Noise reduction calibration method and system for wireless headphones

Abstract

The present application relates to the technical field of headphone noise reduction, and discloses a noise reduction calibration method and system for wireless headphones. The method includes: initializing the internal acoustic model of the wireless headphones to obtain a first initial headphone acoustic model and a second initial headphone acoustic model; collecting a test environment noise signal and performing noise filtering processing to generate a target noise signal; performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals; performing inverse phase sound wave calculation and fusion to obtain a first initial inverse phase sound wave and a second initial inverse phase sound wave; performing inverse filtering frequency equalization processing to obtain a first target inverse phase sound wave and a second target inverse phase sound wave; performing headphone acoustic model parameter compensation through the first target inverse phase sound wave and the second target inverse phase sound wave to obtain a first target headphone acoustic model and a second target headphone acoustic model. The present application realizes intelligent noise reduction calibration of wireless headphones and improves the noise reduction effect of wireless headphones.

Description

Technical Field

[0001] The present application relates to the technical field of headphone noise reduction, and in particular to a noise reduction calibration method and system for wireless headphones. Background Art

[0002] Wireless headphones have become an indispensable part of daily life for consumers, especially in noisy environments, where the noise reduction function of wireless headphones is particularly important. With the rapid development of the wireless headphone market, users have higher and higher requirements for the noise reduction effect of headphones, which has promoted the continuous progress and innovation of noise reduction technology. However, despite the continuous advancement of technology, wireless headphones still face many challenges in noise reduction, especially how to provide stable and effective noise reduction effects in different environments, and how to accurately calibrate and optimize the acoustic model of headphones to adapt to the specific needs of users.

[0003] Existing noise reduction technologies mainly rely on two methods: passive noise reduction and active noise reduction. Passive noise reduction uses physical structures to block the introduction of external noise, while active noise reduction offsets noise by generating sound waves opposite to the external noise. Although these methods can reduce noise to a certain extent, in actual applications, due to the complexity and variability of external environmental noise, it is difficult for headphones to achieve good noise reduction effects in all scenarios. In addition, the differences in ear shape and hearing preferences of each user also make the noise reduction effect of headphones vary significantly among different users. Summary of the invention

[0004] The present application provides a noise reduction calibration method and system for wireless headphones, which are used to implement intelligent noise reduction calibration of wireless headphones and improve the noise reduction effect of wireless headphones.

[0005] In a first aspect, the present application provides a noise reduction calibration method for a wireless headset, the noise reduction calibration method for the wireless headset comprising:

[0006] Initializing an internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model;

[0007] Collecting a test environment noise signal and performing noise filtering on the test environment noise signal to generate a target noise signal;

[0008] Performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals;

[0009] Using the first initial headphone acoustic model and the second initial headphone acoustic model, respectively, the multiple sub-noise signals are subjected to inverse phase sound wave calculation and fusion to obtain a first initial inverse phase sound wave and a second initial inverse phase sound wave;

[0010] Performing inverse filtering frequency equalization processing on the first initial reverse phase sound wave and the second initial reverse phase sound wave to obtain a first target reverse phase sound wave and a second target reverse phase sound wave;

[0011] The first target antiphase sound wave and the second target antiphase sound wave are used to perform headphone acoustic model parameter compensation to obtain a first target headphone acoustic model and a second target headphone acoustic model.

[0012] In a second aspect, the present application provides a noise reduction calibration system for a wireless headset, the noise reduction calibration system for a wireless headset comprising:

[0013] An initialization module, used to initialize the internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model;

[0014] A filtering module is used to collect a test environment noise signal and perform noise filtering on the test environment noise signal to generate a target noise signal;

[0015] A decomposition module, used for performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals;

[0016] a calculation module, configured to respectively calculate and fuse the multiple sub-noise signals using the first initial headphone acoustic model and the second initial headphone acoustic model to obtain a first initial inverted phase sound wave and a second initial inverted phase sound wave;

[0017] an equalization module, configured to perform inverse filtering frequency equalization processing on the first initial inverted phase sound wave and the second initial inverted phase sound wave to obtain a first target inverted phase sound wave and a second target inverted phase sound wave;

[0018] The compensation module is used to compensate the headphone acoustic model parameters by using the first target antiphase sound wave and the second target antiphase sound wave to obtain the first target headphone acoustic model and the second target headphone acoustic model.

[0019] A third aspect of the present application provides a noise reduction calibration device for wireless headphones, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the noise reduction calibration device for the wireless headphones executes the above-mentioned noise reduction calibration method for the wireless headphones.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned noise reduction calibration method for wireless headphones.

[0021] In the technical solution provided by the present application, by collecting the test environment noise signal and performing refined noise filtering processing, the noise characteristics in various complex environments can be accurately captured, thereby providing an accurate data basis for subsequent noise reduction processing. This precise identification and analysis of environmental noise ensures that the noise reduction system can adapt to changing environmental conditions and provide a stable and effective noise reduction effect. Through the initialization and parameter compensation of the acoustic model inside the headset, combined with advanced acoustic processing models including noise recognition network, anti-phase sound wave synthesis network and anti-phase sound wave fusion network, personalized noise reduction effect adjustment can be performed according to the use environment and personal preferences of different users. This personalized adjustment can not only improve the user's satisfaction, but also enhance the market competitiveness of headphone products. The genetic optimization algorithm is used to compensate the parameters of the headphone acoustic model, so that the noise reduction performance of the headphone can be dynamically optimized and adjusted according to real-time feedback. This dynamic adjustment mechanism not only enhances the adaptability of the headset to different noise environments, but also enables the headset to better meet the user's hearing needs that change over time. By performing NKP decomposition on the target noise signal, multiple sub-noise signals are obtained, and the inverse sound waves of these sub-signals are calculated and fused, which can process the noise signal more finely and improve the accuracy and effect of noise reduction. At the same time, the inverse filtering frequency equalization processing further improves the quality of the inverse sound waves to ensure the maximum noise reduction effect. From the initialization of the acoustic model to the final model parameter compensation, a closed-loop optimization system is formed. This not only improves the noise reduction effect, but also helps to improve the overall performance of the headphones, thereby realizing the intelligent noise reduction calibration of wireless headphones and improving the noise reduction effect of wireless headphones. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of an embodiment of a noise reduction calibration method for wireless headphones in an embodiment of the present application;

[0024] Figure 2 Schematic diagram of an embodiment of a noise reduction calibration system for wireless headphones in an embodiment of the present application. DETAILED DESCRIPTION

[0025] Embodiments of the present application provide a method and system for calibrating noise reduction of wireless headphones. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the noise reduction calibration method of the wireless headset in the embodiment of the present application includes:

[0027] Step S101, initializing an internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model;

[0028] It is understandable that the execution subject of the present application may be a noise reduction calibration system for wireless headphones, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0029] Specifically, the internal acoustic characteristics of the wireless headset are simulated and analyzed by ANSYS Acoustics software, and the acoustic characteristic parameter set of the wireless headset at different frequencies is calculated based on the acoustic simulation capability of the software. Based on the acoustic characteristic parameters obtained by simulation, corresponding internal acoustic processing models are created for the left earphone and the right earphone in the wireless headset, respectively, and the two models process the acoustic data of the left and right earphones respectively to ensure the coordinated and consistent noise reduction effect of both ears. The internal acoustic processing model mainly includes a noise recognition network, an anti-phase sound wave synthesis network, and an anti-phase sound wave fusion network, which effectively recognizes the environmental noise, accurately synthesizes the anti-phase sound waves, and finally performs fusion processing. The initialization network hyperparameter set of the left and right earphones is generated respectively through the acoustic characteristic parameter set, and the hyperparameter set directly affects the performance and effect of the internal acoustic processing model. By applying the respective initialization network hyperparameter sets to the left and right internal acoustic processing models for hyperparameter initialization, the first initial earphone acoustic model and the second initial earphone acoustic model are obtained. The two models represent the initial states of the left and right earphones in the noise reduction calibration method, respectively.

[0030] Step S102: collecting a test environment noise signal and performing noise filtering on the test environment noise signal to generate a target noise signal;

[0031] Specifically, noise signals are collected from the test environment, and a highly sensitive microphone or other sound collection device is used to ensure that various sounds in the environment can be captured. The noise reduction frequency range of the wireless headset is defined. This range indicates the specific frequency band that the noise reduction processing needs to pay attention to, which is usually the frequency range that the human ear is most sensitive to, such as the medium and low frequency area, so as to more effectively improve the noise reduction effect and ensure the sound quality. The collected environmental noise signal is amplified to increase the signal strength. The amplified environmental noise signal is subjected to signal standardization processing to ensure that the amplitude of the signal reaches a standard level. The high-frequency noise in the standard environmental noise signal is removed by a filter, and only the signal below a certain frequency is retained according to the pre-defined noise reduction frequency range to obtain the first noise signal. The low-frequency noise in the first noise signal is removed by a filter, and the final second noise signal is ensured to be concentrated in the most critical frequency range according to the noise frequency range to improve the noise reduction effect. The second noise signal is subjected to noise enhancement processing to strengthen the noise components that are important to the noise reduction performance, while removing or reducing the influence of those unimportant noise components. The target noise signal finally generated will be used as a benchmark in the noise reduction calibration process to adjust and optimize the noise reduction algorithm and acoustic model of the wireless headset.

[0032] Step S103, performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals;

[0033] Specifically, a fast Fourier transform is performed on the target noise signal, and the target noise signal is converted to the frequency domain to obtain the distribution characteristics of the noise signal in the frequency domain. The frequency domain noise signal contains the frequency components of the noise. Feature recognition is performed on the frequency domain noise signal. By identifying the key characteristic frequencies or patterns in the frequency domain signal, a noise feature set is obtained to reflect the main frequency components and energy distribution of the target noise signal. Decomposition feature point prediction is performed based on the noise feature set, and the identified noise features are matched with the predefined decomposition model to predict the decomposition feature points that can represent the key features of the noise signal. Signal decomposition interval prediction is performed on the signal decomposition feature points, and the feature interval suitable for signal decomposition is predicted by analyzing the relationship and distribution between each feature point. The feature interval is determined based on the characteristics of the noise signal and the decomposition target, and the most representative and influential part of the noise signal is captured. Based on the predicted multiple signal decomposition feature intervals, the target noise signal is decomposed to obtain multiple sub-noise signals. The signal decomposition process is carried out according to the specific characteristics of each feature interval, aiming to refine the target noise signal into several sub-noise signals, each of which represents a specific part or feature in the original noise signal.

[0034] Step S104: calculating and fusing the inverse phase sound waves of the plurality of sub-noise signals respectively by using the first initial headphone acoustic model and the second initial headphone acoustic model to obtain the first initial inverse phase sound wave and the second initial inverse phase sound wave;

[0035] Specifically, multiple sub-noise signals are input into the noise recognition network in the two initial headphone acoustic models respectively. The noise recognition network uses a deep learning algorithm to extract the key noise features in each sub-noise signal. The noise recognition network adopts the structure of the first double-layer long short-term memory (LSTM) network and the gated recurrent unit (GRU) network and the linear normalization layer. The composite network structure helps to capture and extract the timing characteristics and dynamic changes in the noise signal, so as to more accurately identify the noise. The noise feature vector is input into the anti-phase sound wave synthesis network, which includes an encoder and a decoder. The encoder part processes and compresses the noise features through the input layer and the two-layer convolutional neural network (CNN) to form a more compact and advanced noise feature representation. The decoder part consists of a recurrent neural network (RNN) and a two-layer residual network to synthesize the anti-phase sound wave according to the encoded noise features. The synthesis process takes into account the basic characteristics of the noise and the time series characteristics of the noise to ensure that the generated anti-phase sound wave can effectively offset the original noise signal. The synthesized inverse sound wave of each sub-noise signal is input into the inverse sound wave fusion network to fuse all the sub-inverse sound waves into a complete inverse sound wave. The fusion network adopts the structure of attention weighted fusion layer and output layer, in which the attention mechanism can automatically adjust the weight of each sub-inverse sound wave according to its importance to ensure that more important features are highlighted during the fusion process, thereby improving the quality of the overall inverse sound wave and the noise reduction effect. The whole process is carried out in parallel in two initial headphone acoustic models, and each model processes the noise signal of one side of the headphone separately to ensure that the noise reduction performance of the left and right ears is coordinated and consistent. Through processing, each model can obtain an initial inverse sound wave for its processed signal, and these inverse sound waves are used in the noise reduction calibration of the headphone to achieve a better noise reduction experience for the user.

[0036] Step S105, performing inverse filtering frequency equalization processing on the first initial inverted phase sound wave and the second initial inverted phase sound wave to obtain a first target inverted phase sound wave and a second target inverted phase sound wave;

[0037] Specifically, the frequency band intensity characteristics of the first initial anti-phase sound wave are analyzed to understand the energy distribution of the anti-phase sound wave in different frequency bands, and the first frequency band intensity characteristics are obtained. The characteristics reflect which frequency bands of the sound in the anti-phase sound wave are over-enhanced or weakened. Similarly, the frequency band intensity characteristics of the second initial anti-phase sound wave are analyzed to obtain the second frequency band intensity characteristics. Based on the first frequency band intensity characteristics, the first initial anti-phase sound wave is subjected to inverse filtering frequency intensity equalization adjustment. The intensity of the sound wave in each frequency band is adjusted by inverse filtering technology to ensure that the sound intensity of each frequency band can reach a more ideal and balanced state. Inverse filtering adjusts the sound wave based on the algorithm and acoustic model to ensure that the adjusted sound wave can more accurately produce a counteracting effect with the external noise. Similarly, the second initial anti-phase sound wave is subjected to inverse filtering frequency intensity equalization adjustment according to the second frequency band intensity characteristics to obtain the second target anti-phase sound wave. Through adjustment, the noise reduction effect is improved, and the sound quality distortion problem caused by uneven frequency intensity is avoided. Finally, the obtained first target anti-phase sound wave and second target anti-phase sound wave are more balanced in frequency intensity, and can better adapt to the auditory needs of different users and different external noise environments.

[0038] Step S106: Compensate the headphone acoustic model parameters using the first target antiphase sound wave and the second target antiphase sound wave to obtain the first target headphone acoustic model and the second target headphone acoustic model.

[0039] Specifically, the first noise reduction evaluation index of the first target antiphase sound wave and the second noise reduction evaluation index of the second target antiphase sound wave are calculated respectively. These indicators comprehensively consider multiple aspects of the noise reduction effect, such as noise reduction depth, frequency response balance, etc., and provide accurate evaluation basis for parameter optimization. The first and second noise reduction evaluation indicators obtained reflect the noise reduction performance of the respective antiphase sound waves. A multi-objective fitness function is defined based on the noise reduction evaluation index, which is used in the genetic optimization algorithm to simultaneously optimize multiple noise reduction performance indicators to ensure that the final noise reduction effect is both efficient and balanced. The genetic optimization algorithm is a search algorithm that simulates the principles of natural selection and genetics. It continuously optimizes the quality of the solution through selection, crossover and mutation operations in an iterative process. The genetic optimization algorithm is used to initialize the parameter population of the first initial headphone acoustic model and the second initial headphone acoustic model to generate multiple first compensation parameter sets. This is the starting point for finding the optimal noise reduction parameter configuration, and each parameter set represents a possible headphone acoustic model parameter configuration. The first compensation parameter set is evaluated by the multi-objective fitness function, the fitness value of each parameter set is calculated, and which parameter configurations can better improve the noise reduction effect are identified. According to the calculated fitness value, multiple corresponding second compensation parameter sets are generated, and the parameter sets are optimized to finally obtain the target compensation parameter set. Through continuous iterative optimization, the optimal headphone acoustic model parameter configuration is gradually approached to ensure that the optimal noise reduction effect can be achieved. Based on the obtained target compensation parameter set, the first initial headphone acoustic model and the second initial headphone acoustic model are compensated for the headphone acoustic model parameters to obtain the first and second target headphone acoustic models.

[0040] In the embodiment of the present application, by collecting the test environment noise signal and performing refined noise filtering processing, the noise characteristics in various complex environments can be accurately captured, thereby providing an accurate data basis for subsequent noise reduction processing. This precise identification and analysis of environmental noise ensures that the noise reduction system can adapt to changing environmental conditions and provide a stable and effective noise reduction effect. Through the initialization and parameter compensation of the acoustic model inside the headset, combined with advanced acoustic processing models including noise recognition network, anti-phase sound wave synthesis network and anti-phase sound wave fusion network, personalized noise reduction effect adjustment can be performed according to the use environment and personal preferences of different users. This personalized adjustment can not only improve the user's satisfaction, but also enhance the market competitiveness of the headset product. The genetic optimization algorithm is used to compensate the parameters of the headset acoustic model, so that the noise reduction performance of the headset can be dynamically optimized and adjusted according to real-time feedback. This dynamic adjustment mechanism not only enhances the adaptability of the headset to different noise environments, but also enables the headset to better meet the user's hearing needs that change over time. By performing NKP decomposition on the target noise signal, multiple sub-noise signals are obtained, and the inverse sound waves of these sub-signals are calculated and fused, which can process the noise signal more finely and improve the accuracy and effect of noise reduction. At the same time, the inverse filtering frequency equalization processing further improves the quality of the inverse sound waves to ensure the maximum noise reduction effect. From the initialization of the acoustic model to the final model parameter compensation, a closed-loop optimization system is formed. This not only improves the noise reduction effect, but also helps to improve the overall performance of the headphones, thereby realizing the intelligent noise reduction calibration of wireless headphones and improving the noise reduction effect of wireless headphones.

[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0042] (1) Use ANSYS Acoustics software to simulate the internal acoustic characteristics of wireless headphones and calculate the acoustic characteristic parameter set of wireless headphones at different frequencies;

[0043] (2) creating a left-side internal acoustic processing model corresponding to the left earphone in the wireless headset, and creating a right-side internal acoustic processing model corresponding to the right earphone in the wireless headset, wherein the left-side internal acoustic processing model and the right-side internal acoustic processing model have the same model architecture, and the left-side internal acoustic processing model or the right-side internal acoustic processing model includes: a noise recognition network, an anti-phase sound wave synthesis network, and an anti-phase sound wave fusion network;

[0044] (3) Generate a left-side initialization network hyperparameter set and a right-side initialization network hyperparameter set through the acoustic characteristic parameter set;

[0045] (4) Initializing the network hyperparameters of the left internal acoustic processing model by using the left-side initialization network hyperparameter set to obtain a first initial headphone acoustic model, and initializing the network hyperparameters of the right internal acoustic processing model by using the right-side initialization network hyperparameter set to obtain a second initial headphone acoustic model.

[0046] Specifically, the acoustic characteristics of wireless headphones are simulated by ANSYS Acoustics software, simulating phenomena such as sound wave propagation and reflection inside the headphones, as well as the influence of headphone materials on acoustic performance. The software calculates the acoustic response of wireless headphones at different frequencies to obtain a comprehensive set of acoustic characteristic parameters. These parameter sets include the response of the headphones at various frequency points and also reflect the influence of the headphone structure on sound propagation. Based on the obtained set of acoustic characteristic parameters, corresponding internal acoustic processing models are created for the left and right headphones in the wireless headphones. The two models are consistent in structure, but correspond to the left and right channels of the headphones respectively to ensure that the noise reduction processing can be performed independently on both sides. Each internal acoustic processing model includes a noise recognition network, an anti-phase sound wave synthesis network, and an anti-phase sound wave fusion network. These three networks are responsible for identifying specific features in environmental noise, synthesizing effective anti-phase sound waves based on these features, and effectively fusing multiple synthesized anti-phase sound waves into a single anti-phase sound wave to achieve the best noise reduction effect. In order to enable these internal acoustic processing models to perform effective noise reduction processing according to the actual acoustic characteristics of the headphones, the left and right initialization network hyperparameter sets are generated through the acoustic characteristic parameter set. The network hyperparameter set refers to the parameters that control the network training process and network structure configuration, such as learning rate, number of network layers, number of nodes per layer, etc. The purpose of generating these hyperparameter sets is to enable the network to better adapt to the acoustic characteristics of the headphones and improve the accuracy and efficiency of the network in identifying noise and synthesizing anti-phase sound waves. The network hyperparameters of the left internal acoustic processing model are initialized by the left initialization network hyperparameter set. Similarly, the right internal acoustic processing model is initialized using the right initialization network hyperparameter set to obtain the first initial headphone acoustic model and the second initial headphone acoustic model, which respectively represent the acoustic processing capabilities of the left and right headphones after preliminary optimization.

[0047] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0048] (1) Collect the test environment noise signal and define the noise reduction frequency range of the wireless headset;

[0049] (2) Amplifying the test environment noise signal to obtain an amplified environment noise signal;

[0050] (3) Performing signal standardization processing on the amplified ambient noise signal to obtain a standard ambient noise signal;

[0051] (4) removing high-frequency noise from the standard environmental noise signal according to the noise reduction frequency range to obtain a first noise signal;

[0052] (5) removing low-frequency noise from the first noise signal according to the noise reduction frequency range to obtain a second noise signal;

[0053] (6) Perform noise enhancement processing on the second noise signal to generate a target noise signal.

[0054] Specifically, the noise signal in the test environment is collected. During the collection process, a highly sensitive microphone device is used to collect noise data in an actual or simulated use environment to ensure that the captured noise signal reflects the noise characteristics in the real environment as much as possible. Define the noise reduction frequency range of wireless headphones. Noises of different frequencies interfere with users to different degrees. The effective noise reduction frequency range should cover the frequency bands that the human ear is most sensitive to, which usually involves the mid- and low-frequency ranges, where human voices and most environmental noises are concentrated. Amplify the test environment noise signal to ensure the identifiability of the noise signal in subsequent processing and make the signal characteristics more obvious. Standardize the amplified environmental noise signal and adjust the signal strength to a certain standard level. Process the standardized environmental noise signal through a filter to remove high-frequency noise that is not within the noise reduction frequency range. The filter can effectively shield frequency components above the set threshold and only allow the low-frequency part to pass. This can eliminate unnecessary high-frequency interference and focus on the noise frequency band that has a greater impact on users. Perform low-frequency noise removal on the filtered noise signal to eliminate frequency components below the noise reduction frequency range to ensure that the remaining noise signal is concentrated in the most critical frequency band. The second noise signal is subjected to noise enhancement processing to enhance key features in the noise signal without changing the noise characteristics, so that the key features in the noise signal are easier to be identified and offset in the subsequent noise reduction processing.

[0055] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0056] (1) Perform fast Fourier transform on the target noise signal to obtain the frequency domain noise signal;

[0057] (2) Perform feature recognition on the frequency domain noise signal to obtain a noise feature set;

[0058] (3) Decomposing the target noise signal according to the noise feature set to predict the decomposition feature points and obtain multiple signal decomposition feature points;

[0059] (4) Predicting the signal decomposition interval for multiple signal decomposition feature points to obtain multiple signal decomposition feature intervals;

[0060] (5) Based on multiple signal decomposition feature intervals, the target noise signal is decomposed to obtain multiple sub-noise signals.

[0061] Specifically, the target noise signal is subjected to a fast Fourier transform to convert the noise signal in the time domain into a frequency domain representation. The noise signal is easier to analyze and process in the frequency domain. Fast Fourier transform allows the system to identify the presence and intensity of each frequency component in the noise signal. Then, feature recognition is performed on the frequency domain noise signal. The frequency domain noise signal is analyzed to identify those representative noise features, such as the peak value of a specific frequency, the energy distribution pattern, etc. These noise feature sets contain the key information of the noise signal. Decomposition feature point prediction is performed based on the identified noise feature set. According to the feature set of the noise signal, the feature points that play a key role in the noise signal are predicted. These feature points are the most representative parts of the noise signal and can be regarded as the key nodes of noise processing. Then, the signal decomposition interval prediction is performed for the predicted multiple signal decomposition feature points. The identified feature points are used as references to predict a series of signal decomposition feature intervals. These intervals are divided according to the position of the feature points and the characteristics of the noise signal. The noise components in each interval have certain commonalities or continuity. The purpose of signal decomposition interval prediction is to divide the noise signal more accurately so that the subsequent signal decomposition and processing can be more targeted. Based on the predicted multiple signal decomposition feature intervals, the target noise signal is decomposed to finally obtain multiple sub-noise signals. Each sub-noise signal represents a part of the original noise signal, but has more specific characteristics and frequency components.

[0062] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0063] (1) inputting the plurality of sub-noise signals into a noise recognition network in a first initial headphone acoustic model respectively, and extracting a first noise feature vector of each sub-noise signal through the noise recognition network, wherein the noise recognition network includes: a first double-layer LSTM network, a first GRU network, and a first linear normalization layer;

[0064] (2) respectively inputting the first noise feature vector of each sub-noise signal into the anti-phase sound wave synthesis network in the first initial headphone acoustic model, and synthesizing the first sub-anti-phase sound wave of each sub-noise signal through the anti-phase sound wave synthesis network, wherein the anti-phase sound wave synthesis network comprises: a first encoder and a first decoder, wherein the first encoder comprises a first input layer and a first two-layer CNN network, and the first decoder comprises a first RNN network and a first two-layer residual network in sequence;

[0065] (3) respectively inputting the first sub-inverted sound wave of each sub-noise signal into the inverted sound wave fusion network in the first initial headphone acoustic model, and performing inverted sound wave fusion on the first sub-inverted sound wave of each sub-noise signal through the inverted sound wave fusion network to obtain a first initial inverted sound wave, wherein the inverted sound wave fusion network includes: a first attention weighted fusion layer and a first output layer;

[0066] (4) inputting the plurality of sub-noise signals into the noise recognition network in the second initial headphone acoustic model respectively, and extracting the second noise feature vector of each sub-noise signal through the noise recognition network, wherein the noise recognition network includes: a second double-layer LSTM network, a second GRU network, and a second linear normalization layer;

[0067] (5) respectively inputting the second noise feature vector of each sub-noise signal into the anti-phase sound wave synthesis network in the second initial headphone acoustic model, and synthesizing the second sub-anti-phase sound wave of each sub-noise signal through the anti-phase sound wave synthesis network, wherein the anti-phase sound wave synthesis network includes: a second encoder and a second decoder, wherein the second encoder includes a second input layer and a second two-layer CNN network, and the second decoder includes a second RNN network and a second two-layer residual network in sequence;

[0068] (6) The second sub-inverse sound wave of each sub-noise signal is respectively input into the inverse sound wave fusion network in the second initial headphone acoustic model, and the second sub-inverse sound wave of each sub-noise signal is inversely fused through the inverse sound wave fusion network to obtain a second initial inverse sound wave. The inverse sound wave fusion network includes: a second attention weighted fusion layer and a second output layer.

[0069] Specifically, the sub-noise signal is input into the noise recognition network in the headphone acoustic model, and each sub-noise signal is identified and feature extracted. The noise recognition network adopts deep learning technology, including a two-layer long short-term memory (LSTM) network and a gated recurrent unit (GRU) network, as well as a linear normalization layer. These networks work together to extract key feature vectors from complex noise signals. The first two-layer LSTM network and the first GRU network work together to capture the time series dependency in the noise signal, thereby effectively extracting temporally continuous noise features. Through the linear normalization layer processing, the numerical stability of the feature vector is ensured, providing a standardized input for the subsequent anti-phase sound wave synthesis. The feature vector of each sub-noise signal is input into the anti-phase sound wave synthesis network, which is designed to synthesize effective anti-phase sound waves based on the noise feature vector. The synthesis network consists of an encoder and a decoder, wherein the encoder includes an input layer and a two-layer convolutional neural network (CNN), and the CNN layer extracts and compresses noise features. The decoder works through the included recurrent neural network (RNN) and a two-layer residual network to restore and generate anti-phase sound waves with a target phase based on the compressed noise features. This relies on a neural network model to simulate the complexity of noise, while using a residual network to improve the network's learning ability to ensure that the generated anti-phase sound waves can effectively cancel out the original noise signal. Each generated sub-anti-phase sound wave is input into the anti-phase sound wave fusion network. The network uses an attention weighted fusion layer, which can automatically adjust the weights of each sub-anti-phase sound wave in the final output according to their contribution to achieve the optimal noise cancellation effect. Through weighted fusion, it can be ensured that those sub-anti-phase sound waves that contribute more to the noise reduction effect are given higher priority, thereby generating high-quality initial anti-phase sound waves in the output layer. For the left and right channels of the headphones, the above steps are performed separately to ensure the consistency and optimization of binaural noise reduction effects. Although the noise recognition, anti-phase sound wave synthesis and fusion networks of the left and right channels are consistent in structure, the signals they process and the anti-phase sound waves they generate are different, reflecting the independent noise reduction requirements of the headphones' binaural channels.

[0070] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0071] (1) Analyzing the frequency band intensity characteristics of the first initial antiphase sound wave to obtain the first frequency band intensity characteristics;

[0072] (2) analyzing the frequency band intensity characteristics of the second initial antiphase sound wave to obtain the second frequency band intensity characteristics;

[0073] (3) performing inverse filtering frequency intensity equalization adjustment on the first initial anti-phase sound wave according to the intensity characteristics of the first frequency band to obtain a first target anti-phase sound wave;

[0074] (4) Performing inverse filtering frequency intensity equalization adjustment on the second initial antiphase sound wave according to the intensity characteristics of the second frequency band to obtain a second target antiphase sound wave.

[0075] Specifically, the frequency band intensity characteristics of the first initial anti-phase sound wave are analyzed to quantify the energy distribution of the anti-phase sound wave in different frequency bands, and to understand which frequency bands have sound intensities exceeding or falling below the expected level. At the same time, the frequency band intensity characteristics of the second initial anti-phase sound wave are analyzed to obtain the second frequency band intensity characteristics. Ensure that the dual-channel output of the headset is fully considered so that the noise reduction processing can be evenly and effectively applied to the left and right sides of the headset to ensure the consistency and balance of the binaural hearing experience. Then, the initial anti-phase sound wave is subjected to inverse filtering frequency intensity equalization adjustment according to the frequency band intensity characteristics. According to the first frequency band intensity characteristics, the first initial anti-phase sound wave is adjusted by inverse filtering technology to balance the sound intensity of each frequency band so that it is closer to the ideal noise reduction performance curve. Inverse filtering processing includes reducing excessive frequency components and enhancing certain weak frequency bands to ensure the balance and effectiveness of the anti-phase sound wave in the entire frequency range. Similarly, the second initial anti-phase sound wave undergoes an inverse filtering frequency intensity equalization adjustment process according to the second frequency band intensity characteristics. Ensure that the right ear anti-phase sound wave also achieves the most optimized frequency response characteristics. Through the above steps, the first and second initial anti-phase sound waves are adjusted to the first and second target anti-phase sound waves respectively. After the inverse filtering frequency intensity equalization adjustment, these target anti-phase sound waves can more accurately offset the external noise and minimize the sound quality distortion that may be introduced, providing a more natural and clearer listening experience.

[0076] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0077] (1) calculating a first noise reduction evaluation index of a first target antiphase sound wave and a second noise reduction evaluation index of a second target antiphase sound wave respectively;

[0078] (2) defining a multi-objective fitness function of the genetic optimization algorithm based on the first noise reduction evaluation index and the second noise reduction evaluation index;

[0079] (3) initializing the noise reduction calibration compensation parameter population of the first initial headphone acoustic model and the second initial headphone acoustic model by a genetic optimization algorithm to obtain a plurality of first compensation parameter sets;

[0080] (4) Calculating the fitness of multiple first compensation parameter sets through a multi-objective fitness function to obtain a fitness value of each first compensation parameter set;

[0081] (5) generating a plurality of corresponding second compensation parameter sets according to the fitness values, and optimizing the plurality of second compensation parameter sets to obtain a target compensation parameter set;

[0082] (6) Performing headphone acoustic model parameter compensation on the first initial headphone acoustic model and the second initial headphone acoustic model based on the target compensation parameter set to obtain a first target headphone acoustic model and a second target headphone acoustic model.

[0083] Specifically, the first noise reduction evaluation index of the first target antiphase sound wave and the second noise reduction evaluation index of the second target antiphase sound wave are calculated respectively. These indexes can quantitatively reflect the quality of the noise reduction effect, including the noise reduction depth, the uniformity of the frequency response, and the efficiency of suppressing the ambient noise. The multi-objective fitness function of the genetic optimization algorithm is defined based on the evaluation index. The genetic optimization algorithm is a search algorithm that imitates the natural selection and genetic mechanism in the biological evolution process, and finds the optimal solution to the problem through the iterative evolution of the population. The multi-objective fitness function aims to optimize multiple noise reduction performance indicators at the same time to achieve a comprehensive optimal noise reduction effect, ensuring that in the optimization process, not only a single performance indicator is focused on, but all aspects of the noise reduction effect are comprehensively considered to obtain the best overall noise reduction performance. The first initial headphone acoustic model and the second initial headphone acoustic model are initialized with a population of noise reduction calibration compensation parameters by using a genetic optimization algorithm. Multiple candidate compensation parameter sets are generated, each of which represents a possible headphone acoustic model configuration. The fitness of the candidate compensation parameter sets is calculated by using the multi-objective fitness function. The fitness value of each parameter set reflects the performance of its corresponding configuration in terms of noise reduction performance, and a high fitness value means a better noise reduction effect. By comparing the fitness values ​​of different parameter sets, it is possible to identify which parameter configurations are closer to the ideal noise reduction effect. According to the fitness value, the algorithm generates multiple optimized compensation parameter sets. The optimized parameter set is optimized and solved, and the final target compensation parameter set is found through iterative optimization. The target parameter set represents the headphone acoustic model configuration that can achieve the optimal noise reduction effect under the current algorithm and model conditions. Based on the target compensation parameter set, the first initial headphone acoustic model and the second initial headphone acoustic model are finally parameter compensated to obtain the first target headphone acoustic model and the second target headphone acoustic model. These two models provide the best noise reduction configuration for the left and right channels of the headphones, respectively, ensuring that the wireless headphones can provide excellent noise reduction effects in various environments.

[0084] The above describes the noise reduction calibration method of the wireless headset in the embodiment of the present application. The following describes the noise reduction calibration system of the wireless headset in the embodiment of the present application. Figure 2 In one embodiment of the present application, a noise reduction calibration system for wireless headphones includes:

[0085] An initialization module 201 is used to initialize the internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model;

[0086] The filtering module 202 is used to collect the test environment noise signal and perform noise filtering on the test environment noise signal to generate a target noise signal;

[0087] A decomposition module 203 is used to perform NKP decomposition on the target noise signal to obtain multiple sub-noise signals;

[0088] The calculation module 204 is used to calculate and fuse the anti-phase sound waves of the plurality of sub-noise signals respectively through the first initial headphone acoustic model and the second initial headphone acoustic model to obtain the first initial anti-phase sound wave and the second initial anti-phase sound wave;

[0089] An equalization module 205 is used to perform inverse filtering frequency equalization processing on the first initial reverse phase sound wave and the second initial reverse phase sound wave to obtain a first target reverse phase sound wave and a second target reverse phase sound wave;

[0090] The compensation module 206 is used to compensate the headphone acoustic model parameters by using the first target anti-phase sound wave and the second target anti-phase sound wave to obtain the first target headphone acoustic model and the second target headphone acoustic model.

[0091] Through the collaboration of the above components, by collecting the test environment noise signal and performing refined noise filtering processing, the noise characteristics in various complex environments can be accurately captured, thus providing an accurate data basis for subsequent noise reduction processing. This precise identification and analysis of environmental noise ensures that the noise reduction system can adapt to changing environmental conditions and provide stable and effective noise reduction effects. Through the initialization and parameter compensation of the internal acoustic model of the headset, combined with advanced acoustic processing models including noise recognition network, anti-phase sound wave synthesis network and anti-phase sound wave fusion network, personalized noise reduction effect adjustment can be performed according to the use environment and personal preferences of different users. This personalized adjustment can not only improve user satisfaction, but also enhance the market competitiveness of headphone products. The genetic optimization algorithm is used to compensate the parameters of the headphone acoustic model, so that the noise reduction performance of the headphone can be dynamically optimized and adjusted according to real-time feedback. This dynamic adjustment mechanism not only enhances the adaptability of the headphone to different noise environments, but also enables the headphone to better meet the user's changing listening needs over time. By performing NKP decomposition on the target noise signal, multiple sub-noise signals are obtained, and the inverse sound waves of these sub-signals are calculated and fused, which can process the noise signal more finely and improve the accuracy and effect of noise reduction. At the same time, the inverse filtering frequency equalization processing further improves the quality of the inverse sound waves to ensure the maximum noise reduction effect. From the initialization of the acoustic model to the final model parameter compensation, a closed-loop optimization system is formed. This not only improves the noise reduction effect, but also helps to improve the overall performance of the headphones, thereby realizing the intelligent noise reduction calibration of wireless headphones and improving the noise reduction effect of wireless headphones.

[0092] The present application also provides a noise reduction calibration device for wireless headphones, the noise reduction calibration device for wireless headphones comprising a memory and a processor, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the noise reduction calibration method for wireless headphones in the above-mentioned embodiments.

[0093] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the noise reduction calibration method for wireless headphones.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0096] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A noise reduction calibration method for wireless headphones, characterized in that: The noise reduction calibration method of the wireless headset comprises: Initializing the internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model; specifically comprising: simulating the internal acoustic characteristics of the wireless headset through ANSYS Acoustics software, and calculating the acoustic characteristic parameter set of the wireless headset at different frequencies; creating a left internal acoustic processing model corresponding to the left headset of the wireless headset, and creating a right internal acoustic processing model corresponding to the right headset of the wireless headset, wherein the left internal acoustic processing model has the same model architecture as the right internal acoustic processing model, and the left internal acoustic processing model or the right internal acoustic processing model includes: a noise recognition network, an anti-phase sound wave synthesis network, and an anti-phase sound wave fusion network; generating a left initialization network hyperparameter set and a right initialization network hyperparameter set through the acoustic characteristic parameter set; initializing the network hyperparameters of the left internal acoustic processing model through the left initialization network hyperparameter set to obtain a first initial headset acoustic model, and initializing the network hyperparameters of the right internal acoustic processing model through the right initialization network hyperparameter set to obtain a second initial headset acoustic model; Collecting a test environment noise signal and performing noise filtering on the test environment noise signal to generate a target noise signal; Performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals; The method comprises: performing inverse phase sound wave calculation and fusion on the plurality of sub-noise signals respectively through the first initial headphone acoustic model and the second initial headphone acoustic model to obtain a first initial inverse phase sound wave and a second initial inverse phase sound wave; specifically comprising: inputting the plurality of sub-noise signals into a noise recognition network in the first initial headphone acoustic model respectively, and extracting a first noise feature vector of each sub-noise signal through the noise recognition network, wherein the noise recognition network comprises: a first double-layer LSTM network, a first GRU network and a first linear normalization layer; inputting the first noise feature vector of each sub-noise signal into the first initial headphone acoustic model respectively The inverse sound wave synthesis network in the acoustic model is used to synthesize the first sub-inverse sound wave of each sub-noise signal respectively through the inverse sound wave synthesis network, and the inverse sound wave synthesis network includes: a first encoder and a first decoder, wherein the first encoder includes a first input layer and a first two-layer CNN network, and the first decoder includes a first RNN network and a first two-layer residual network in sequence; the first sub-inverse sound wave of each sub-noise signal is respectively input into the inverse sound wave fusion network in the first initial headphone acoustic model, and the first sub-inverse sound wave of each sub-noise signal is inversely fused through the inverse sound wave fusion network to obtain the first An initial anti-phase sound wave, the anti-phase sound wave fusion network includes: a first attention weighted fusion layer and a first output layer; the multiple sub-noise signals are respectively input into the noise recognition network in the second initial headphone acoustic model, and the second noise feature vector of each sub-noise signal is extracted through the noise recognition network, the noise recognition network includes: a second double-layer LSTM network, a second GRU network and a second linear normalization layer; the second noise feature vector of each sub-noise signal is respectively input into the anti-phase sound wave synthesis network in the second initial headphone acoustic model, and each sub-noise signal is synthesized respectively through the anti-phase sound wave synthesis network The second sub-inverted sound wave of the sub-noise signal is synthesized into the inverted sound wave synthesis network, the inverted sound wave synthesis network includes: a second encoder and a second decoder, wherein the second encoder includes a second input layer and a second two-layer CNN network, and the second decoder includes a second RNN network and a second two-layer residual network in sequence; the second sub-inverted sound wave of each sub-noise signal is respectively input into the inverted sound wave fusion network in the second initial headphone acoustic model, and the second sub-inverted sound wave of each sub-noise signal is inverted through the inverted sound wave fusion network to obtain a second initial inverted sound wave, and the inverted sound wave fusion network includes: a second attention weighted fusion layer and a second output layer; The first initial inverted sound wave and the second initial inverted sound wave are subjected to inverse filtering frequency equalization processing to obtain a first target inverted sound wave and a second target inverted sound wave; specifically comprising: performing frequency segment intensity characteristic analysis on the first initial inverted sound wave to obtain a first frequency segment intensity characteristic; performing frequency segment intensity characteristic analysis on the second initial inverted sound wave to obtain a second frequency segment intensity characteristic; performing inverse filtering frequency intensity equalization adjustment on the first initial inverted sound wave according to the first frequency segment intensity characteristic to obtain a first target inverted sound wave; performing inverse filtering frequency intensity equalization adjustment on the second initial inverted sound wave according to the second frequency segment intensity characteristic to obtain a second target inverted sound wave; The first target antiphase sound wave and the second target antiphase sound wave are used to compensate the headphone acoustic model parameters to obtain the first target headphone acoustic model and the second target headphone acoustic model; specifically, the method comprises: respectively calculating the first noise reduction evaluation index of the first target antiphase sound wave and the second noise reduction evaluation index of the second target antiphase sound wave; defining the multi-objective fitness function of the genetic optimization algorithm based on the first noise reduction evaluation index and the second noise reduction evaluation index; initializing the noise reduction calibration compensation parameter population of the first initial headphone acoustic model and the second initial headphone acoustic model by the genetic optimization algorithm to obtain multiple first compensation parameter sets; calculating the fitness of the multiple first compensation parameter sets by the multi-objective fitness function to obtain the fitness value of each first compensation parameter set; generating multiple corresponding second compensation parameter sets according to the fitness values, and optimizing the multiple second compensation parameter sets to obtain the target compensation parameter set; compensating the headphone acoustic model parameters of the first initial headphone acoustic model and the second initial headphone acoustic model based on the target compensation parameter set to obtain the first target headphone acoustic model and the second target headphone acoustic model.

2. The noise reduction calibration method for wireless headphones according to claim 1, characterized in that: The collecting of the test environment noise signal and performing noise filtering on the test environment noise signal to generate a target noise signal includes: Collecting a test environment noise signal and defining a noise reduction frequency range of the wireless headset; Amplifying the test environment noise signal to obtain an amplified environment noise signal; Performing signal standardization processing on the amplified ambient noise signal to obtain a standard ambient noise signal; Perform high-frequency noise removal on the standard environmental noise signal according to the noise reduction frequency range to obtain a first noise signal; Remove low-frequency noise from the first noise signal according to the noise reduction frequency range to obtain a second noise signal; The second noise signal is subjected to noise enhancement processing to generate a target noise signal.

3. The noise reduction calibration method for wireless headphones according to claim 2, characterized in that: The NKP decomposition of the target noise signal to obtain a plurality of sub-noise signals includes: Performing a fast Fourier transform on the target noise signal to obtain a frequency domain noise signal; Performing feature recognition on the frequency domain noise signal to obtain a noise feature set; Decomposing feature points of the target noise signal according to the noise feature set to obtain a plurality of signal decomposition feature points; Performing signal decomposition interval prediction on the multiple signal decomposition feature points to obtain multiple signal decomposition feature intervals; The target noise signal is decomposed based on the multiple signal decomposition feature intervals to obtain multiple sub-noise signals.

4. A noise reduction calibration system for wireless headphones, characterized in that: The noise reduction calibration system for wireless headphones comprises: An initialization module is used to initialize the internal acoustic model of the wireless headset to obtain a first initial headset acoustic model and a second initial headset acoustic model; specifically comprising: simulating the internal acoustic characteristics of the wireless headset through ANSYS Acoustics software, and calculating the acoustic characteristic parameter set of the wireless headset at different frequencies; creating a left internal acoustic processing model corresponding to the left headset of the wireless headset, and creating a right internal acoustic processing model corresponding to the right headset of the wireless headset, wherein the left internal acoustic processing model has the same model architecture as the right internal acoustic processing model, and the left internal acoustic processing model or the right internal acoustic processing model includes: a noise recognition network, an anti-phase sound wave synthesis network, and an anti-phase sound wave fusion network; generating a left initialization network hyperparameter set and a right initialization network hyperparameter set through the acoustic characteristic parameter set; initializing the network hyperparameters of the left internal acoustic processing model through the left initialization network hyperparameter set to obtain a first initial headset acoustic model, and initializing the network hyperparameters of the right internal acoustic processing model through the right initialization network hyperparameter set to obtain a second initial headset acoustic model; A filtering module is used to collect a test environment noise signal and perform noise filtering on the test environment noise signal to generate a target noise signal; A decomposition module, used for performing NKP decomposition on the target noise signal to obtain multiple sub-noise signals; A calculation module, used to respectively calculate and fuse the inverse phase sound waves of the plurality of sub-noise signals through the first initial headphone acoustic model and the second initial headphone acoustic model to obtain the first initial inverse phase sound wave and the second initial inverse phase sound wave; specifically comprising: respectively inputting the plurality of sub-noise signals into the noise recognition network in the first initial headphone acoustic model, and extracting the first noise feature vector of each sub-noise signal through the noise recognition network, wherein the noise recognition network comprises: a first double-layer LSTM network, a first GRU network and a first linear normalization layer; respectively inputting the first noise feature vector of each sub-noise signal into the first An anti-phase sound wave synthesis network in an initial headphone acoustic model is used to synthesize the first sub-anti-phase sound wave of each sub-noise signal respectively through the anti-phase sound wave synthesis network, and the anti-phase sound wave synthesis network includes: a first encoder and a first decoder, wherein the first encoder includes a first input layer and a first two-layer CNN network, and the first decoder includes a first RNN network and a first two-layer residual network in sequence; the first sub-anti-phase sound wave of each sub-noise signal is input into the anti-phase sound wave fusion network in the first initial headphone acoustic model respectively, and the first sub-anti-phase sound wave of each sub-noise signal is fused through the anti-phase sound wave fusion network , obtain a first initial anti-phase sound wave, the anti-phase sound wave fusion network includes: a first attention weighted fusion layer and a first output layer; respectively input the multiple sub-noise signals into the noise recognition network in the second initial headphone acoustic model, and extract the second noise feature vector of each sub-noise signal through the noise recognition network, the noise recognition network includes: a second double-layer LSTM network, a second GRU network and a second linear normalization layer; respectively input the second noise feature vector of each sub-noise signal into the anti-phase sound wave synthesis network in the second initial headphone acoustic model, and synthesize each sub-noise signal through the anti-phase sound wave synthesis network The second sub-inverted sound wave of the signal, the inverted sound wave synthesis network includes: a second encoder and a second decoder, wherein the second encoder includes a second input layer and a second two-layer CNN network, and the second decoder includes a second RNN network and a second two-layer residual network in sequence; the second sub-inverted sound wave of each sub-noise signal is respectively input into the inverted sound wave fusion network in the second initial headphone acoustic model, and the second sub-inverted sound wave of each sub-noise signal is inverted through the inverted sound wave fusion network to obtain a second initial inverted sound wave, and the inverted sound wave fusion network includes: a second attention weighted fusion layer and a second output layer; An equalization module is used to perform inverse filtering frequency equalization processing on the first initial inverse phase sound wave and the second initial inverse phase sound wave to obtain a first target inverse phase sound wave and a second target inverse phase sound wave; specifically comprising: performing frequency segment intensity characteristic analysis on the first initial inverse phase sound wave to obtain a first frequency segment intensity characteristic; performing frequency segment intensity characteristic analysis on the second initial inverse phase sound wave to obtain a second frequency segment intensity characteristic; performing inverse filtering frequency intensity equalization adjustment on the first initial inverse phase sound wave according to the first frequency segment intensity characteristic to obtain a first target inverse phase sound wave; performing inverse filtering frequency intensity equalization adjustment on the second initial inverse phase sound wave according to the second frequency segment intensity characteristic to obtain a second target inverse phase sound wave; A compensation module is used to compensate the headphone acoustic model parameters through the first target antiphase sound wave and the second target antiphase sound wave to obtain the first target headphone acoustic model and the second target headphone acoustic model; specifically comprising: respectively calculating the first noise reduction evaluation index of the first target antiphase sound wave and the second noise reduction evaluation index of the second target antiphase sound wave; defining the multi-objective fitness function of the genetic optimization algorithm based on the first noise reduction evaluation index and the second noise reduction evaluation index; initializing the noise reduction calibration compensation parameter population of the first initial headphone acoustic model and the second initial headphone acoustic model through the genetic optimization algorithm to obtain multiple first compensation parameter sets; performing fitness calculation on the multiple first compensation parameter sets through the multi-objective fitness function to obtain the fitness value of each first compensation parameter set; generating multiple corresponding second compensation parameter sets according to the fitness values, and optimizing the multiple second compensation parameter sets to obtain the target compensation parameter set; performing headphone acoustic model parameter compensation on the first initial headphone acoustic model and the second initial headphone acoustic model based on the target compensation parameter set to obtain the first target headphone acoustic model and the second target headphone acoustic model.

5. A noise reduction calibration device for wireless headphones, characterized in that: The noise reduction calibration device for wireless headphones includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the noise reduction calibration device of the wireless headset to execute the noise reduction calibration method of the wireless headset according to any one of claims 1 to 3.

6. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the noise reduction calibration method for the wireless headset according to any one of claims 1 to 3 is implemented.

Images