Noise reduction method, system and electronic device for high-performance TWS Bluetooth audio chip

Abstract

The present application discloses a noise reduction method, system and electronic device for a high-performance TWS Bluetooth audio chip, which uses a motion augmented feature map of a motion differential feature map to improve the expression capability of the motion differential feature map for the relative motion information between a mobile phone and a Bluetooth headset. Meanwhile, considering that the feature distribution of the motion differential feature map in the time dimension in the corresponding high-dimensional feature space is relatively sparse, a motion feature space that is more suitable for reflecting the relative motion characteristics between the mobile phone and the Bluetooth headset is constructed in the present application. In this way, by increasing the density of the feature distribution, the regression operation performed by the decoder model can achieve better results.

Description

Technical Field

[0001] The present invention relates to the field of intelligent audio noise reduction, and more specifically, to a noise reduction method, system and electronic device for a high-performance TWS Bluetooth audio chip. Background Art

[0002] The full name of TWS is True Wireless Stereo, which means true wireless stereo. TWS technology is developed based on Bluetooth chip technology. Its working principle is that the mobile phone connects to the main earphone, and then the main earphone quickly connects to the auxiliary earphone wirelessly, realizing the true Bluetooth left and right channel wireless separation. That is, the audio data is first transmitted from the mobile phone to the main earphone, and then from the main earphone to the auxiliary earphone.

[0003] At present, for example, in a scenario where a user uses a TWS Bluetooth headset while exercising, there is usually a large amount of signal noise in the surrounding environment, which will interfere with the audio transmission quality. In addition to general environmental noise, there is also motion noise caused by the asynchronous movement between the Bluetooth transmitter, that is, the receiving end of the mobile phone and the Bluetooth headset, and the Bluetooth headset.

[0004] Therefore, in order to make the TWS Bluetooth headset achieve satisfactory results for people, a noise reduction solution for high-performance TWS Bluetooth audio chips is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a noise reduction method, system and electronic device for a high-performance TWS Bluetooth audio chip, which uses a motion augmentation feature map of a motion differential feature map to improve the expression ability of the motion differential feature map for the relative motion information between the mobile phone and the Bluetooth headset. At the same time, considering that the motion differential feature map has a relatively sparse feature distribution in the time dimension in the high-dimensional feature space to which it corresponds, a motion feature space that is more suitable for reflecting the relative motion characteristics between the mobile phone and the Bluetooth headset is constructed in the present application. In this way, by increasing the density of the feature distribution, the regression operation performed by the decoder model can achieve better results.

[0006] According to one aspect of the present application, a noise reduction method for a high-performance TWS Bluetooth audio chip is provided, which includes:

[0007] Obtaining motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of the X-axis, the Y-axis and the Z-axis;

[0008] Calculate the difference between the motion information of the mobile phone and the Bluetooth headset at the same time point to obtain a plurality of nine-element arrays, wherein the nine-element arrays include the difference in displacement, velocity, and acceleration in the directions of the X-axis, the Y-axis, and the Z-axis;

[0009] Constructing the plurality of nine-element arrays into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points and the number of columns is 9;

[0010] Using a convolutional neural network to obtain a motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix;

[0011] Performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map;

[0012] Obtaining a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset;

[0013] The first audio signal is converted into a first audio vector and then passed through an encoder model to obtain a transmission feature vector;

[0014] Convert the second audio signal into a second audio vector and pass it through the encoder model to obtain a received feature vector;

[0015] Calculating a difference between the transmitted feature vector and the received feature vector to obtain a differential feature vector;

[0016] multiplying the differential feature vector as a query vector with the motion augmented feature map to obtain a compensated feature vector;

[0017] Inputting the compensated feature vector into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model has a symmetrical structure with the encoder model; and

[0018] A denoised audio signal is generated based on the denoised frequency vector for compensation and the transmission feature vector.

[0019] According to another aspect of the present application, a noise reduction system for a high-performance TWS Bluetooth audio chip is provided, comprising:

[0020] A motion information acquisition unit, used to obtain motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of X-axis, Y-axis and Z-axis;

[0021] a motion information difference calculation unit, used to calculate the difference between the motion information of the mobile phone obtained by the motion information acquisition unit and the motion information of the Bluetooth headset obtained by the motion information acquisition unit at the same time point, so as to obtain a plurality of nine-element arrays, wherein the nine-element arrays include the difference in displacement, velocity and acceleration in the directions of the X-axis, the Y-axis and the Z-axis;

[0022] a motion difference matrix construction unit, configured to construct the plurality of nine-element arrays obtained by the motion information difference calculation unit into a motion difference matrix according to a time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9;

[0023] A convolutional neural network processing unit, configured to obtain a motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix obtained by the motion difference matrix construction unit using a convolutional neural network;

[0024] A motion augmentation feature map generating unit, configured to perform data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map obtained by the convolutional neural network processing unit in the channel dimension, so as to obtain a motion augmentation feature map;

[0025] An audio signal acquisition unit, used to acquire a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset;

[0026] a sending feature vector generating unit, configured to convert the first audio signal obtained by the audio signal obtaining unit into a first audio vector and then obtain a sending feature vector through an encoder model;

[0027] a receiving feature vector generating unit, configured to convert the second audio signal obtained by the audio signal obtaining unit into a second audio vector and then pass the second audio vector through the encoder model to obtain a receiving feature vector;

[0028] a feature vector difference calculation unit, configured to calculate a difference between the sending feature vector obtained by the sending feature vector generation unit and the receiving feature vector obtained by the receiving feature vector generation unit to obtain a differential feature vector;

[0029] a compensation feature vector generating unit, configured to multiply the differential feature vector obtained by the feature vector difference calculating unit as a query vector with the motion augmented feature map obtained by the motion augmented feature map generating unit to obtain a compensation feature vector;

[0030] a denoised frequency vector generating unit, configured to input the compensated feature vector obtained by the compensated feature vector generating unit into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model has a symmetrical structure with the encoder model; and

[0031] The denoised audio signal generating unit generates a denoised audio signal based on the denoised frequency vector for compensation obtained by the denoised frequency vector generating unit and the transmission feature vector obtained by the transmission feature vector generating unit.

[0032] According to another aspect of the present application, an electronic device is provided, comprising: a processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the noise reduction method for a high-performance TWS Bluetooth audio chip as described above.

[0033] According to yet another aspect of the present application, a computer-readable medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the noise reduction method for a high-performance TWS Bluetooth audio chip as described above.

[0034] Compared with the prior art, the noise reduction method, system and electronic device for a high-performance TWS Bluetooth audio chip provided in the present application use a motion augmented feature map of the motion differential feature map to improve the expression capability of the motion differential feature map for the relative motion information between the mobile phone and the Bluetooth headset. Meanwhile, considering that the feature distribution of the motion differential feature map in the time dimension in the corresponding high-dimensional feature space is relatively sparse, a motion feature space that is more suitable for reflecting the relative motion characteristics between the mobile phone and the Bluetooth headset is constructed in the present application. In this way, by increasing the density of the feature distribution, the regression operation performed by the decoder model can achieve better results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0036] Figure 1 This is an application scenario diagram of a noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0037] Figure 2It is a flowchart of a noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0038] Figure 3 A schematic diagram of the system architecture of a noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0039] Figure 4 In the noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application, data augmentation is performed on each feature matrix of the motion difference feature map in the channel dimension based on constructing Gaussian distribution and random sampling to obtain a flow chart of the motion augmentation feature map;

[0040] Figure 5 A flowchart of converting the first audio signal into a first audio vector and then obtaining a sending feature vector through an encoder model in a noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0041] Figure 6 It is a block diagram of a noise reduction system for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0042] Figure 7 It is a block diagram of a motion augmentation feature map generation unit in a noise reduction system for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0043] Figure 8 A block diagram of a sending feature vector generation unit in a noise reduction system for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application;

[0044] Fig. 9 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0046] Scenario Overview

[0047] As mentioned earlier, in the scenario where the user uses TWS Bluetooth headsets while exercising, there is usually a large amount of signal noise in the surrounding environment, which will interfere with the audio transmission quality. In addition to general environmental noise, there is also motion noise caused by the Bluetooth transmitter, that is, the receiving end of the mobile phone and the Bluetooth headset, and the asynchronous movement between the Bluetooth headsets. Therefore, in order to make the TWS Bluetooth headsets achieve satisfactory results, a noise reduction solution for high-performance TWS Bluetooth audio chips is expected.

[0048] Based on this, in the technical solution of the present application, firstly, the inertial sensors in the mobile phone and the Bluetooth headset are used to obtain the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process, including the displacement, velocity and acceleration in the three directions of X, Y and Z. Then, the difference of these motion information between the mobile phone and the Bluetooth headset is calculated respectively to obtain the nine-element array of (Δdx, Δdy, Δdz, Δvx, Δvy, Δvz, Δax, Δay, Δaz).

[0049] The above values ​​correspond to the three dimensions of distance, speed and acceleration respectively. First, they are normalized to their maximum value, that is, each value is mapped to the space between [0,1] according to its maximum value, and then the motion difference matrix is ​​obtained according to each time point. Here, the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9.

[0050] Next, the motion difference matrix is ​​input into the convolutional neural network to obtain a motion difference feature map, thereby extracting the dimensions of different motion data and the high-dimensional correlation features between different motion sampling times. Here, the convolutional neural network does not change the scale of the motion difference matrix, so the number of rows of each feature matrix in the motion difference feature map is still the number of time points, and the number of columns is still 9. In this way, a Gaussian distribution with the mean and variance of the nine eigenvalues ​​of each row in each feature matrix can be constructed, and then the Gaussian distribution is sampled to obtain more data points that conform to the Gaussian distribution to obtain a motion augmented feature map of the motion difference feature map.

[0051] Then, the audio vectors of the audio signal sent by the mobile phone and the audio vectors of the audio signal received by the Bluetooth headset are obtained respectively. First, they are converted to a high-dimensional feature space through an encoder model composed of multiple fully connected layers to obtain a sending feature vector and a receiving feature vector, and then the difference between the sending feature vector and the receiving feature vector is calculated to obtain a differential feature vector. Then, the differential feature vector is used as a query vector and multiplied by the motion augmented feature map to obtain a compensation feature vector. Finally, the compensation feature vector is input into a decoder model symmetrical to the encoder model to obtain a denoised frequency vector for compensation.

[0052] Based on this, the present application proposes a noise reduction method for a high-performance TWS Bluetooth audio chip, which includes: obtaining motion information of the mobile phone and the Bluetooth headset at a series of time points during the movement process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of X-axis, Y-axis and Z-axis; calculating the difference between the motion information of the mobile phone and the Bluetooth headset at the same time point to obtain a plurality of nine-element arrays, wherein the nine-element array includes the difference in displacement, velocity and acceleration in the directions of X-axis, Y-axis and Z-axis; constructing the plurality of nine-element arrays into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9; using a convolutional neural network to obtain a motion difference feature map for representing the dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix; The invention relates to a method for generating a motion augmented feature map, comprising: performing data augmentation on each feature matrix of the feature map in the channel dimension based on constructing a Gaussian distribution and random sampling to obtain a motion augmented feature map; obtaining a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset; converting the first audio signal into a first audio vector and then passing it through an encoder model to obtain a sending feature vector; converting the second audio signal into a second audio vector and then passing it through the encoder model to obtain a receiving feature vector; calculating the difference between the sending feature vector and the receiving feature vector to obtain a differential feature vector; multiplying the differential feature vector with the motion augmented feature map as a query vector to obtain a compensation feature vector; inputting the compensation feature vector into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model and the encoder model have a symmetrical structure; and generating a denoised audio signal based on the denoised frequency vector for compensation and the sending feature vector.

[0053] Figure 1 The figure shows an application scenario diagram of the noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application. Figure 1 As shown, in this application scenario, first, by installing on a mobile phone (for example, Figure 1 ) and the master earphone of the Bluetooth headset (e.g., Figure 1 The inertial sensor (e.g., P2) in FIG. Figure 1T) as shown in the figure obtains the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process, and obtains the first audio signal sent by the mobile phone through the mobile phone terminal and obtains the second audio signal received by the Bluetooth headset through the Bluetooth headset terminal, wherein the motion information includes the displacement, speed and acceleration in the three directions of the X-axis, Y-axis and Z-axis. Of course, in other application scenarios, other sensors can also be used to obtain the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process, which is not limited to this application.

[0054] It should be understood that the main earphone of the Bluetooth headset is also equipped with an audio chip (for example, Figure 1 C), wherein the audio chip can perform noise reduction processing on the received second audio signal, and transmit the audio signal to the auxiliary earphone of the Bluetooth headset (for example, Figure 1 P3 shown in Figure 1).

[0055] Then, the obtained motion information and the first and second audio signals are input into a server that is deployed with a noise reduction algorithm for a high-performance TWS Bluetooth audio chip (for example, Figure 1 S) as shown in , wherein the server is capable of processing the motion information and the first and second audio signals using a noise reduction algorithm for a high-performance TWS Bluetooth audio chip to generate a denoised audio signal.

[0056] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0057] Exemplary Methods

[0058] Figure 2 The figure shows a flow chart of a noise reduction method for a high-performance TWS Bluetooth audio chip. Figure 2As shown, according to the noise reduction method for a high-performance TWS Bluetooth audio chip of an embodiment of the present application, the method includes: S110, obtaining motion information of the mobile phone and the Bluetooth headset at a series of time points during the movement process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of X-axis, Y-axis and Z-axis; S120, calculating the difference between the motion information of the mobile phone and the Bluetooth headset at the same time point to obtain a plurality of nine-element arrays, wherein the nine-element array includes the difference in displacement, velocity and acceleration in the directions of X-axis, Y-axis and Z-axis; S130, constructing the plurality of nine-element arrays into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9; S140, using a convolutional neural network to obtain a motion difference feature map for representing the dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix; S150, performing convolution on the motion difference feature map in Each feature matrix on the channel dimension is augmented based on constructing Gaussian distribution and random sampling to obtain a motion augmented feature map; S160, obtaining a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset; S170, converting the first audio signal into a first audio vector and then passing it through an encoder model to obtain a sending feature vector; S180, converting the second audio signal into a second audio vector and then passing it through the encoder model to obtain a receiving feature vector; S190, calculating the difference between the sending feature vector and the receiving feature vector to obtain a differential feature vector; S200, multiplying the differential feature vector as a query vector with the motion augmented feature map to obtain a compensation feature vector; S210, inputting the compensation feature vector into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model and the encoder model have a symmetrical structure; and, S220, generating a denoised audio signal based on the denoised frequency vector for compensation and the sending feature vector.

[0059] Figure 3 The figure shows a schematic diagram of the architecture of the noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application. Figure 3 As shown, in the network architecture of the noise reduction method for the high-performance TWS Bluetooth audio chip, first, the motion information of the mobile phone at the same time point is calculated (for example, Figure 3 P1 shown in FIG) and the motion information of the Bluetooth headset (for example, Figure 3 to obtain a plurality of nine-element arrays (e.g., Figure 3 A shown in FIG); S130, constructing the plurality of nine-element arrays into motion difference matrices according to the time dimension (for example, Figure 3 M as shown in FIG); S140, using a convolutional neural network (e.g., Figure 3 The CNN shown in FIG. 10 ) obtains a motion difference feature map (for example, as shown in FIG. 10 ) for representing the dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix. Figure 3 S150, performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map (for example, Figure 3 S170, obtaining the first audio signal sent by the mobile phone (for example, Figure 3 Q1) is converted into a first audio vector (e.g., Figure 3 V1 as shown in FIG. 1 ) and then passes through an encoder model (e.g., Figure 3 E as shown in FIG. 1 ) to obtain a transmission feature vector (e.g., Figure 3 S180, obtaining the second audio signal received by the Bluetooth headset (for example, Figure 3 Q2) is converted into a second audio vector (e.g., Figure 3 V2) is then passed through the encoder model to obtain a received feature vector (e.g., Figure 3 S190, calculating the difference between the transmitted feature vector and the received feature vector to obtain a differential feature vector (for example, Figure 3 S200, multiplying the differential feature vector as a query vector with the motion augmented feature map to obtain a compensated feature vector (for example, Figure 3 S210, input the compensated feature vector into a decoder model (e.g., Figure 3 D as shown in FIG) to obtain a de-noised frequency vector for compensation (eg, as Figure 3 V), wherein the decoder model and the encoder model have a symmetrical structure; and, S220, based on the denoised frequency vector for compensation and the transmitted feature vector, generating a denoised audio signal (for example, Figure 3 OUT as shown in the figure.

[0060] In step S110 and step S120, the motion information of the mobile phone and the Bluetooth headset at a series of time points during the movement is obtained by inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of the X-axis, the Y-axis and the Z-axis, and the difference between the motion information of the mobile phone and the Bluetooth headset at the same time point is calculated to obtain a plurality of nine-element arrays, wherein the nine-element array includes the difference in displacement, velocity and acceleration in the directions of the X-axis, the Y-axis and the Z-axis. As described above, the present application takes into account the specific usage scenario of high-performance TWS Bluetooth headsets, that is, the scenario where the user uses the Bluetooth headset while exercising. In addition to general environmental noise, motion noise is also generated due to the Bluetooth transmitter, that is, the transmitter of the mobile phone and the receiver of the Bluetooth headset, and the asynchronous movement between the Bluetooth headset. Therefore, it is expected to be able to model and reduce noise by applying compensating audio.

[0061] Therefore, in the technical solution of the present application, first, it is necessary to obtain the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process. Here, the motion information includes the displacement, velocity and acceleration in the three directions of the X-axis, Y-axis and Z-axis. In a specific example, the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process can be obtained by inertial sensors installed in the mobile phone and the Bluetooth headset. Of course, in other specific examples, other sensors can also be used to obtain the motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process, which is not limited to this application. Then, the difference between these motion information at the same time points between the mobile phone and the Bluetooth headset is calculated respectively to obtain a nine-element array of (Δdx, Δdy, Δdz, Δvx, Δvy, Δvz, Δax, Δay, Δaz).

[0062] In step S130, the multiple nine-element arrays are constructed into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9. That is, the data corresponding to the three dimensions of distance, speed and acceleration in the nine-element array are first normalized to the maximum value, that is, each value is mapped to the space between [0,1] according to its maximum value; then the multiple nine-element arrays after normalization are arranged according to each time point to obtain the motion difference matrix. It is worth mentioning that here, the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9.

[0063] Specifically, in the embodiment of the present application, the process of constructing the plurality of nine-element arrays into motion difference matrices according to the time dimension includes: first, performing a normalization process based on the maximum value on each value in the plurality of nine-element arrays, so as to map each value in the plurality of nine-element arrays to a space between [0, 1] according to its maximum value. Accordingly, in a specific example, the formula X can be used norm =(X i -X min ) / (X max -X min ) calculates the maximum value normalized data of each value in the plurality of nine-element arrays, where X i Represents each value in the multiple nine-element arrays. Then, the normalized multiple nine-element arrays are arranged into the motion difference matrix according to the time dimension. It should be understood that since distance, speed, and acceleration are data between three different units, such processing can not only eliminate the dimensional influence between data features, facilitate measurement and subsequent calculation, but also facilitate the subsequent training and convergence of the convolutional neural network.

[0064] In step S140, a motion difference feature map for representing the dimensions of different motion information and high-dimensional correlation features between different motion sampling times is obtained from the motion difference matrix using a convolutional neural network. That is, the motion difference matrix is ​​processed using a convolutional neural network to extract the dimensions of different motion data in the motion difference matrix and high-dimensional correlation features between different motion sampling times, thereby obtaining a motion difference feature map.

[0065] Specifically, in an embodiment of the present application, a process of using a convolutional neural network to obtain a motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix includes: the convolutional neural network obtains the motion difference feature map from the motion difference matrix using the following formula;

[0066] Wherein, the formula is:

[0067] f i =Sigmoid(N i ×f i-1 +B i )

[0068] Among them, f i-1 is the input of the i-th layer of convolutional neural network, f i is the output of the i-th layer of convolutional neural network, N i is the filter of the i-th layer of convolutional neural network, and B iis the bias vector of the i-th layer convolutional neural network, and Sigmoid represents the nonlinear activation function.

[0069] In step S150, data augmentation based on constructing Gaussian distribution and random sampling is performed on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map. It should be understood that, considering that the feature distribution of the motion difference feature map in the time dimension in the high-dimensional feature space to which it corresponds is relatively sparse, therefore, in the technical solution of the present application, the motion augmented feature map of the motion difference feature map is used to improve the expression ability of the motion difference feature map for the relative motion information between the mobile phone and the Bluetooth headset, and at the same time, a motion feature space that is more suitable for reflecting the relative motion characteristics between the mobile phone and the Bluetooth headset can be constructed to solve the feature expression problem used for modeling motion noise. In this way, by increasing the density of the feature distribution, the regression operation performed by the decoder model can achieve better results.

[0070] Specifically, in an embodiment of the present application, the process of performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map in the channel dimension to obtain the motion augmented feature map includes: first, based on the mean and variance of the nine eigenvalues ​​of each row in each of the feature matrices, constructing a Gaussian distribution with the variance and the mean. It should be understood that since the scale of the motion difference matrix is ​​not changed after being processed by the convolutional neural network, the number of rows of each feature matrix in the motion difference feature map is still the number of time points, and the number of columns is still 9. In this way, in the technical solution of the present application, a Gaussian distribution with the mean and variance can be constructed based on the mean and variance of the nine eigenvalues ​​of each row in each of the feature matrices. Then, the Gaussian distribution is randomly sampled to obtain more data points as the data points of a row of the feature matrix to obtain the motion augmented feature map. That is, the Gaussian distribution is randomly sampled to obtain more data points that conform to the Gaussian distribution to obtain the motion augmented feature map of the motion difference feature map.

[0071] Figure 4 The figure shows a flow chart of performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion differential feature map in the channel dimension in the noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application to obtain a motion augmented feature map. Figure 4As shown, in an embodiment of the present application, data augmentation based on constructing Gaussian distribution and random sampling is performed on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map, including: S310, based on the mean and variance of the nine eigenvalues ​​of each row in each of the feature matrices, constructing a Gaussian distribution with the variance and the mean; and, S320, randomly sampling the Gaussian distribution to obtain more data points as data points of a row of the feature matrix to obtain the motion augmented feature map.

[0072] In step S160, step S170 and step S180, the first audio signal sent by the mobile phone and the second audio signal received by the Bluetooth headset are obtained, and the first audio signal is converted into a first audio vector and then passed through the encoder model to obtain a sending feature vector, and then the second audio signal is converted into a second audio vector and then passed through the encoder model to obtain a receiving feature vector. It should be understood that the present application takes into account the specific use scenario of high-performance TWS Bluetooth headsets, that is, the scenario where the user uses the Bluetooth headset while exercising. In addition to general environmental noise, motion noise is also generated due to the asynchronous movement between the Bluetooth transmitter, that is, the mobile phone and the Bluetooth receiver, that is, the Bluetooth headset. Therefore, it is expected to be able to model and reduce noise by applying compensation audio. Therefore, in the technical solution of the present application, after obtaining the motion augmentation feature map, it is first necessary to obtain the first audio signal sent by the mobile phone from the mobile phone terminal and the second audio signal received by the Bluetooth headset from the Bluetooth headset terminal. Then, the first audio signal sent by the mobile phone and the second audio signal received by the Bluetooth headset are converted into the first audio vector and the second audio vector. Finally, the first audio vector and the second audio vector are passed through an encoder model composed of multiple fully connected layers to convert the first audio vector and the second audio vector into a high-dimensional feature space, thereby obtaining a sending feature vector and a receiving feature vector.

[0073] Specifically, in the embodiment of the present application, the process of converting the first audio signal into a first audio vector and then obtaining a transmission feature vector through an encoder model includes: first, using an analog-to-digital converter to convert the first audio signal into a first digital signal, so that the computer can process it later. That is, the obtained first audio signal is converted into a digital signal by analog-to-digital conversion, so as to convert the analog signal into a digital signal. Specifically, it includes two steps: sampling and quantization, that is, converting the continuous waveform of the sound into discrete data points at a certain sampling rate and sampling bit number.

[0074] Next, the first digital signal is subjected to spectrum analysis using a spectrum analyzer, and the audio analysis includes discrete Fourier transforming the first digital signal, and Mel filtering the first digital signal after discrete Fourier transforming. It should be understood that sound is an analog signal, and the time domain waveform of sound only represents the relationship between the sound pressure and time, and cannot well represent the characteristics of the sound. Therefore, in the technical solution of the present application, the sound waveform must be subjected to audio analysis to convert it into an acoustic feature vector. That is, in a specific example, the audio analysis includes: first, discrete Fourier transforming the first digital signal to convert the obtained time domain characteristics of the digitized first digital signal into the spectrum frequency domain. It should be understood that the discrete Fourier transform can extract the information of the discrete frequency band spectrum from a discrete signal. Then, Mel filtering is performed on the digitized first digital signal after discrete Fourier transforming. It should be understood that the sensitivity of human hearing to different frequency bands is different. The human ear is not as sensitive to high frequencies as low frequencies. This dividing line is about 1000Hz. Therefore, simulating this property of human hearing when extracting sound features can improve recognition performance. It is worth mentioning that here, the corresponding relationship between frequency (unit Hz) and Mel scale is linear below 1000Hz and logarithmic above 1000Hz. The calculation formula is as follows: Mel (f) = 1127ln (1 + f / 700).

[0075] Then, the first digital signal analyzed by the spectral analysis is processed by a parameter transformer to obtain the first audio vector. Finally, the first audio vector is fully connected encoded using multiple fully connected layers of the encoder model to convert it into a high-dimensional feature space, thereby obtaining the transmission feature vector.

[0076] Similarly, in a specific example, the process of converting the second audio signal into a second audio vector and then obtaining a received feature vector through the encoder model includes: first, using the analog-to-digital converter to convert the second audio signal into a second digital signal; then, using the spectrogram analyzer to perform spectrum analysis on the second digital signal, and the audio analysis includes performing a discrete Fourier transform on the second digital signal, and performing a Mel filter on the second digital signal after the discrete Fourier transform; then, using the parameter converter to process the second digital signal after the spectrum analysis to obtain the second audio vector; finally, using multiple fully connected layers of the encoder model to fully connect encode the second audio vector to obtain the received feature vector.

[0077] Figure 5The figure shows a flow chart of converting the first audio signal into a first audio vector and then passing the encoder model to obtain a transmission feature vector in a noise reduction method for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application. Figure 5 As shown, in an embodiment of the present application, the first audio signal is converted into a first audio vector and then passed through an encoder model to obtain a sending feature vector, including: S410, using an analog-to-digital converter to convert the first audio signal into a first digital signal to convert the currently allocated network resources of the multiple terminal devices into the range of 0 to 1; S420, using a spectrogram analyzer to perform spectrum analysis on the first digital signal, and the audio analysis includes discrete Fourier transforming the first digital signal, and Mel filtering the first digital signal after the discrete Fourier transform; S430, using a parameter converter to process the first digital signal after the spectrum analysis to obtain the first audio vector; and, S440, using multiple fully connected layers of the encoder model to fully connect the first audio vector to obtain the sending feature vector.

[0078] In step S190 and step S200, the difference between the transmitted feature vector and the received feature vector is calculated to obtain a differential feature vector, and the differential feature vector is used as a query vector to be multiplied with the motion augmentation feature map to obtain a compensation feature vector. That is, first, the difference between the transmitted feature vector and the received feature vector is calculated to obtain a differential feature vector representing the degree of difference between the signal transmitted by the mobile phone and the signal received by the Bluetooth headset in a high-dimensional space. Then, the differential feature vector is used as a query vector to be multiplied with the motion augmentation feature map to fuse the correlation characteristics between the asynchronous motion noise and the environmental loss noise, thereby obtaining a compensation feature vector.

[0079] In step S210 and step S220, the compensation feature vector is input into the decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model has a symmetrical structure with the encoder model, and based on the denoised frequency vector for compensation and the sending feature vector, a denoised audio signal is generated. That is, first, the compensation feature vector is processed by the decoder model to restore the denoised frequency vector for compensation. It is worth mentioning that, here, the decoder model has a symmetrical structure with the encoder model. Then, based on the denoised frequency vector for compensation and the sending feature vector, a denoised audio signal is generated. It should be understood that the present application can achieve better results in the regression operation performed by the decoder model by increasing the density of the feature distribution.

[0080] Specifically, in an embodiment of the present application, the process of generating a denoised audio signal based on the denoised frequency vector for compensation and the transmitted feature vector includes: first, calculating the positional sum of the denoised frequency vector for compensation and the transmitted feature vector to obtain a denoised feature vector; and then, converting the denoised feature vector into the denoised audio signal.

[0081] In summary, the noise reduction method for a high-performance TWS Bluetooth audio chip in an embodiment of the present application is explained, which improves the expression capability of the motion differential feature map for the relative motion information between the mobile phone and the Bluetooth headset by using the motion augmentation feature map of the motion differential feature map, and at the same time, considering that the feature distribution of the motion differential feature map in the time dimension in the corresponding high-dimensional feature space is relatively sparse, therefore, a motion feature space that is more suitable for reflecting the relative motion characteristics between the mobile phone and the Bluetooth headset is constructed in the present application, so that by increasing the density of the feature distribution, the regression operation performed by the decoder model can obtain better results.

[0082] Exemplary Systems

[0083] Figure 6 The figure shows a block diagram of a noise reduction system for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application. Figure 6As shown, according to the embodiment of the present application, the noise reduction system 600 for a high-performance TWS Bluetooth audio chip includes: a motion information acquisition unit 610, which is used to obtain the motion information of the mobile phone and the Bluetooth headset at a series of time points during the movement process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, speed and acceleration in three directions of X-axis, Y-axis and Z-axis; a motion information difference calculation unit 620, which is used to calculate the difference between the motion information of the mobile phone obtained by the motion information acquisition unit 610 and the motion information of the Bluetooth headset obtained by the motion information acquisition unit at the same time point, so as to obtain multiple nine-element arrays, wherein the nine-element array includes displacement in three directions of X-axis, Y-axis and Z-axis. The difference in displacement, velocity and acceleration of the motion information difference calculation unit 620; a motion difference matrix construction unit 630, used to construct the multiple nine-element arrays obtained by the motion information difference calculation unit 620 into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9; a convolutional neural network processing unit 640, used to use a convolutional neural network to obtain a motion difference feature map for representing the dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix obtained by the motion difference matrix construction unit 630; a motion augmented feature map generation unit 650, used to generate a motion difference feature map obtained by the convolutional neural network processing unit 640 in the channel The feature matrices on the dimension are augmented based on constructing Gaussian distribution and random sampling to obtain motion augmented feature maps; an audio signal acquisition unit 660 is used to obtain a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset; a sending feature vector generation unit 670 is used to convert the first audio signal obtained by the audio signal acquisition unit 660 into a first audio vector and then pass it through an encoder model to obtain a sending feature vector; a receiving feature vector generation unit 680 is used to convert the second audio signal obtained by the audio signal acquisition unit 660 into a second audio vector and then pass it through the encoder model to obtain a receiving feature vector; a feature vector difference calculation unit 690 is used to calculate the sending feature vector. The difference between the sent feature vector obtained by the sending feature vector generating unit 670 and the received feature vector obtained by the receiving feature vector generating unit 680 is used to obtain a differential feature vector; the compensation feature vector generating unit 700 is used to multiply the differential feature vector obtained by the feature vector difference calculating unit 690 with the motion augmented feature map obtained by the motion augmented feature map generating unit 650 as a query vector to obtain a compensation feature vector; the denoised frequency vector generating unit 710 is used to input the compensated feature vector obtained by the compensated frequency vector generating unit 700 into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model and the encoder model have a symmetrical structure;And, the denoised audio signal generating unit 720 generates a denoised audio signal based on the denoised frequency vector for compensation obtained by the denoised frequency vector generating unit 710 and the transmission feature vector obtained by the transmission feature vector generating unit. ;

[0084] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, the motion difference matrix construction unit 630 includes: a normalization subunit, used to perform normalization processing based on the maximum value on each value in the multiple nine-element arrays; and a motion difference matrix arrangement subunit, used to arrange the multiple nine-element arrays obtained by the normalization subunit into the motion difference matrix in rows according to the time dimension.

[0085] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, the convolutional neural network processing unit 640 is further used for: the convolutional neural network obtains the motion difference feature map from the motion difference matrix using the following formula; wherein the formula is:

[0086] f i =Sigmoid(N i ×f i-1 +B i )

[0087] Among them, f i-1 is the input of the i-th layer of convolutional neural network, f i is the output of the i-th layer of convolutional neural network, N i is the filter of the i-th layer of convolutional neural network, and B i is the bias vector of the i-th layer convolutional neural network, and Sigmoid represents the nonlinear activation function.

[0088] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, as Figure 7 As shown, the motion augmented feature map generation unit 650 includes: a Gaussian distribution construction subunit 651, which is used to construct a Gaussian distribution with the variance and the mean based on the mean and variance of the nine eigenvalues ​​of each row in each of the feature matrices; and a sampling subunit 652, which is used to randomly sample the Gaussian distribution obtained by the Gaussian distribution construction subunit 651 to obtain more data points as data points of a row of the feature matrix to obtain the motion augmented feature map.

[0089] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, as Figure 8As shown, the sending feature vector generating unit 670 includes: a digital signal conversion subunit 671, which is used to convert the first audio signal into a first digital signal using an analog-to-digital converter; a spectrum analysis subunit 672, which is used to perform spectrum analysis on the first digital signal obtained by the digital signal conversion subunit 671 using a spectrum analyzer, and the audio analysis includes discrete Fourier transforming the first digital signal, and Mel filtering the first digital signal after discrete Fourier transform; a parameter transformation subunit 673, which is used to process the first digital signal obtained by the spectrum analysis subunit 672 after spectrum analysis using a parameter transformer to obtain the first audio vector; and a fully connected encoding subunit 674, which is used to perform fully connected encoding on the first audio vector obtained by the parameter transformation subunit 673 using multiple fully connected layers of the encoder model to obtain the sending feature vector.

[0090] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, the receiving feature vector generating unit 680 is further used to: use the analog-to-digital converter to convert the second audio signal into a second digital signal; use the spectrogram analyzer to perform spectrum analysis on the second digital signal, and the audio analysis includes discrete Fourier transforming the second digital signal, and Mel filtering the second digital signal after the discrete Fourier transform; use the parameter converter to process the second digital signal after the spectrum analysis to obtain the second audio vector; and use multiple fully connected layers of the encoder model to fully connect the second audio vector to obtain the receiving feature vector.

[0091] In one example, in the above-mentioned noise reduction system 600 for a high-performance TWS Bluetooth audio chip, the denoised audio signal generating unit 720 is further used to: calculate the positional sum of the denoised frequency vector for compensation and the transmitted feature vector to obtain a denoised feature vector; and, convert the denoised feature vector into the denoised audio signal.

[0092] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the noise reduction system 600 for a high-performance TWS Bluetooth audio chip have been described above with reference to Figures 1 to 5 The noise reduction method for a high-performance TWS Bluetooth audio chip has been described in detail, and therefore, its repeated description will be omitted.

[0093] As described above, the noise reduction system 600 for a high-performance TWS Bluetooth audio chip according to an embodiment of the present application can be implemented in various terminal devices, such as the audio chip of the main earphone of a TWS Bluetooth headset.

[0094] Exemplary Electronic Devices

[0095] Below, reference Fig. 9 To describe the electronic device according to the embodiment of the present application. Fig. 9 As shown, the electronic device 10 includes one or more processors 11 and a memory 12. The processor 11 may be a central processing unit (CPU) or other processing units having data processing capability and / or instruction execution capability, and may control other components in the electronic device 10 to perform desired functions.

[0096] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the functions of the noise reduction method for high-performance TWS Bluetooth audio chip of the various embodiments of the present application described above and / or other desired functions. Various contents such as motion difference matrix, differential eigenvector, etc. may also be stored in the computer-readable storage medium.

[0097] In one example, the electronic device 10 may further include: an input system 13 and an output system 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0098] The input system 13 may include, for example, a keyboard, a mouse, and the like.

[0099] The output system 14 can output various information to the outside, including the denoised audio signal, etc. The output system 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0100] Of course, to simplify, Fig. 9 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0101] Exemplary computer program products and computer-readable storage media

[0102] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to perform the steps in the functions of the noise reduction method for a high-performance TWS Bluetooth audio chip according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0103] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0104] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, cause the processor to execute the steps of the noise reduction method for a high-performance TWS Bluetooth audio chip described in the above “Exemplary Method” section of this specification.

[0105] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0106] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0107] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0108] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0109] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0110] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A noise reduction method for a high-performance TWS Bluetooth audio chip, characterized in that: include: Obtaining motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of the X-axis, the Y-axis and the Z-axis; Calculate the difference between the motion information of the mobile phone and the Bluetooth headset at the same time point to obtain a plurality of nine-element arrays, wherein the nine-element arrays include the difference in displacement, velocity, and acceleration in the directions of the X-axis, the Y-axis, and the Z-axis; Constructing the plurality of nine-element arrays into a motion difference matrix according to the time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points and the number of columns is 9; Using a convolutional neural network to obtain a motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix; Performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map; Obtaining a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset; The first audio signal is converted into a first audio vector and then passed through an encoder model to obtain a transmission feature vector; Convert the second audio signal into a second audio vector and pass it through the encoder model to obtain a received feature vector; Calculating the difference between the transmitted feature vector and the received feature vector to obtain a differential feature vector; multiplying the differential feature vector as a query vector with the motion augmented feature map to obtain a compensated feature vector; Inputting the compensated feature vector into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model has a symmetrical structure with the encoder model; and A denoised audio signal is generated based on the denoised frequency vector for compensation and the transmission feature vector.

2. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 1, wherein: The plurality of nine-element arrays are constructed into a motion difference matrix according to a time dimension, including: performing a maximum value-based normalization process on each value in the plurality of nine-element arrays; and The normalized multiple nine-element arrays are arranged in rows along the time dimension to form the motion difference matrix.

3. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 1, wherein: A motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times is obtained from the motion difference matrix using a convolutional neural network, including: The convolutional neural network obtains the motion difference feature map from the motion difference matrix using the following formula; Wherein, the formula is: f i =Sigmoid(N i ×f i-1 +B i ) Among them, f i-1 is the input of the i-th layer of convolutional neural network, f i is the output of the i-th layer of convolutional neural network, N i is the filter of the i-th layer of convolutional neural network, and B i is the bias vector of the i-th layer convolutional neural network, and Sigmoid represents the nonlinear activation function.

4. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 3, wherein: Performing data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map in the channel dimension to obtain a motion augmented feature map, including: Based on the mean and variance of the nine eigenvalues ​​in each row of each of the feature matrices, construct a Gaussian distribution with the variance and the mean; and The Gaussian distribution is randomly sampled to obtain more data points as data points of a row of the feature matrix to obtain the motion augmented feature map.

5. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 1, wherein: The first audio signal is converted into a first audio vector and then passed through an encoder model to obtain a transmission feature vector, including: Convert the first audio signal into a first digital signal using an analog-to-digital converter; Performing spectrum analysis on the first digital signal using a spectrum analyzer, wherein the audio analysis includes performing discrete Fourier transform on the first digital signal, and performing Mel filtering on the first digital signal after discrete Fourier transform; Processing the first digital signal after spectral analysis using a parameter transformer to obtain the first audio vector; and The first audio vector is fully-connected encoded using multiple fully-connected layers of the encoder model to obtain the transmission feature vector.

6. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 5, wherein: The step of converting the second audio signal into a second audio vector and then passing the second audio signal through the encoder model to obtain a received feature vector includes: Converting the second audio signal into a second digital signal using the analog-to-digital converter; Using the spectrum analyzer to perform spectrum analysis on the second digital signal, the audio analysis includes performing discrete Fourier transform on the second digital signal, and performing Mel filtering on the second digital signal after discrete Fourier transform; Processing the second digital signal analyzed by the spectrum analyzer using the parameter transformer to obtain the second audio vector; and The second audio vector is fully-connected encoded using multiple fully-connected layers of the encoder model to obtain the received feature vector.

7. The noise reduction method for a high-performance TWS Bluetooth audio chip according to claim 1, wherein: Generating a denoised audio signal based on the denoised frequency vector for compensation and the sending feature vector, comprising: Calculating the position-wise sum of the denoised frequency vector for compensation and the transmitted feature vector to obtain a denoised feature vector; and The denoised feature vector is converted into the denoised audio signal.

8. A noise reduction system for a high-performance TWS Bluetooth audio chip, characterized in that: include: A motion information acquisition unit, used to obtain motion information of the mobile phone and the Bluetooth headset at a series of time points during the motion process through inertial sensors installed in the mobile phone and the Bluetooth headset, wherein the motion information includes displacement, velocity and acceleration in three directions of X-axis, Y-axis and Z-axis; a motion information difference calculation unit, used to calculate the difference between the motion information of the mobile phone obtained by the motion information acquisition unit and the motion information of the Bluetooth headset obtained by the motion information acquisition unit at the same time point, so as to obtain a plurality of nine-element arrays, wherein the nine-element arrays include the difference in displacement, velocity and acceleration in the directions of the X-axis, the Y-axis and the Z-axis; a motion difference matrix construction unit, configured to construct the plurality of nine-element arrays obtained by the motion information difference calculation unit into a motion difference matrix according to a time dimension, wherein the number of rows of the motion difference matrix is ​​the number of time points, and the number of columns is 9; A convolutional neural network processing unit, configured to obtain a motion difference feature map for representing dimensions of different motion information and high-dimensional correlation features between different motion sampling times from the motion difference matrix obtained by the motion difference matrix construction unit using a convolutional neural network; A motion augmentation feature map generating unit, configured to perform data augmentation based on constructing Gaussian distribution and random sampling on each feature matrix of the motion difference feature map obtained by the convolutional neural network processing unit in the channel dimension, so as to obtain a motion augmentation feature map; An audio signal acquisition unit, used to acquire a first audio signal sent by the mobile phone and a second audio signal received by the Bluetooth headset; a sending feature vector generating unit, configured to convert the first audio signal obtained by the audio signal obtaining unit into a first audio vector and then obtain a sending feature vector through an encoder model; a receiving feature vector generating unit, configured to convert the second audio signal obtained by the audio signal obtaining unit into a second audio vector and then pass the second audio vector through the encoder model to obtain a receiving feature vector; a feature vector difference calculation unit, configured to calculate a difference between the sending feature vector obtained by the sending feature vector generation unit and the receiving feature vector obtained by the receiving feature vector generation unit to obtain a differential feature vector; a compensation feature vector generating unit, configured to multiply the differential feature vector obtained by the feature vector difference calculating unit as a query vector with the motion augmented feature map obtained by the motion augmented feature map generating unit to obtain a compensation feature vector; a denoised frequency vector generating unit, configured to input the compensated feature vector obtained by the compensated feature vector generating unit into a decoder model to obtain a denoised frequency vector for compensation, wherein the decoder model has a symmetrical structure with the encoder model; and The denoised audio signal generating unit generates a denoised audio signal based on the denoised frequency vector for compensation obtained by the denoised frequency vector generating unit and the transmission feature vector obtained by the transmission feature vector generating unit.

9. The noise reduction system for a high-performance TWS Bluetooth audio chip according to claim 8, wherein: The motion difference matrix construction unit comprises: a normalization subunit, configured to perform normalization processing based on a maximum value on each value in the plurality of nine-element arrays; and The motion difference matrix arrangement subunit is used to arrange the multiple nine-element arrays obtained by the normalization subunit in normalization processing into the motion difference matrix in rows according to the time dimension.

10. An electronic device comprising: processor; as well as A memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the noise reduction method for a high-performance TWS Bluetooth audio chip as described in any one of claims 1-7.

Images