Filter coefficient matching model training method, active noise reduction method and system

Abstract

The invention relates to the technical field of audio processing, in particular to a filter coefficient matching model training method, an active noise reduction method and an active noise reduction system. The training method comprises the steps that a plurality of tasks are obtained, each task comprises a support set and a query set, the support set comprises a plurality of support samples from a plurality of noise categories and corresponding optimal filter coefficient labels, and the query set comprises a plurality of query samples from the same noise categories; maximizing the probability of matching each query sample to a target label as a training target, and constructing a target training function, the target label being an optimal filter coefficient label corresponding to a support sample belonging to the same category as the query sample; and performing iterative training on the matching network according to the plurality of tasks and the target training function to obtain a filter coefficient matching model. By adopting the method, the noise reduction effect can be improved.

Description

Technical Field

[0001] The present application relates to the field of audio processing technology, and in particular to a training method for a filter coefficient matching model, an active noise reduction method and system. Background Art

[0002] Active Noise Cancellation (ANC) is a noise cancellation method based on the principle of acoustic signal superposition. Its purpose is to generate an anti-noise signal with the same amplitude and opposite phase as the noise signal to eliminate or attenuate the noise signal.

[0003] Related technologies propose a selective fixed-filtered active noise cancellation (SFANC) method based on convolutional neural networks. The SFANC method uses traditional deep learning methods and conducts targeted training based on a large number of training samples, ultimately matching the best filter coefficients according to the noise type to solve the noise problem in a specific scenario. The premise for its applicability is that the sample distribution can well represent the overall distribution.

[0004] However, the actual application scenarios are very complex, and the noise signals generated in different application scenarios are various. For example, in the vehicle active noise reduction scenario, the noise signal depends on different influencing factors such as vehicle condition, driving habits and road conditions. Therefore, when there is a significant difference between the online noise and the training sample, the SFANC method may not be able to match the optimal filter coefficient, resulting in poor noise reduction effect. Summary of the invention

[0005] Based on this, a training method for a filter coefficient matching model, an active noise reduction method and a system are provided to improve the problem that the related SFANC method cannot match the optimal filter coefficient when there is a significant difference between the online noise and the training sample, so as to improve the noise reduction effect.

[0006] In a first aspect, a method for training a filter coefficient matching model is provided, the method comprising:

[0007] Acquire multiple tasks, each of the tasks includes a support set and a query set, the support set includes multiple support samples from multiple noise categories and their corresponding optimal filter coefficient labels, and the query set includes multiple query samples from the same multiple noise categories;

[0008] Taking maximizing the probability of each query sample matching a target label as a training objective, constructing a target training function, wherein the target label is the optimal filter coefficient label corresponding to the support sample belonging to the same category as the query sample;

[0009] The matching network is iteratively trained according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

[0010] In combination with the first aspect, in a first implementable embodiment of the first aspect, before iteratively training the matching network according to the multiple tasks and the target training function, the method further includes:

[0011] Constructing a first feature extraction subnetwork, wherein the first feature extraction subnetwork is used to extract implicit features of each of the supporting samples;

[0012] Constructing a second feature extraction subnetwork, where the second feature extraction subnetwork is used to extract implicit features of each of the query samples;

[0013] Constructing an attention score calculation subnetwork, wherein the attention score calculation subnetwork is used to calculate the probability that each query sample matches the target label according to the implicit features of each query sample and the implicit features of each support sample;

[0014] The matching network is obtained based on the first feature extraction subnetwork, the second feature extraction subnetwork and the attention score calculation subnetwork.

[0015] In combination with the first possible implementation of the first aspect, in a second possible implementation of the first aspect, an attention score calculation subnetwork is constructed, including:

[0016] Constructing a similarity calculation function, wherein the similarity calculation function is used to calculate the similarity between each query sample and each supporting sample according to the implicit features of each query sample and the implicit features of each supporting sample;

[0017] Constructing an attention weight calculation function, wherein the attention weight calculation function is used to assign a corresponding attention weight according to the similarity between each query sample and each support sample;

[0018] Constructing an attention score calculation function, wherein the attention score calculation function is used to obtain the probability of each query sample matching the target label according to the weighted sum of each attention weight and each optimal filter coefficient label;

[0019] The attention score calculation subnetwork is obtained according to the similarity calculation function, the attention weight calculation function and the attention score calculation function.

[0020] In combination with the first aspect, in a third implementable embodiment of the first aspect, before acquiring the plurality of tasks, the method further includes:

[0021] Building a training database according to a preset sound effect library and a preset noise database, wherein the preset sound effect library includes a variety of non-speech environmental noises, and the preset noise database includes background noises in a variety of environments;

[0022] The build steps for each of these tasks include:

[0023] Selecting multiple noise categories from the training database, and selecting multiple noise samples from the noise samples corresponding to the multiple noise categories as corresponding support samples to form the support set, and selecting multiple noise samples from the remaining samples corresponding to the same multiple noise categories as corresponding query samples to form the query set;

[0024] Querying each of the supporting samples in a preset filter coefficient database to obtain a corresponding optimal filter coefficient and generate a corresponding optimal filter coefficient label;

[0025] The task is constructed according to the support set and the query set.

[0026] In combination with the third implementable embodiment of the first aspect, in a fourth implementable embodiment of the first aspect, before querying each of the supporting samples in a preset filter coefficient database, the method further includes:

[0027] In the simulation experiment, multiple Gaussian noises of different frequencies are obtained;

[0028] Determine an interference signal generated by each Gaussian noise passing through a main path, wherein the main path represents a path from a reference microphone to an error microphone;

[0029] Based on each of the Gaussian noises and each of the interference signals, an adaptive filtering algorithm is used to obtain an optimal filter coefficient corresponding to each of the Gaussian noises;

[0030] The filter coefficient database is constructed according to the optimal filter coefficients corresponding to each of the Gaussian noises.

[0031] In combination with the fourth implementable embodiment of the first aspect, in a fifth implementable embodiment of the first aspect, determining the interference signal generated by each Gaussian noise passing through the main path includes:

[0032] In a given noise reduction environment, acoustic signal tests are performed at the reference microphone and the error microphone respectively to obtain the corresponding noise source signal and the signal at the first human ear;

[0033] Determining a main path transfer function according to the noise source signal and the first human ear signal;

[0034] A corresponding interference signal is obtained according to the main path transfer parameter and each Gaussian noise.

[0035] In combination with the fourth implementable embodiment of the first aspect, in a sixth implementable embodiment of the first aspect, based on each of the Gaussian noises and each of the interference signals, an adaptive filtering algorithm is used to obtain an optimal filter coefficient corresponding to each of the Gaussian noises, including:

[0036] In a given noise reduction environment, acoustic signal tests are performed at the cancellation speaker and the error microphone respectively to obtain the corresponding noise reduction signal and the signal at the second human ear;

[0037] Determine a secondary path transfer function according to the noise reduction signal and the second human ear signal, wherein the secondary path represents a path from the cancellation speaker to the error microphone;

[0038] Determine a corresponding anti-noise signal according to each of the interference signals;

[0039] Based on each of the Gaussian noises, the secondary path transfer function, the preset cancellation speaker transfer function and each of the anti-noise signals, an adaptive filtering algorithm is used to obtain the optimal filter coefficient corresponding to each of the Gaussian noises.

[0040] In a second aspect, an active noise reduction method is provided, the method comprising:

[0041] Acquire an online support set, wherein the online support set includes a plurality of real noises from an actual vehicle use environment and their corresponding optimal filter coefficients;

[0042] Collecting online noise in the current vehicle environment, inputting the online support set and the online noise into a trained filter coefficient matching model for processing to obtain a current optimal filter coefficient, wherein the trained filter coefficient matching model is obtained according to the training method for the filter coefficient matching model described in any one of the first embodiment to the sixth embodiment of the first aspect;

[0043] The online noise is reduced according to the current optimal filter coefficient.

[0044] In a third aspect, a training system for a filter coefficient matching model is provided, the system comprising:

[0045] An acquisition module, configured to acquire a plurality of tasks, each of the tasks comprising a support set and a query set, the support set comprising a plurality of support samples from a plurality of noise categories and their corresponding optimal filter coefficient labels, and the query set comprising a plurality of query samples from the same plurality of noise categories;

[0046] A construction module is used to take maximizing the probability of each query sample matching a target label as a training goal and construct a target training function, wherein the target label is an optimal filter coefficient label corresponding to a support sample belonging to the same category as the query sample;

[0047] A training module is used to iteratively train the matching network according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

[0048] In a fourth aspect, an active noise reduction system is provided, the system comprising:

[0049] A preprocessing module, used to obtain an online support set, wherein the online support set includes a plurality of real noises from an actual vehicle environment and their corresponding optimal filter coefficients;

[0050] a matching module, configured to collect online noise in a current vehicle-using environment, and input the online support set and the online noise into a trained filter coefficient matching model for processing to obtain a current optimal filter coefficient, wherein the trained filter coefficient matching model is obtained according to the training method for the filter coefficient matching model described in any one of the first embodiment to the sixth embodiment of the first aspect;

[0051] A noise reduction module is used to reduce the online noise according to the current optimal filter coefficient.

[0052] The training method, active noise reduction method and system of the filter coefficient matching model are obtained by acquiring multiple tasks, each task includes a support set and a query set, the support set includes multiple support samples from multiple noise categories and their corresponding optimal filter coefficient labels, and the query set includes multiple query samples from the same multiple noise categories; the probability of maximizing the matching of each query sample to the target label is used as the training goal, and a target training function is constructed, and the target label is the optimal filter coefficient label corresponding to the support sample belonging to the same category as the query sample; the matching network is iteratively trained according to the multiple tasks and the target training function to obtain the filter coefficient matching model. The filter coefficient matching model can be applied to match the current optimal filter coefficient with the best online noise reduction effect according to multiple real noises from the actual vehicle environment and matched with the optimal filter coefficient and the online noise from the current vehicle environment. The training method of the filter coefficient matching model of the present application uses a matching network for training. When training the model, there is no need to collect training samples that can cover all actual active noise reduction scenarios. The matching network calculates the similarity between the query sample and each support sample in the support set, thereby classifying the query samples and support samples of the same category into the same category, and then using the same optimal filter coefficient for the query sample, which allows the trained model to be instantly applied to noise signals that have not been encountered before, without the need for additional training for each new category of noise signal. Therefore, compared with related technologies, the present application can match the optimal filter coefficient when there is a significant difference between the actual noise and the training sample, thereby improving the noise reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 FIG. 1 is an application environment diagram of a training method for a filter coefficient matching model in an embodiment;

[0054] Figure 2 A schematic diagram of a flow chart of a method for training a filter coefficient matching model in one embodiment;

[0055] Figure 3 A schematic diagram of a process of extracting hidden features using a one-dimensional convolutional recurrent network in one embodiment;

[0056] Figure 4 is a structural block diagram of a typical active noise reduction system in an embodiment;

[0057] Figure 5 is a schematic diagram of a process of an active noise reduction method in an embodiment;

[0058] Figure 6 is a structural block diagram of a training system for a filter coefficient matching model in one embodiment;

[0059] Figure 7 is a structural block diagram of an active noise reduction system in one embodiment;

[0060] Figure 8 is an internal structure diagram of a computer device in one embodiment;

[0061] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] It should be noted that the illustrations provided in this embodiment only illustrate the basic concept of the present application in a schematic manner, and therefore the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0064] The structures, proportions, sizes, etc. illustrated in the drawings in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which this application can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by this application, should still fall within the scope of the technical contents disclosed in this application.

[0065] The directions or positional relationships indicated by the terms "upper", "lower", "left", "right", "middle", "longitudinal", "lateral", "horizontal", "inner", "outer", "radial", "circumferential" and the like in this specification are based on the directions or positional relationships shown in the drawings and are only for the convenience of simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore cannot be understood as limiting the present application. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0066] In feedforward active noise cancellation (ANC) systems, adaptive filters are usually used to minimize the error signal in order to adjust and optimize the noise cancellation effect in real time. The FxLMS algorithm is a widely used adaptive filtering algorithm that compensates for the secondary path delay and achieves optimal noise control around the error microphone. However, adaptive filtering methods represented by the FxLMS algorithm have the following limitations: when the secondary path changes greatly or the system output is saturated due to excessive noise levels, such methods may lose stability; the slow convergence of the adaptive algorithm limits the effect of dynamic noise and overall noise reduction; traditional adaptive filtering algorithms need to run at a higher sampling frequency, which requires higher processor performance.

[0067] On the contrary, the active noise reduction method using fixed filters does not have the above problems and has become another widely used active noise reduction method. Among them, the selective fixed-filter active noise reduction method (Selective Fixed-filtered Active Noise Cancellation, SFANC) can select the filter coefficient with the best noise reduction effect for different noises. The related SFANC method uses a noise data set of a general scene to purposefully train the neural network to solve the filter coefficient matching problem of a specific scene; however, this method has a major limitation, that is, in a new, complex scene that is different from the distribution of training samples, the accuracy of selecting filter coefficients decreases, resulting in poor noise reduction effect. The premise for the application of this method is that the samples meet the assumption of independent and identical distribution, that is, the sample distribution can well represent the overall distribution.

[0068] However, the actual application scenarios are very complex, and the noise signals generated in different application scenarios are also different. Taking the vehicle active noise reduction scenario as an example, the actual driving environment is very complex, and the noise signal depends on different influencing factors such as vehicle condition, driving habits and road conditions. In the actual noise reduction scenario, it is very likely to encounter a new, complex scene that is completely different from the training sample distribution. At this time, this method is difficult to match the filter coefficients with the best noise reduction effect, resulting in poor noise reduction effect.

[0069] To this end, this application proposes a training method for a filter coefficient matching model, which can be applied to Figure 1In the application environment shown, the vehicle end 102 communicates with the server 104 through a network. The server 104 obtains multiple tasks, each of which includes a support set and a query set, wherein the support set includes multiple support samples from multiple noise categories and their corresponding optimal filter coefficient labels, and the query set includes query samples from the same multiple noise categories; the probability of maximizing each query sample matching the target label is used as the training goal, and a target training function is constructed, wherein the target training is the optimal filter coefficient label corresponding to the support sample belonging to the same category as the query sample; the matching network is iteratively trained according to the multiple tasks and the target training function to obtain a filter coefficient matching model. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0070] Unlike traditional deep learning methods, this application uses a matching network for training, and there is no need to retrain for each new category using a large amount of data. It finds the most similar support samples between the query sample and the support set, and classifies them according to the noise categories to which these similar samples belong, and uses the same optimal filter coefficients for samples of the same category. This mechanism helps to improve the model's generalization ability for unknown noise, and even when the distribution of noise categories changes, such as when new noise categories are generated, it can maintain good filter coefficient matching performance and ensure better noise reduction effects.

[0071] Next, the training method of the filter coefficient matching model of the present application will be described in detail through the following embodiments.

[0072] In one embodiment, Figure 2 As shown, a training method for a filter coefficient matching model is provided, and the method is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0073] Step 202, obtaining multiple tasks, each of which includes a support set and a query set, the support set includes multiple support samples from multiple noise categories and their corresponding optimal filter coefficient labels, and the query set includes multiple query samples from the same multiple noise categories.

[0074] Since the actual application scenarios of active noise reduction are very complex, the noise generated by different application scenarios is also different. For example, in the vehicle active noise reduction scenario, the noise signal depends on different influencing factors such as vehicle condition, driving habits and road conditions. Moreover, when training the model, it is difficult to collect noise that can cover all actual application scenarios as training samples due to the difficulty of sample collection and high testing costs. Therefore, the supporting samples and query samples of this application are not limited to the noise of certain specific scenarios. In some embodiments, a variety of non-speech environmental noises from the preset sound effect library and / or background noises from a variety of environments in the preset noise database are used.

[0075] In a specific implementation, a training data set is constructed based on a preset sound effect library and a preset noise database; multiple categories are randomly selected from the training data set, multiple noises are randomly selected from each category, and these noises are respectively used as support samples to construct a support set; from the remaining noises of the same multiple categories, multiple noises are randomly selected from each category, and these noises are respectively used as query samples to construct a query set; since the matching network is trained in tasks, the support set and the query set are taken as a whole to construct a task; the above steps of constructing tasks are repeated multiple times to obtain multiple tasks, and the multiple tasks form a task set.

[0076] For any noise, the test can determine which optimal filter coefficient can achieve the best noise reduction effect. Therefore, for each supporting sample, in a simulation test environment, such as a Matlab simulation test environment, the supporting sample is input as a reference signal, and the corresponding optimal filter coefficient is queried from the pre-built filter coefficient database, and the optimal filter coefficient label corresponding to the supporting sample is generated based on the optimal filter coefficient obtained by the query. Among them, the filter coefficient database includes the optimal filter coefficients corresponding to multiple Gaussian noises of different frequencies, and the optimal filter coefficient refers to the filter coefficient with the best noise reduction effect on the supporting sample.

[0077] Step 204: Taking maximizing the probability of each query sample matching a target label as a training objective, a target training function is constructed, wherein the target label is an optimal filter coefficient label corresponding to a support sample belonging to the same category as the query sample.

[0078] Since each supporting sample is a noise of a different noise category and has been marked with the best filter coefficient label, each query sample is a noise of a different noise category but has not been marked with the best filter coefficient label; and the working principle of the matching network is: by calculating the similarity between the query sample and each supporting sample, the attention score of the best filter coefficient label corresponding to the query sample and each supporting sample is calculated according to each similarity. The attention score represents the probability that the best filter coefficient label corresponding to the query sample and the support sample matches each other. The highest attention score is obtained by comparing the attention scores, and the best filter coefficient corresponding to the support sample corresponding to the highest attention score is matched as the target label of the query sample.

[0079] In the above process, when the similarity between the query sample and the support sample of the same category is higher, the corresponding attention score is higher, that is, the probability that the query sample matches the target label is higher, for example, it is constantly tending to 1; on the contrary, when the similarity between the query sample and the support sample of the same category is lower, the corresponding attention score is lower, that is, the probability that the query sample matches the non-target label is lower, for example, it is constantly tending to 0, then the model's optimal filter coefficient matching accuracy for noise is higher. Therefore, this application takes maximizing the probability of each query sample matching the target label as the training goal, constructs a target training function, and iteratively trains the matching network with the target training function to obtain the optimal filter coefficient matching model.

[0080] Step 206: Iteratively train the matching network according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

[0081] The supporting samples and query samples of each task are input into the matching network, and the matching network is iteratively trained using the target training function to obtain the optimal model parameters, and the optimal model parameters are assigned to the matching network to obtain a trained filter coefficient matching model. The filter coefficient matching model is used to match the current optimal filter coefficient with the best online noise reduction effect based on multiple real noises from actual noise reduction scenarios that have been matched with the optimal filter coefficients and the online noise from the current noise reduction scenario.

[0082] Taking the vehicle active noise reduction scenario as an example, although it is impossible to collect a large amount of real noise from different actual vehicle use environments, a small amount of real noise from different categories can be collected during the test drive or internal testing process; for these small amounts of real noise, an optimal filter coefficient with the best noise reduction effect is selected from a pre-constructed filter coefficient database through manual matching or automatic matching; based on these real noises and their corresponding optimal filter coefficients, a support set in an online environment (hereinafter referred to as the online support set) is constructed; the online noise in the current vehicle use environment and each real noise in the online support set and its corresponding optimal filter coefficient are input into the optimal filter coefficient for processing to obtain the current optimal filter coefficient with the best noise reduction effect on the online noise; the filter is configured according to the current optimal filter coefficient, and the online noise is reduced according to the configured filter.

[0083] The training method of the filter coefficient matching model mentioned above uses a matching network for training. During training, there is no need to collect training samples that can cover all active noise reduction scenarios. Instead, by finding support samples of the same category as the query sample from the support set, the same optimal filter coefficient is used for the query sample. This allows the trained model to be instantly applied to new categories of noise signals without the need for additional training for each new category of noise signal. Therefore, compared with the prior art, the present application can match the optimal filter coefficient when there is a significant difference between the actual noise and the training sample, thereby ensuring the noise reduction effect.

[0084] Next, a complete filter coefficient matching model training process will be explained through the following example.

[0085] Constructing a data set: constructing a data set according to a preset sound effect library and a preset noise database. The preset sound effect library includes a variety of non-speech environmental noises, and the preset noise database includes background noises in various environments. For example, the sound effect library disclosed at the website http: / / www.sound-ideas.com records more than 10,000 non-speech environmental noises, and NOISE-92 records engine noise, factory noise, speech spectrum noise (Speech Spectrum-Shaped Noise, SSN) and multi-person conversation noise (Babble Noise, BN); dividing the data set into a training data set, a verification data set and a test data set according to the ratio of 70%, 15% and 15%; in a Matlab simulation test environment, inputting each noise data in the data set as a reference signal, querying in a pre-constructed filter coefficient database, obtaining the best filter coefficient with the best noise reduction effect and generating the corresponding best filter coefficient label, and finally constructing a supervised data set.

[0086] Construct the task set: randomly extract C noise categories from the training data set, with m noise samples for each noise category, for a total of C*m noise samples, and use these C*m noise samples as corresponding support samples to form a support set; randomly extract n noise samples from the remaining noise samples corresponding to the same C noise categories for each noise category, for a total of C*n noise samples, and use these C*n noise samples as corresponding query samples to form a query set; construct the task by taking the support set and the query set as a whole; repeat the above steps N times to obtain N tasks, and the N tasks constitute the task set.

[0087] Construct matching network: Construct the first feature extraction subnetwork, which is used to extract the implicit features g(x i );Construct the second feature extraction subnetwork, which is used to extract the implicit features of each query sample Construct an attention score calculation subnetwork, which is used to calculate the implicit features of each query sample and the implicit features g(x i ), calculate the attention score of each query sample and the target label The attention score It represents the probability that the query sample matches the target label given the support set, where y represents the target label and S represents the support set formed by each support sample.

[0088] For the step of constructing the first feature extraction subnetwork and the second feature extraction subnetwork, illustratively, a one-dimensional convolutional neural network (1D CNN) is used to construct the first feature extraction subnetwork and the second feature extraction subnetwork; the one-dimensional convolutional neural network can directly process the normalized original noise waveform and strike a balance between computational complexity and model performance, such as Figure 3As shown in the figure, the process of extracting hidden features includes: the input layer receives the noise signal collected by the reference microphone; the noise signal is normalized to the minimum or maximum; the normalized noise signal is input into a 1x5 convolution layer with 10 channels; the upper layer output is connected to a 1x32 convolution layer with 20 channels, and then connected to a maximum pooling layer with a size of 512; the upper layer output is connected to a flatten layer and converted into a one-dimensional array form; the upper layer output is connected to a batch normalization layer, and then connected to a fully connected layer with 32 neurons, and then connected to a hyperbolic tangent layer. The output of the upper layer is connected to the batch normalization layer, and then connected to the fully connected layer of 64 neurons, and then connected to the Tanh activation function; the output of the upper layer is connected to the batch normalization layer, and then connected to the fully connected layer of 128 neurons, and then connected to the Tanh activation function; the output of the upper layer is connected to the batch normalization layer, and then connected to the fully connected layer of 256 neurons, and connected to the Tanh activation function; the output of the upper layer is connected to a fully connected layer of 256 neurons to obtain the implicit features of the noise signal.

[0089] The step of constructing the attention score calculation subnetwork specifically includes: constructing a similarity calculation function, which is used to calculate the similarity based on the implicit features of each query sample. and the implicit features g(x i ), calculate the similarity between each query sample and each supporting sample, for example, cosine similarity Construct an attention weight calculation function, which is used to assign corresponding attention weights according to the similarity between each query sample and each supporting sample Exemplarily, the attention weight is calculated using the following mathematical expression: k represents the total number of supporting samples, i and j represent the indexes of supporting samples; construct an attention score calculation function, which is used to calculate the attention score according to the attention weights and the labels y for the best filter coefficients i The weighted sum of is used to obtain the probability that each query sample matches the target label. The mathematical expressions used include: The attention scores are compared to get the highest attention score, and the best filter coefficient label corresponding to the support sample corresponding to the highest attention score is matched as the target label of the query sample.

[0090] Among them, y iis a matrix. Given a support set, the values ​​of each bit in the matrix depend on the attention score of the query sample and the best filter coefficient label of the support sample in the support set. For example, assuming that the given support set includes 4 support samples, when calculating the attention score of the query sample and the best filter coefficient label of the first support sample, y i =(1 0 0 0), when calculating the attention score of the best filter coefficient label of the query sample and the second support sample, y i =(0 1 0 0), and the other cases are similar and are not listed here exhaustively.

[0091] Constructing the target training function: We hope to train a model that can: randomly extract multiple tasks from the task set, then randomly extract each task multiple times from the samples of the support set and the query set, and input these samples into the matching network. In general, the probability that each query sample matches the target label can be maximized. Therefore, according to the training goal of maximizing the probability that each query sample matches the target label, the target training function is constructed.

[0092] Exemplary: randomly extract N tasks from the task set; randomly extract several different noise categories from the support set for each task, extract a support sample from each category, extract several identical categories from the query set, extract a query sample from each category, several support samples and several query samples constitute a batch, and each task is extracted M times to form M batches; for each batch, the matching network can only match one query sample with each support sample at a time, therefore, poll each query sample, input each query sample and each support sample into the matching network respectively, and obtain the probability of each query sample matching the target label; add the probabilities of each query sample matching the target label under the same batch to obtain the probability corresponding to a batch; calculate the mathematical expectation based on the probabilities corresponding to each batch of a task to obtain the probability corresponding to a task; calculate the mathematical expectation based on the probabilities corresponding to several tasks to obtain the final probability; in order to maximize the probability of query samples matching the target label in general, take the maximization of the final probability as the training goal, and construct a target training function, whose expression includes:

[0093]

[0094] Among them, argmax represents the maximization operator, E represents the mathematical expectation operator, L represents the task, T represents the task set, S represents the support set, N represents the total number of tasks extracted from the task set, M represents the number of batches, B represents a batch, and E represents the number of tasks extracted from the task set. L～T represents the expected probability of randomly selecting N tasks from the task set, E S～L,B～L represents the expected probability of randomly sampling support and batches from the task.

[0095] In some implementations, the automatic gradient calculation function provided by the Pytorch framework is used, the Adam optimizer is used, the learning rate is set to 0.001, and the model gradient is automatically updated; the epochs is set to 50, and if the current epochs reaches 50, the model is considered to have converged and the training is completed.

[0096] In some embodiments, constructing a filter coefficient database includes the following steps: in a simulation experiment, obtaining multiple Gaussian noises of different frequencies; determining the interference signal generated by each Gaussian noise through a main path, where the main path represents the path from a reference microphone to an error microphone; based on each Gaussian noise and each interference signal, using an adaptive filtering algorithm to obtain the optimal filter coefficient corresponding to each Gaussian noise; and constructing a filter coefficient database based on the optimal filter coefficient corresponding to each Gaussian noise.

[0097] For example, Figure 4 The figure shows a typical block diagram of a feedforward ANC system, including a reference microphone, an error microphone, an active noise reduction system, and a cancellation speaker. In the Matlab simulation experiment, it is assumed that the frequency interval of the Gaussian noise is [0.02, 7.08] kHz. A Gaussian noise of a frequency is randomly extracted from the interval. The Gaussian noise is used as a reference signal x(n) and is transmitted from the main path to the error microphone to generate an interference signal d(n). The Gaussian noise is adaptively filtered by the active noise reduction system and then transmitted to the error microphone through the secondary path to generate an anti-noise signal a(n).

[0098] The interference signal d(n) is obtained by convolving the reference signal with the main path transfer function p(n), and its mathematical expression includes: d(n)=x(n)*p(n); the anti-noise signal is obtained by filtering the Gaussian noise through the active noise reduction system, collecting it through the elimination speaker, and then convolving it with the secondary path transfer function s(n), and its mathematical expression includes: a(n)=s(n)*f LS {w(n) T x(n)}, f LS The symbol of the transfer function for canceling the loudspeaker, w(n) T represents the matrix transpose of the filter coefficients of the active noise reduction system; e(n) is the residual noise measured by the error microphone, e(n) = d(n) - a(n);

[0099] When the anti-noise signal and the interference signal are completely offset, that is, when the anti-noise signal is a signal with the same amplitude and opposite phase as the interference signal, the active noise reduction system has the best noise reduction performance; therefore, through this relationship, the corresponding anti-noise signal can be determined according to the interference signal; based on Gaussian noise, secondary path transfer function, eliminated speaker transfer function and anti-noise signal, an adaptive filtering algorithm, such as the LMS algorithm, is used to reverse the mathematical expression for calculating the anti-noise signal, and the optimal filter coefficient with the best noise reduction effect on Gaussian noise can be obtained; the above steps are repeated multiple times to obtain the optimal filter coefficients corresponding to multiple Gaussian noises, thereby constructing a filter coefficient database.

[0100] In the above process, the main path transfer function and the secondary path transfer function need to be determined in advance. Figure 4 The structural block diagram shown is used to illustrate that the main path transfer function is determined by the physical characteristics of the path from the reference microphone to the error microphone. In a given noise reduction environment, such as a vehicle noise reduction environment, acoustic signal tests are performed at the reference microphone and the error microphone, respectively, to obtain the corresponding noise source signal and the first ear signal; based on the relationship that the first ear signal is obtained after the convolution operation between the noise source signal and the main path transfer function, the main path transfer function is reversely derived.

[0101] Similarly, the secondary path is determined by the physical properties of the path from the cancellation speaker to the error microphone. In a given noise reduction environment, such as a vehicle noise reduction environment, acoustic signal tests are performed at the cancellation speaker and the error microphone, respectively, to obtain the corresponding noise reduction signal and the signal at the second ear. The secondary path transfer function is reversely derived based on the relationship between the noise reduction signal and the convolution operation of the secondary path transfer function to obtain the signal at the second ear.

[0102] In one embodiment, Figure 5 As shown, an active noise reduction method is provided, which is applied to Figure 1 Taking the vehicle end 102 in the example as an example, the method includes the following steps:

[0103] Step 302 : obtaining an online support set, where the online support set includes a plurality of real noises from an actual vehicle use environment and their corresponding optimal filter coefficients.

[0104] In the vehicle active noise reduction scenario, the noise signal depends on different influencing factors such as vehicle condition, driving habits and road conditions. Although it is impossible to collect a large amount of real noise from different actual vehicle use environments, a small amount of real noise from different categories can be collected during test drives or internal tests. For these small amounts of real noise, an optimal filter coefficient with the best noise reduction effect is selected from a pre-built filter coefficient database through manual matching or automatic matching. Based on these real noises and their corresponding optimal filter coefficients, an online support set is constructed.

[0105] Step 304 , collecting online noise in the current vehicle environment, inputting the online support set and the online noise into a trained filter coefficient matching model for processing, and obtaining the current optimal filter coefficient.

[0106] The trained filter coefficient matching model is obtained according to the training method of the filter coefficient matching model described in the above embodiment. Since each real noise in the online support set has been matched with the optimal filter coefficient, each real noise in the online support set and the online noise are input into the filter coefficient matching model, and the same category of real noise can be matched for the online noise, so that the same optimal filter coefficient is used for the online noise.

[0107] In some embodiments, when there are many real noises in the online support set, for example, when there is more than one real noise applicable to each optimal filter coefficient, the real noise in the online support set is classified according to the category of the optimal filter coefficient, and the real noise using the same optimal filter coefficient is classified into the same category; when in use, a real noise and its corresponding optimal filter coefficient are randomly selected from each category in the online support set, and the real noises from different categories and their corresponding optimal filter coefficients and the online noise are input into the filter coefficient matching network for processing, so as to screen out the real noise most similar to the online noise, and then use the same optimal filter coefficient for the online noise, so as to avoid matching the online noise with multiple real noises of the same category and not matching with real noises of different categories, thereby improving the adequacy of noise matching, thereby improving the accuracy of filter coefficient matching, and ensuring the noise reduction effect.

[0108] Step 306: Reduce the online noise according to the current optimal filter coefficient.

[0109] The filter coefficient of the filter is configured as the current optimal filter coefficient, and then the online noise is reduced according to the configured filter. Among them, the above-mentioned online support set is only sent to the vehicle side, and only a small number of noise samples and their labels are added or updated. There is no need to update the parameters of the filter coefficient matching model, and there is no need to perform additional training on the model. This can not only alleviate the overfitting phenomenon of the model, greatly expand the actual application scenarios of the model, but also enable the model to be immediately applied to new categories of noise signals. Therefore, compared with the prior art, the present application can match the optimal filter coefficient when there is a significant difference between the actual noise and the training sample, thereby improving the noise reduction effect.

[0110] It should be understood that although Figure 2 , 5 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 , 5 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0111] In one embodiment, Figure 6 As shown, a training system for a filter coefficient matching model is provided, including: an acquisition module, a construction module and a training module, wherein:

[0112] An acquisition module, configured to acquire a plurality of tasks, each of the tasks comprising a support set and a query set, the support set comprising a plurality of support samples from a plurality of noise categories and their corresponding optimal filter coefficient labels, and the query set comprising a plurality of query samples from the same plurality of noise categories;

[0113] A construction module is used to take maximizing the probability of each query sample matching a target label as a training goal and construct a target training function, wherein the target label is an optimal filter coefficient label corresponding to a support sample belonging to the same category as the query sample;

[0114] A training module is used for iteratively training a matching network according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

[0115] In some embodiments, the construction module is also used to: construct a first feature extraction subnetwork, the first feature extraction subnetwork is used to extract the implicit features of each of the supporting samples; construct a second feature extraction subnetwork, the second feature extraction subnetwork is used to extract the implicit features of each of the query samples; construct an attention score calculation subnetwork, the attention score calculation subnetwork is used to calculate the probability of each of the query samples matching the target label based on the implicit features of each of the query samples and the implicit features of each of the supporting samples; based on the first feature extraction subnetwork, the second feature extraction subnetwork and the attention score calculation subnetwork, the matching network is obtained.

[0116] In some embodiments, the construction module is used to construct an attention score calculation subnetwork, specifically including: constructing a similarity calculation function, the similarity calculation function is used to calculate the similarity between each query sample and each supporting sample according to the implicit features of each query sample and the implicit features of each supporting sample; constructing an attention weight calculation function, the attention weight calculation function is used to assign corresponding attention weights according to the similarity between each query sample and each supporting sample; constructing an attention score calculation function, the attention score calculation function is used to obtain the probability that each query sample matches the target label according to the weighted sum of each attention weight and each optimal filter coefficient label; according to the similarity calculation function, the attention weight calculation function and the attention score calculation function, the attention score calculation subnetwork is obtained.

[0117] In some embodiments, before acquiring multiple tasks, the acquisition module is also used to: construct a training database based on a preset sound effect library and a preset noise database, wherein the preset sound effect library includes multiple non-speech environmental noises, and the preset noise database includes background noises in multiple environments; the acquisition module is also used to construct tasks, and the construction steps of each task include: selecting multiple noise categories from the training database, and selecting multiple noise samples from the noise samples corresponding to the multiple noise categories as corresponding support samples to form the support set, and selecting multiple noise samples from the remaining samples corresponding to the same multiple noise categories as corresponding query samples to form the query set; querying each of the support samples in a preset filter coefficient database to obtain the corresponding optimal filter coefficient and generate the corresponding optimal filter coefficient label; and constructing the task based on the support set and the query set.

[0118] In some embodiments, before querying each of the supporting samples in a preset filter coefficient database, the acquisition module is also used to: obtain multiple Gaussian noises of different frequencies in a simulation experiment; determine the interference signal generated by each of the Gaussian noises through the main path, and the main path represents the path from the reference microphone to the error microphone; based on each of the Gaussian noises and each of the interference signals, use an adaptive filtering algorithm to obtain the optimal filter coefficient corresponding to each of the Gaussian noises; and construct the filter coefficient database based on the optimal filter coefficient corresponding to each of the Gaussian noises.

[0119] In some embodiments, the acquisition module is used to determine the interference signal generated by each Gaussian noise passing through the main path, specifically including: in a given noise reduction environment, performing acoustic signal tests at a reference microphone and an error microphone, respectively, to obtain a corresponding noise source signal and a signal at a first human ear; determining the main path transfer function based on the noise source signal and the first human ear signal; and obtaining the corresponding interference signals based on the main path transfer parameters and each Gaussian noise.

[0120] In some embodiments, the acquisition module is used to obtain the optimal filter coefficient corresponding to each Gaussian noise based on each Gaussian noise and each interference signal using an adaptive filtering algorithm, specifically including: in a given noise reduction environment, performing acoustic signal tests at the cancellation speaker and the error microphone, respectively, to obtain the corresponding noise reduction signal and the second human ear signal; determining the secondary path transfer function based on the noise reduction signal and the second human ear signal, wherein the secondary path represents the path from the cancellation speaker to the error microphone; determining the corresponding anti-noise signal based on each interference signal; and obtaining the optimal filter coefficient corresponding to each Gaussian noise based on each Gaussian noise, the secondary path transfer function, the preset cancellation speaker transfer function, and the anti-noise signal using an adaptive filtering algorithm.

[0121] In one embodiment, Figure 7 As shown, an active noise reduction system is provided, including a preprocessing module, a matching module and a noise reduction module, wherein:

[0122] A preprocessing module, used to obtain an online support set, wherein the online support set includes a plurality of real noises collected in an actual vehicle environment and corresponding optimal filter coefficients;

[0123] A matching module, used for collecting online noise in the current vehicle environment, inputting the online support set and the online noise into a trained filter coefficient matching model for processing, and obtaining a current optimal filter coefficient, wherein the trained filter coefficient matching model is obtained according to the training method of the filter coefficient matching model described in the above embodiment;

[0124] A noise reduction module is used to reduce the online noise according to the current optimal filter coefficient.

[0125] For the specific limitations of the training system for the filter coefficient matching model, please refer to the limitations of the training method for the filter coefficient matching model mentioned above. For the specific limitations of the active noise reduction system, please refer to the limitations of the active noise reduction method mentioned above, which will not be repeated here. The various modules in the above-mentioned training system for the filter coefficient matching model and the active noise reduction system can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0126] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as different types of noise and their corresponding optimal filter coefficients. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a training method for a filter coefficient matching model is implemented.

[0127] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an active noise reduction method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0128] Those skilled in the art will understand that Figure 8 , 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0129] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the training method of the filter coefficient matching model are implemented.

[0130] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above active noise reduction method are implemented.

[0131] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0132] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A training method for a filter coefficient matching model, characterized in that: include: Acquire multiple tasks, each of the tasks includes a support set and a query set, the support set includes multiple support samples from multiple noise categories and their corresponding optimal filter coefficient labels, and the query set includes multiple query samples from the same multiple noise categories; Taking maximizing the probability of each query sample matching a target label as a training objective, constructing a target training function, wherein the target label is the optimal filter coefficient label corresponding to the support sample belonging to the same category as the query sample; The matching network is iteratively trained according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

2. The training method of the filter coefficient matching model according to claim 1, characterized in that: Before iteratively training the matching network according to the multiple tasks and the target training function, the method further includes: Constructing a first feature extraction subnetwork, wherein the first feature extraction subnetwork is used to extract implicit features of each of the supporting samples; Constructing a second feature extraction subnetwork, where the second feature extraction subnetwork is used to extract implicit features of each of the query samples; Constructing an attention score calculation subnetwork, wherein the attention score calculation subnetwork is used to calculate the probability that each query sample matches the target label according to the implicit features of each query sample and the implicit features of each support sample; The matching network is obtained based on the first feature extraction subnetwork, the second feature extraction subnetwork and the attention score calculation subnetwork.

3. The training method of the filter coefficient matching model according to claim 2, characterized in that: Construct the attention score calculation subnetwork, including: Constructing a similarity calculation function, wherein the similarity calculation function is used to calculate the similarity between each query sample and each supporting sample according to the implicit features of each query sample and the implicit features of each supporting sample; Constructing an attention weight calculation function, wherein the attention weight calculation function is used to assign a corresponding attention weight according to the similarity between each query sample and each support sample; Constructing an attention score calculation function, wherein the attention score calculation function is used to obtain the probability of each query sample matching the target label according to the weighted sum of each attention weight and each optimal filter coefficient label; The attention score calculation subnetwork is obtained according to the similarity calculation function, the attention weight calculation function and the attention score calculation function.

4. The training method of the filter coefficient matching model according to claim 1, characterized in that: Before acquiring the plurality of tasks, the method further includes: Building a training database according to a preset sound effect library and a preset noise database, wherein the preset sound effect library includes a variety of non-speech environmental noises, and the preset noise database includes background noises in a variety of environments; The build steps for each of these tasks include: Selecting multiple noise categories from the training database, and selecting multiple noise samples from the noise samples corresponding to the multiple noise categories as corresponding support samples to form the support set, and selecting multiple noise samples from the remaining samples corresponding to the same multiple noise categories as corresponding query samples to form the query set; Querying each of the supporting samples in a preset filter coefficient database to obtain a corresponding optimal filter coefficient and generate a corresponding optimal filter coefficient label; The task is constructed according to the support set and the query set.

5. The training method of the filter coefficient matching model according to claim 4, characterized in that: Before querying each of the supporting samples in a preset filter coefficient database, the method further includes: In the simulation experiment, multiple Gaussian noises of different frequencies are obtained; Determine an interference signal generated by each Gaussian noise passing through a main path, wherein the main path represents a path from a reference microphone to an error microphone; Based on each of the Gaussian noises and each of the interference signals, an adaptive filtering algorithm is used to obtain an optimal filter coefficient corresponding to each of the Gaussian noises; The filter coefficient database is constructed according to the optimal filter coefficients corresponding to each of the Gaussian noises.

6. The training method of the filter coefficient matching model according to claim 5, characterized in that: Determining interference signals generated by each Gaussian noise passing through a main path includes: In a given noise reduction environment, acoustic signal tests are performed at the reference microphone and the error microphone respectively to obtain the corresponding noise source signal and the signal at the first human ear; Determining a main path transfer function according to the noise source signal and the first human ear signal; A corresponding interference signal is obtained according to the main path transfer parameter and each Gaussian noise.

7. The training method of the filter coefficient matching model according to claim 5, characterized in that: Based on each of the Gaussian noises and each of the interference signals, an adaptive filtering algorithm is used to obtain an optimal filter coefficient corresponding to each of the Gaussian noises, including: In a given noise reduction environment, acoustic signal tests are performed at the cancellation speaker and the error microphone respectively to obtain the corresponding noise reduction signal and the signal at the second human ear; Determine a secondary path transfer function according to the noise reduction signal and the second human ear signal, wherein the secondary path represents a path from the cancellation speaker to the error microphone; Determine a corresponding anti-noise signal according to each of the interference signals; Based on each of the Gaussian noises, the secondary path transfer function, the preset cancellation speaker transfer function and each of the anti-noise signals, an adaptive filtering algorithm is used to obtain the optimal filter coefficient corresponding to each of the Gaussian noises.

8. An active noise reduction method, characterized in that: include: Acquire an online support set, wherein the online support set includes a plurality of real noises from an actual vehicle use environment and their corresponding optimal filter coefficients; Collecting online noise in the current vehicle environment, inputting the online support set and the online noise into a trained filter coefficient matching model for processing to obtain a current optimal filter coefficient, wherein the trained filter coefficient matching model is obtained according to the training method for the filter coefficient matching model according to any one of claims 1 to 7; The online noise is reduced according to the current optimal filter coefficient.

9. A training system for a filter coefficient matching model, characterized in that: include: An acquisition module, configured to acquire a plurality of tasks, each of the tasks comprising a support set and a query set, the support set comprising a plurality of support samples from a plurality of noise categories and their corresponding optimal filter coefficient labels, and the query set comprising a plurality of query samples from the same plurality of noise categories; A construction module is used to take maximizing the probability of each query sample matching a target label as a training goal and construct a target training function, wherein the target label is an optimal filter coefficient label corresponding to a support sample belonging to the same category as the query sample; A training module is used for iteratively training a matching network according to the multiple tasks and the target training function to obtain a filter coefficient matching model.

10. An active noise reduction system, characterized in that: include: A preprocessing module, used to obtain an online support set, wherein the online support set includes a plurality of real noises collected in an actual vehicle environment and corresponding optimal filter coefficients; A matching module, used for collecting online noise in the current vehicle environment, inputting the online support set and the online noise into a trained filter coefficient matching model for processing, and obtaining a current optimal filter coefficient, wherein the trained filter coefficient matching model is obtained according to the training method of the filter coefficient matching model according to any one of claims 1 to 7; A noise reduction module is used to reduce the online noise according to the current optimal filter coefficient.

Images