibration sensor is arranged on the target automobile.
In an embodiment, the vibration sensor is an acceleration sensor, so as to obtain acceleration information of the target automobile through the acceleration sensor, and then generate a vibration signal through the acceleration information. Further, the vibration sensor is a triaxial acceleration sensor.
In one embodiment, as shown in FIG. 2, the vibration sensor includes four accelerometers, each of which is located at the suspension system of each tire of the bottom of the target automobile.
Here, the operating state may include, but is not limited to, at least one of: speed of the target car, acceleration of the target car, steering of the target car, etc. The operating state can be detected by an inherent component of the automobile, and will not be described in detail herein.
Here, the input signal is used to characterize vibration information of the target car and operation state information of the target car. In other words, the vibration signal of the target automobile and the running state of the target automobile are fused to obtain the input signal, and the specific fusion mode is not particularly limited herein, and only the information that the input signal can simultaneously represent the two needs to be ensured.
In an embodiment, the vibration signal of the target automobile is converted into a first complex spectrum signal, the running state of the target automobile is converted into a second complex spectrum signal, and the first complex spectrum signal and the second complex spectrum signal are fused to obtain an input signal, wherein the input signal is also the complex spectrum signal. Based on the above, the complex spectrum signal is used as the input of the noise cancellation model, and the output of the noise cancellation model is also the complex spectrum signal; accordingly, outputting the noise reduction signal requires converting the noise reduction signal into a signal in the time domain. The complex spectrum signal is input as a model, so that the feature extraction capability of the noise elimination model on the input signal can be improved, and the noise reduction effect of the noise elimination model is further improved.
In another embodiment, the vibration signal of the target automobile and the running state of the target automobile are fused to obtain input data; the input data is converted into a complex spectrum signal (input signal). Illustratively, fourier transforming the input data to obtain an input signal; for example, the sampling rate is 16kHz and the length of the fourier transform is 320 points. Based on the above, the complex spectrum signal is used as the input of the noise cancellation model, and the output of the noise cancellation model is also the complex spectrum signal; accordingly, outputting the noise reduction signal requires converting the noise reduction signal into a signal in the time domain. The complex spectrum signal is input as a model, so that the feature extraction capability of the noise elimination model on the input signal can be improved, and the noise reduction effect of the noise elimination model is further improved.
And step 120, inputting the input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer.
The feature extraction layer comprises a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting nonlinear feature tensors of the input signals.
Here, the noise cancellation model is used to generate a corresponding noise reduction signal based on the input signal, and the feature extraction layer is used to perform feature extraction on the input signal. It should be understood that the feature extraction layer includes a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting a nonlinear feature tensor of an input signal, that is, extracting features of nonlinear noise, so as to provide nonlinear modeling capability, further effectively eliminate the nonlinear noise, and finally improve the noise reduction effect.
In some embodiments, an input signal is input to a nonlinear feature extraction layer in the feature extraction layers, resulting in a feature tensor output by the nonlinear feature extraction layer, the nonlinear feature extraction layer comprising a nonlinear activation function layer, the feature tensor being a nonlinear feature tensor.
In an embodiment, the nonlinear feature extraction layer further comprises a linear layer (fully connected layer). The nonlinear feature extraction layer includes a linear layer, and a nonlinear activation function layer that are sequentially connected, that is, the nonlinear feature extraction layer includes two linear layers and a nonlinear activation function layer that are sequentially connected, and of course, the number of layers of the linear layer may not be specifically limited.
In one embodiment, if the input signal is a complex spectrum signal, the nonlinear feature extraction layer includes a linear layer that is a two-dimensional linear layer, such that the output of the two-dimensional linear layer is a feature tensor, rather than a one-dimensional feature vector.
And 130, inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer.
Here, the signal generation layer generates a corresponding noise reduction signal based on the feature tensor extracted by the feature extraction layer. It should be understood that the signal generation layer may be set according to actual needs.
In some embodiments, the feature tensor is input to a deconvolution mapping layer in the signal generation layer to obtain the noise reduction signal output by the deconvolution mapping layer, which includes the deconvolution layer.
In some embodiments, the feature tensor is input to a causal convolution layer in the signal generation layer to obtain a first target feature tensor output by the causal convolution layer; and inputting the first target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer comprises a deconvolution layer. Further, inputting the first target feature tensor and the feature tensor 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; and inputting the second target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer.
Wherein the deconvolution layer is used to deconvolve the input. The causal convolution layer can buffer the input at the previous moment, so that the input is modeled in time, namely the problem of time sequence speculation is realized, and the generation accuracy of the noise reduction signal is improved.
Further, the deconvolution mapping layer also includes a normalization layer, e.g., the normalization layer is an example normalization layer.
Further, the deconvolution mapping layer also includes a non-linear threshold layer.
In one embodiment, if the input signal is a complex spectrum signal, the deconvolution layer is a two-dimensional deconvolution layer to two-dimensionally deconvolve the input.
In one embodiment, the deconvolution mapping layer comprises a plurality of mapping layers, any of which comprises a deconvolution layer. Further, the mapping layer also includes a normalization layer, e.g., the normalization layer is an instance normalization layer.
And 140, outputting the noise reduction signal.
Specifically, a noise reduction signal (control signal) is output through a speaker to achieve in-vehicle noise reduction. The speaker is disposed inside the target automobile. As an example, as shown in fig. 2, 6 speakers are provided in the cabin of the target car.
In one embodiment, if the input signal is a complex spectrum signal, the noise reduction signal needs to be converted into a control signal in the time domain, and then the control signal is output.
Further, the control signal is a Digital signal, and before outputting the control signal, the control signal needs to be converted into an Analog signal by a Digital-to-Analog Converter (DAC), and then the Analog signal is output by a speaker in the vehicle.
In addition, the speaker is tuned in advance in consideration of the difference in specification and parameters of each speaker, so that the noise reduction effect can be ensured. In tuning a loudspeaker, parameters such as frequency response characteristics, sensitivity and the like of the loudspeaker need to be determined first, and then the loudspeaker is tested and calibrated by using a corresponding acoustic testing instrument. During calibration, the performance of the speaker may be evaluated using frequency response flatness, phase response, distortion, sensitivity, etc., and adjusted to achieve optimal results.
It should be understood that the noise cancellation model may generate a corresponding noise reduction signal (control signal) in real time according to the vibration signal and the running state of the target automobile, thereby realizing real-time in-vehicle noise reduction.
The noise elimination model is trained 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 running state.
Here, the sample vibration signal may be acquired by a vibration sensor provided in the automobile.
In one embodiment, as shown in FIG. 3, the vibration sensor includes four accelerometers, each of which is located at the suspension system of each tire of the vehicle's bottom.
Here, the sample operating state may include, but is not limited to, at least one of: speed of the car, acceleration of the car, steering of the car, etc. The sample operating state may be detected by an inherent component of the vehicle, and will not be described in detail herein.
Here, the sample input signal is used to characterize vibration information of the vehicle and operational state information of the vehicle. In other words, the sample vibration signal and the sample operation state are fused to obtain the sample input signal, and the specific fusion mode is not specifically limited herein, and only the information that the sample input signal can simultaneously represent the two needs to be ensured.
In one embodiment, the sample vibration signal and the sample running state are fused to obtain sample input data; the sample input data is converted into a complex spectrum signal (sample input signal). Illustratively, fourier transforming the sample input data results in a sample input signal.
In an embodiment, the tag signal may be a noise reduction signal tag corresponding to the marked sample input signal.
In another embodiment, the tag signal is a noise signal collected based on a microphone in the car, i.e. the noise signal is collected at the same time as the sample vibration signal and the sample operating state are collected.
In some embodiments, to improve noise reduction, the training process of the noise cancellation model and the data acquisition process of the application process (reasoning process) are all performed based on the same target car. The input signal is input to a noise elimination model corresponding to the target automobile to obtain a noise reduction signal output by the noise elimination model, and the sample vibration signal and the sample running state are acquired based on the target automobile. In other words, model training is performed on the vehicle type needing noise reduction in advance to obtain a noise elimination model corresponding to the vehicle type.
In one embodiment, as shown in fig. 3, 4 accelerometers (acceleration sensors), 6 speakers, and 4 microphones are mounted on the target car. Further, the 4 microphones are located at the main driving position, the co-driving position, the rear left side position and the rear right side position, respectively. Of course, the number and arrangement of the above-described devices may be set according to the vehicle type. Wherein, 4 accelerometers are used for gathering vibration signal and sample vibration signal, and 6 speakers are used for the output noise reduction signal, and 4 microphones are used for gathering the label signal. The 4 microphones can be removed in the application process and only need to be installed in the training process.
In an embodiment, the sample vibration signal, the sample running state and the corresponding tag signal may be data collected by the target automobile on different road surfaces and at different speeds, that is, it is ensured that the training sample covers different vehicle conditions and road conditions, thereby ensuring richness of the training sample and further improving the training effect of the noise elimination model.
In an embodiment, after the sample vibration signal, the sample running state and the corresponding tag signal are collected, the collected training data may be subjected to data cleaning to remove the data which does not meet the condition; for example, the data of the automobile during idling is removed, specifically, the idling section in the training data is identified and removed by monitoring parameters such as the speed of the automobile or the rotation speed of the engine, only the data related to the actual driving situation is reserved, and any abnormal value or unreasonable data point also needs to be detected and removed in the idling section; these outliers may be due to sensor errors or other problems that can adversely affect the training of the model. The cleaned data set is ensured to have enough balance under different conditions so as to avoid the excessive dependence or bias of the model on the data under certain conditions.
In one embodiment, the tag signal may be obtained by framing the audio data collected by the microphone, where the framing of the audio data is to segment the continuous audio signal into short time periods, each time period being referred to as a frame. This step facilitates finer granularity processing and analysis of the audio data. First, the time length of each frame and the time interval between adjacent frames need to be selected, the common frame length can be between 20 ms and 40 ms, and the frame shift is typically 10 ms to 20 ms.
According to the automobile noise elimination method provided by the embodiment of the invention, based on the vibration signal of the target automobile and the running state of the target automobile, the input signal is determined, so that the vibration signal is considered, the running state of the automobile is considered, the input signal is input to the feature extraction layer in the noise elimination model, the feature tensor output by the feature extraction layer is obtained, so that the features of the vibration signal are extracted, the features of the running state are also extracted, the feature tensor is input to the signal generation layer in the noise elimination model, and further a more accurate noise reduction signal is obtained, and the noise reduction effect is improved; meanwhile, the feature extraction layer comprises a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting nonlinear feature tensors of input signals, namely, the features of nonlinear noise can be extracted, so that nonlinear modeling capacity is provided, the nonlinear noise can be effectively eliminated, and finally, the noise reduction effect is improved.
Based on any of the above embodiments, the method in step 130 includes:
and inputting the characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer comprises a deconvolution layer and a nonlinear activation function layer.
Here, the deconvolution layer is used to deconvolute the input. The nonlinear activation function layer is used for extracting nonlinear feature tensors.
In one embodiment, if the input signal is a complex spectrum signal, the deconvolution layer is a two-dimensional deconvolution layer to two-dimensionally deconvolve the input.
In an embodiment, the deconvolution mapping layer further comprises a normalization layer, e.g., the normalization layer is an instance normalization layer.
In an embodiment, the deconvolution mapping layer further includes a nonlinear threshold layer to further extract the characteristics of nonlinear noise, thereby further providing nonlinear modeling capability and finally further improving noise reduction effect.
In one embodiment, the deconvolution mapping layer includes a plurality of mapping layers, any of which includes a deconvolution layer and a nonlinear activation function layer. Further, the mapping layers further include normalization layers, e.g., the normalization layer is an instance normalization layer, e.g., any mapping layer includes a deconvolution layer, an instance normalization layer, and a nonlinear activation function layer, connected in sequence.
In order to facilitate understanding of the above embodiments, a specific embodiment is described herein. As shown in fig. 4, the deconvolution mapping layer includes a nonlinear threshold layer, a mapping layer, a nonlinear threshold layer, a mapping layer and a nonlinear threshold layer that are sequentially connected, and any mapping layer includes a deconvolution layer, an instance normalization layer and a nonlinear activation function layer that are sequentially connected. Based on this, an accurate noise reduction signal can be generated.
According to the automobile noise elimination method provided by the embodiment of the invention, the feature tensor is input to the deconvolution mapping layer in the signal generation layer to obtain the noise reduction signal output by the deconvolution mapping layer, and the deconvolution mapping layer comprises the deconvolution layer and the nonlinear activation function layer, so that the features of nonlinear noise can be further extracted, the nonlinear modeling capability is further provided, the noise reduction signal for eliminating the nonlinear noise can be further generated, the nonlinear noise can be effectively eliminated, and finally the noise reduction effect is further improved.
Based on any one of the foregoing embodiments, in the method, the inputting the feature tensor to a deconvolution mapping layer in the signal generation layer, to obtain a noise reduction signal output by the deconvolution mapping layer includes:
Inputting the characteristic tensor to a causal convolution layer in the signal generation layer to obtain a first target characteristic tensor output by the causal convolution layer;
inputting the first target feature tensor and the feature tensor 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;
and inputting the second target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer.
Here, the causal convolution layer can buffer the input at the previous moment, so as to model the input in time, namely, realize the time sequence estimation problem, further improve the generation accuracy of the noise reduction signal, and finally further improve the noise reduction effect. Illustratively, the causal convolution layer comprises a plurality of nonlinear one-dimensional convolution layers connected in sequence.
The first feature fusion layer is used for carrying out feature fusion on the first target feature tensor and the feature tensor, namely carrying out causal convolution on one branch and mapping on identity of the one branch, so that a residual network structure is realized, further feature modeling is carried out, robustness of a noise elimination model is improved, and finally noise reduction effect is improved.
According to the automobile noise elimination method provided by the embodiment of the invention, the feature tensor is input to the causal convolution layer in the signal generation layer to obtain the first target feature tensor output by the causal convolution layer, so that the input is modeled in time, namely, the problem of time sequence speculation is realized, the generation accuracy of noise reduction signals is further improved, and finally the noise reduction effect is further improved; the method comprises the steps of inputting a first target feature tensor and a feature tensor into a first feature fusion layer in a signal generation layer to obtain a second target feature tensor output by the first feature fusion layer, so that a residual network structure is realized, further feature modeling is conducted, robustness of a noise elimination model is improved, the noise reduction effect is finally improved, the second target feature tensor is input into a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, the deconvolution mapping layer comprises a deconvolution layer and a nonlinear activation function layer, and therefore the feature of nonlinear noise can be further extracted, nonlinear modeling capacity is further provided, a noise reduction signal for eliminating the nonlinear noise can be generated, the nonlinear noise can be effectively eliminated, and the noise reduction effect is finally further improved.
Based on any of the foregoing embodiments, in the method, the deconvolution mapping layer further includes a nonlinear threshold layer, and the nonlinear threshold layer is constructed based on a nonlinear threshold function as follows:
f(x)＝kx，x>0；
f(x)＝m(exp(x)-n)，x≤0；
wherein 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, exp () represents an exponential function based on a natural constant e.
Illustratively, k is 0.5, m is 0.83, and n is 2, the nonlinear threshold function is as follows:
f(x)＝0.5x，x>0；
f(x)＝0.83(exp(x)-2)，x≤0。
according to the automobile noise elimination method provided by the embodiment of the invention, under the condition that the input is smaller than or equal to 0, the nonlinear processing is carried out on the input through the nonlinear threshold function, so that the characteristics of nonlinear noise are further extracted, the nonlinear modeling capability is further provided, the generation accuracy of noise reduction signals is further improved, and finally the noise reduction effect is further improved.
Based on any one of the above embodiments, the method further includes the step 120 of:
inputting the 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, wherein the nonlinear feature extraction layer comprises the nonlinear activation function layer;
Inputting the input signal to a linear feature extraction layer in the feature extraction layers to obtain a linear feature tensor output by the linear feature extraction layer;
and inputting the nonlinear feature tensor and the linear feature tensor into a second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.
In some embodiments, the nonlinear feature extraction layer further comprises a linear layer (fully connected layer). The nonlinear feature extraction layer includes a linear layer, and a nonlinear activation function layer that are sequentially connected, that is, the nonlinear feature extraction layer includes two linear layers and a nonlinear activation function layer that are sequentially connected, and of course, the number of layers of the linear layer may not be specifically limited.
In some embodiments, the linear feature extraction layer comprises a linear layer to perform linear feature extraction. The linear feature extraction layer includes a plurality of linear layers connected in sequence, for example, the linear feature extraction layer includes three linear layers connected in sequence, and the number of layers of the linear layers may be not particularly limited.
In an 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, and is not a one-dimensional feature vector.
According to the automobile noise elimination method provided by the embodiment of the invention, 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, 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 to the second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer, so that not only the linear feature is extracted, but also the nonlinear feature is extracted, namely, not only the linear modeling is performed, but also the nonlinear modeling is performed, namely, the noise of the linear part is extracted, and the noise of the nonlinear part is extracted, so that the nonlinear noise and the linear noise can be effectively eliminated, and the noise reduction effect is further improved.
Based on any of the above embodiments, the nonlinear activation function layer is constructed based on the nonlinear activation function as follows:
f(x)＝sin(x)+cos(x)；
where x represents the input of the nonlinear activation function layer and f (x) represents the output of the nonlinear activation function layer.
According to the automobile noise elimination method provided by the embodiment of the invention, through the nonlinear activation function layer, nonlinear processing can be performed on the input, so that the characteristics of nonlinear noise are extracted, nonlinear modeling capability is provided, nonlinear noise is effectively eliminated, and finally the noise reduction effect is improved.
In accordance with any of the above embodiments, the tag signal is a noise signal collected based on a microphone in the vehicle.
In some embodiments, to improve noise reduction, the training process of the noise cancellation model and the data acquisition process of the application process (reasoning process) are all performed based on the same target car. Based on this, the tag signal is a noise signal acquired based on a microphone in the target car.
Accordingly, the noise cancellation model is trained based on the following:
inputting the sample input signal to a model to be trained to obtain a sample noise reduction signal output by the model to be trained;
performing convolution operation on the sample noise reduction signal and a preset transmission path to obtain a prediction noise reduction signal transmitted to the microphone by the sample noise reduction signal;
determining a loss value corresponding to a loss function based on the sum of the predicted noise reduction signal and the tag signal;
and training the model to be trained based on the loss value.
Because the tag signal is based on the noise signal acquired by the microphone in the automobile, the sample noise reduction signal is usually output through the loudspeaker, and the path loss exists from the audio output by the loudspeaker to the microphone, in order to improve the noise reduction effect, namely the training effect of the noise elimination model, the sample noise reduction signal and the preset transmission path are required to be subjected to convolution operation to obtain the prediction noise reduction signal transmitted to the microphone by the sample noise reduction signal.
Here, the preset transmission path is preset, and may be set according to transmission paths of a speaker and a microphone in the automobile. The preset transmission path can represent the loss rate of the audio from the loudspeaker to the microphone, so that convolution operation is carried out on the sample noise reduction signal and the preset transmission path, a predicted noise reduction signal transmitted from the loudspeaker to the microphone by the sample noise reduction signal is obtained, the predicted noise reduction signal is a signal with the path loss removed, namely the predicted noise reduction signal is a control signal at the microphone, and the process of transmitting the sample noise reduction signal from the loudspeaker to the microphone is simulated.
Since the model finally outputs an inverted control signal, a loss value corresponding to the loss function is determined based on the sum of the predictive noise reduction signal and the tag signal. Further, a loss value corresponding to the loss function is determined based on the square of the sum value.
Illustratively, the loss function of the model is as follows:
where w represents model parameters, N represents the number of samples in each round of training, y i Representing the tag signal corresponding to the i-th sample input signal,representing the predicted noise reduction signal corresponding to the i-th sample input signal.
It should be noted that, during the training process, the back propagation algorithm may be used to optimize the loss function, and continuously update the parameters of the model, so that the value of the loss function is continuously reduced. Further, after each round of training is completed, the model is evaluated using a validation set to avoid the occurrence of overfitting. Further, the model selection and parameter tuning are performed by using cross verification and other technologies, so that the accuracy and generalization capability of the model are improved. After multiple rounds of training and parameter tuning, the model with the smallest loss function value under the test set is selected as the final noise elimination model.
According to the automobile noise elimination method provided by the embodiment of the invention, as the tag signal is the noise signal acquired based on the microphone in the automobile, and the path loss exists from the audio output by the loudspeaker to the microphone, based on the path loss, the sample noise reduction signal and the preset transmission path are subjected to convolution operation to obtain the predicted noise reduction signal transmitted to the microphone by the sample noise reduction signal, so that the training effect of the noise elimination model can be improved, and the noise reduction effect is further improved; because the model finally outputs the inverted control signal, based on the inverted control signal, the loss value corresponding to the loss function is determined based on the sum value of the predictive noise reduction signal and the label signal, the training effect of the model can be ensured, the noise reduction effect is further improved, and the model training can be well completed through the loss function.
Based on any one of the above embodiments, the method in step 110 includes:
converting the vibration signal into a first complex spectrum signal and converting the operating state into a second complex spectrum signal;
and fusing the first complex spectrum signal and the second complex spectrum signal to obtain an input signal.
Illustratively, fourier transforming the vibration signal to obtain a first complex spectrum signal; for example, the sampling rate is 16kHz and the length of the fourier transform is 320 points.
Illustratively, the value of the operating state is taken as the real part of the second complex spectrum signal and 0 is taken as the imaginary part of the second complex spectrum signal to obtain the second complex spectrum signal.
If the number of the operation states is plural, a plurality of second complex spectrum signals are obtained by conversion, and the first complex spectrum signals and the plurality of second complex spectrum signals are fused to obtain an input signal, and the input signal is also a complex spectrum signal.
Illustratively, the first complex spectral signal and the second complex spectral signal are fused in such a way that the real part and the imaginary part are added separately. In other words, converting the first complex spectrum signal into a first matrix, converting the second complex spectrum signal into a second matrix, fusing the first matrix and the second matrix to obtain a third matrix, and converting the third matrix into an input signal; for example, the first matrix is a 10×10 matrix, the second matrix is a 1×10 matrix, and the third matrix is an 11×10 matrix.
According to the automobile noise elimination method provided by the embodiment of the invention, the vibration signal is converted into the first complex spectrum signal, the running state is converted into the second complex spectrum signal, and the first complex spectrum signal and the second complex spectrum signal are fused to obtain the input signal, so that the complex spectrum signal is input as a model, the characteristic extraction capability of the noise elimination model on the input signal can be improved, and the noise reduction effect of the noise elimination model is improved; and the first complex spectrum signal and the second complex spectrum signal are fused to obtain an input signal, so that the noise reduction signal is generated based on the information of the vibration signal and the information of the running state, and a more accurate noise reduction signal is obtained, and the noise reduction effect is improved.
In order to facilitate understanding of the above embodiments, a specific embodiment is described herein. As shown in fig. 5, the noise cancellation model includes a linear feature extraction layer, a nonlinear feature extraction layer, a causal convolution layer, and a deconvolution mapping layer. Specifically, an input signal is input to a nonlinear feature extraction layer to obtain a nonlinear feature tensor output by the nonlinear feature extraction layer; inputting the input signal to a linear feature extraction layer to obtain a linear feature tensor output by the linear feature extraction layer; carrying out feature fusion on the nonlinear feature tensor and the linear feature tensor to obtain a feature tensor; inputting the feature tensor into a causal convolution layer to obtain a first target feature tensor output by the causal convolution layer; feature fusion is carried out on the first target feature tensor and the feature tensor, and a second target feature tensor is obtained; and inputting the second target characteristic tensor to the deconvolution mapping layer to obtain a noise reduction signal output by the deconvolution mapping layer.
The following describes an automobile noise cancellation device provided by the present invention, and the automobile noise cancellation device described below and the automobile noise cancellation method described above may be referred to correspondingly to each other.
Fig. 6 is a schematic structural diagram of an automotive noise cancellation device according to the present invention, as shown in fig. 6, the automotive noise cancellation device includes:
A signal determining module 610, configured to determine an input signal based on a vibration signal of a target vehicle and an operation state of the target vehicle;
a feature extraction module 620, configured to input the input signal to a feature extraction layer in a noise cancellation model, to obtain a feature tensor output by the feature extraction layer;
a signal generating module 630, configured to input the feature tensor to a signal generating layer in the noise cancellation model, so as to obtain a noise reduction signal output by the signal generating layer;
a signal output module 640 for outputting the noise reduction signal;
the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
According to the automobile noise elimination device provided by the embodiment of the invention, based on the vibration signal of the target automobile and the running state of the target automobile, the input signal is determined, so that the vibration signal is considered, the running state of the automobile is considered, the input signal is input to the feature extraction layer in the noise elimination model, the feature tensor output by the feature extraction layer is obtained, so that the features of the vibration signal are extracted, the features of the running state are also extracted, the feature tensor is input to the signal generation layer in the noise elimination model, and further a more accurate noise reduction signal is obtained, and the noise reduction effect is improved; meanwhile, the feature extraction layer comprises a nonlinear activation function layer, and the nonlinear activation function layer is used for extracting nonlinear feature tensors of input signals, namely, the features of nonlinear noise can be extracted, so that nonlinear modeling capacity is provided, the nonlinear noise can be effectively eliminated, and finally, the noise reduction effect is improved.
Based on any of the above embodiments, the signal generating module 630 is further configured to:
and inputting the characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer comprises a deconvolution layer and a nonlinear activation function layer.
Based on any of the above embodiments, the signal generating module 630 is further configured to:
inputting the characteristic tensor to a causal convolution layer in the signal generation layer to obtain a first target characteristic tensor output by the causal convolution layer;
inputting the first target feature tensor and the feature tensor 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;
and inputting the second target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer.
Based on any of the above embodiments, the deconvolution mapping layer further includes a nonlinear threshold layer, the nonlinear threshold layer being constructed based on a nonlinear threshold function as follows:
f(x)＝kx，x>0；
f(x)＝m(exp(x)-n)，x≤0；
wherein 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, exp () represents an exponential function based on a natural constant e.
Based on any of the above embodiments, the feature extraction module 620 is further configured to:
inputting the 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, wherein the nonlinear feature extraction layer comprises the nonlinear activation function layer;
inputting the input signal to a linear feature extraction layer in the feature extraction layers to obtain a linear feature tensor output by the linear feature extraction layer;
and inputting the nonlinear feature tensor and the linear feature tensor into a second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.
Based on any of the above embodiments, the nonlinear activation function layer is constructed based on the nonlinear activation function as follows:
f(x)＝sin(x)+cos(x)；
where x represents the input of the nonlinear activation function layer and f (x) represents the output of the nonlinear activation function layer.
Based on any of the above embodiments, the tag signal is a noise signal collected based on a microphone in the automobile;
the apparatus further includes a model training module to:
inputting the sample input signal to a model to be trained to obtain a sample noise reduction signal output by the model to be trained;
Performing convolution operation on the sample noise reduction signal and a preset transmission path to obtain a prediction noise reduction signal transmitted to the microphone by the sample noise reduction signal;
determining a loss value corresponding to a loss function based on the sum of the predicted noise reduction signal and the tag signal;
and training the model to be trained based on the loss value.
The invention also provides an automobile comprising a vibration sensor, a loudspeaker and a processor. The vibration sensor is used for acquiring a vibration signal of the automobile; the loudspeaker is used for outputting noise reduction signals; the processor is configured to perform the method of removing noise from an automobile of any of the embodiments described above.
Fig. 7 illustrates a physical schematic diagram of an electronic device, as shown in fig. 7, which may include: processor 710, communication interface (Communications Interface) 720, memory 730, and communication bus 740, wherein processor 710, communication interface 720, memory 730 communicate with each other via communication bus 740. Processor 710 may invoke logic instructions in memory 730 to perform a method of car noise cancellation, the method comprising: determining an input signal based on a vibration signal of a target vehicle and a running state of the target vehicle; inputting the input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer; inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer; outputting the noise reduction signal; the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
Further, the logic instructions in the memory 730 described above may be implemented in the form of software functional units and may be stored in a computer readable storage medium when sold or used as a stand alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the method of car noise cancellation provided by the above methods, the method comprising: determining an input signal based on a vibration signal of a target vehicle and a running state of the target vehicle; inputting the input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer; inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer; outputting the noise reduction signal; the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (12)
1. A method for removing noise from an automobile, comprising:
determining an input signal based on a vibration signal of a target vehicle and a running state of the target vehicle;
inputting the input signal to a feature extraction layer in a noise elimination model to obtain a feature tensor output by the feature extraction layer;
inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer;
outputting the noise reduction signal;
the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
2. The method for removing noise from an automobile according to claim 1, wherein said inputting the feature tensor into the signal generation layer in the noise removal model, to obtain the noise reduction signal of the signal generation layer output, comprises:
and inputting the characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer, wherein the deconvolution mapping layer comprises a deconvolution layer and a nonlinear activation function layer.
3. The method according to claim 2, wherein the inputting the feature tensor to the deconvolution mapping layer in the signal generation layer, to obtain the noise reduction signal output by the deconvolution mapping layer, includes:
inputting the characteristic tensor to a causal convolution layer in the signal generation layer to obtain a first target characteristic tensor output by the causal convolution layer;
inputting the first target feature tensor and the feature tensor 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;
and inputting the second target characteristic tensor to a deconvolution mapping layer in the signal generation layer to obtain a noise reduction signal output by the deconvolution mapping layer.
4. The method of claim 2, wherein the deconvolution mapping layer further comprises a non-linear threshold layer constructed based on a non-linear threshold function as follows:
f(x)＝kx，x>0；
f(x)＝m(exp(x)-n)，x≤0；
wherein 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, exp () represents an exponential function based on a natural constant e.
5. The method for eliminating noise of an automobile according to claim 1, wherein said inputting the input signal to a feature extraction layer in a noise elimination model, obtaining a feature tensor output by the feature extraction layer, comprises:
inputting the 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, wherein the nonlinear feature extraction layer comprises the nonlinear activation function layer;
inputting the input signal to a linear feature extraction layer in the feature extraction layers to obtain a linear feature tensor output by the linear feature extraction layer;
And inputting the nonlinear feature tensor and the linear feature tensor into a second feature fusion layer in the feature extraction layer to obtain the feature tensor output by the second feature fusion layer.
6. The method of claim 1, 2 or 5, wherein the nonlinear activation function layer is constructed based on the nonlinear activation function:
f(x)＝sin(x)+cos(x)；
where x represents the input of the nonlinear activation function layer and f (x) represents the output of the nonlinear activation function layer.
7. The method of any one of claims 1 to 5, wherein the tag signal is a noise signal collected based on a microphone in the vehicle;
the noise cancellation model is trained based on the following:
inputting the sample input signal to a model to be trained to obtain a sample noise reduction signal output by the model to be trained;
performing convolution operation on the sample noise reduction signal and a preset transmission path to obtain a prediction noise reduction signal transmitted to the microphone by the sample noise reduction signal;
determining a loss value corresponding to a loss function based on the sum of the predicted noise reduction signal and the tag signal;
And training the model to be trained based on the loss value.
8. The method of claim 1, wherein determining the input signal based on the vibration signal of the target vehicle and the operating state of the target vehicle comprises:
converting the vibration signal into a first complex spectrum signal and converting the operating state into a second complex spectrum signal;
and fusing the first complex spectrum signal and the second complex spectrum signal to obtain an input signal.
9. An automotive noise abatement device, comprising:
the signal determining module is used for determining an input signal based on a vibration signal of a target automobile and the running state of the target automobile;
the feature extraction module is used for inputting the input signal to a feature extraction layer in the noise elimination model to obtain a feature tensor output by the feature extraction layer;
the signal generation module is used for inputting the characteristic tensor into a signal generation layer in the noise elimination model to obtain a noise reduction signal output by the signal generation layer;
the signal output module is used for outputting the noise reduction signal;
the characteristic extraction layer comprises a nonlinear activation function layer, wherein the nonlinear activation function layer is used for extracting nonlinear characteristic tensors of the input signals; the noise cancellation model is trained based on a sample input signal and a tag signal corresponding to the sample input signal, the sample input signal being determined based on a sample vibration signal and a sample operating state.
10. An automobile, comprising:
the vibration sensor is used for acquiring a vibration signal of the automobile;
a speaker for outputting a noise reduction signal;
a processor for performing the method of automotive noise cancellation of 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, wherein the processor implements the method of car noise cancellation as claimed in any one of claims 1 to 8 when the program is executed by the processor.
12. A non-transitory computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the car noise cancellation method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311629578.9A CN117765912A (en) | 2023-11-30 | 2023-11-30 | Automobile noise elimination method and device, automobile, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311629578.9A CN117765912A (en) | 2023-11-30 | 2023-11-30 | Automobile noise elimination method and device, automobile, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117765912A true CN117765912A (en) | 2024-03-26 |
Family
ID=90309604
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311629578.9A Pending CN117765912A (en) | 2023-11-30 | 2023-11-30 | Automobile noise elimination method and device, automobile, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117765912A (en) |
-
2023
- 2023-11-30 CN CN202311629578.9A patent/CN117765912A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10013967B2 (en) | Method and system for selecting sensor locations on a vehicle for active road noise control | |
| CN105374365A (en) | System and method for controlling vehicle noise | |
| Duan et al. | Combined feedforward–feedback active control of road noise inside a vehicle cabin | |
| US8688309B2 (en) | Active and stateful hyperspectral vehicle evaluation | |
| CN111833841A (en) | Active control system and method for automobile road noise and vehicle system | |
| US20170213550A1 (en) | Adaptive dual collaborative kalman filtering for vehicular audio enhancement | |
| Park et al. | Coherence-based sensor set expansion for optimal sensor placement in active road noise control | |
| JP2006313997A (en) | Noise level estimating device | |
| CN109600696A (en) | System for the frequency spectrum shaping that vehicle noise is eliminated | |
| CN116312448A (en) | Reference sensing optimization method of road noise active control system | |
| Chen et al. | A low-complexity multi-channel active noise control system using local secondary path estimation and clustered control strategy for vehicle interior engine noise | |
| CN117765912A (en) | Automobile noise elimination method and device, automobile, electronic equipment and storage medium | |
| CN111833840B (en) | Noise reduction method, noise reduction device, noise reduction system, electronic equipment and storage medium | |
| Zhou et al. | A Modified Reference Signal Selection Method for Feed-Forward Active Road Noise Control System | |
| Wang et al. | Discrete wavelet transfom for nonstationary signal processing | |
| JP2022156359A (en) | Measurement method of transfer function and active noise reduction device | |
| Mashiana et al. | Speech enhancement using residual convolutional neural network | |
| Lucas et al. | From Test Rig to Road–Experience Virtual Vehicle Acoustics on Powertrain Test Bench BTPA | |
| JP2006084928A (en) | Sound input device | |
| CN112334971A (en) | Concurrent FXLMS system using common reference signal and error signal | |
| CN113393858B (en) | Voice separation method and system, electronic equipment and readable storage medium | |
| Costes et al. | Simulation of Automotive loudspeaker integration for the full auditory spectrum | |
| CN118335047A (en) | Active noise reduction method and device and electronic equipment | |
| Cheng et al. | An optimal sensor layout method based on noise reduction estimation for active road noise control | |
| Mostafavi et al. | Feedback Active Noise Cancellation Using Single Sensor with Deep Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |