Noise reduction method for constructing secondary channel estimation by using neural network, computer readable storage medium and electronic device

Abstract

The present application relates to a kind of denoising method for constructing secondary channel estimation using neural network, computer readable storage medium, electronic equipment, wherein the method comprises: constructing second neural network model, based on adaptive feedforward active noise reduction architecture, using second neural network model as the secondary channel estimation S' (z) in the architecture.The present application is realized by using S' (z) as second neural network model, so that the estimation of secondary channel is more accurate, and the harmonic and intermodulation distortion phenomenon caused by the nonlinear element existing in the propagation path from B bit loudspeaker to C bit error microphone is improved.

Description

[0001] This patent is a divisional application of invention No. 201911056457.3, "Adaptive Feedforward Active Noise Reduction Method Based on Neural Network, Computer-readable Storage Medium, Electronic Device" (application date 2019-10-31). Technical Field

[0002] This invention relates to headphone noise reduction, and more particularly to a noise reduction method that utilizes a neural network to construct a secondary channel estimate, a computer-readable storage medium, and an electronic device. Background Technology

[0003] See Figure 1 The basic principle of adaptive feedforward active noise cancellation architecture in headphones is as follows: At position A, the reference microphone picks up the original noise signal in the environment, and generates a signal that is opposite to the original noise signal (referred to as the reverse noise signal) through the feedforward filter. Then, the reverse noise signal is output through the speaker at position B, so that the original noise signal and the reverse noise signal cancel each other out at position B to generate a residual noise signal. The above process is adaptive feedforward active noise cancellation.

[0004] Based on adaptive feedforward active noise reduction, an error microphone at position C is added. The residual noise signal is collected by the error microphone at position C, and after analysis by the LMS algorithm, a weighting coefficient is generated to adjust the inverse noise signal output by the feedforward filter, thus realizing adaptive feedforward active noise reduction.

[0005] In adaptive feedforward active noise cancellation, the feedforward filter is generally implemented using an FIR (Finite Impulse Response) filter or an IIR (Infinite Impulse Response) filter. Taking FIR as an example, the signal operation process of adaptive feedforward active noise cancellation is as follows: Figure 2 As shown, x(n) is the original noise signal, P(z) is the transfer function of the original noise channel, representing the time delay of the noise from position A to position B, and x(n) is passed through P(z) to output the target signal d(n). W f y(n) is the feedforward filter, and y(n) is W. f (n) is the inverse noise signal output, and S(z) is the secondary channel, i.e., the input W from the acquired x(n). f (n), to W f (n) Output y(n) to the speaker. After the speaker pushes the air, it is transmitted to the error microphone. The transfer function of the entire path is characterized by the time delay of the noise through the path. After y(n) and d(n) are superimposed and cancel each other out, they are absorbed by the error microphone to obtain the residual noise signal e(n). S'(z) is the estimate of S(z). Its essence is a filter. It has an internal coefficient iterative calculation formula. The output weight is multiplied by x(n) through the formula to correct x(n).

[0006] In practice, e(n) is picked up by the error microphone as one input to the LMS algorithm, and x(n) is corrected by S'(z) as another input to the LMS algorithm. The LMS algorithm's iterative formula is then used to output W. f The weight coefficients of (n) are used to obtain W. f After assigning weight coefficients to x(n), we use x(n) as W. f The input signal of (n) is passed through W f The weight coefficients of (n) are multiplied by x(n) to output y(n), thus achieving the purpose of adaptive feedforward active noise reduction.

[0007] Existing adaptive feedforward active noise cancellation solutions have the following problems:

[0008] Because the feedforward filter W f (n) The FIR or IIR filters used are all linear filters. Therefore, if there is a nonlinear element in the propagation path from the noise source at position A to the speaker at position B, such as the original noise being too large causing the reference microphone to become nonlinear, or if there is a nonlinear element in the propagation path from the speaker at position B to the error microphone at position C, such as the speaker being saturated, then the noise reduction effect of the entire link will be significantly reduced because the FIR or IIR filters cannot handle the harmonics and intermodulation distortion caused by the nonlinearity. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides an adaptive feedforward active noise reduction method, which is used to improve harmonics and intermodulation distortion generated by the link.

[0010] Therefore, a noise reduction method using a neural network to construct a secondary channel estimate is provided, including:

[0011] S201. Constructing the theoretical model of the second neural network model;

[0012] S202. Acquire x1(n), h1(n), and e1(n) from the adaptive feedforward active noise reduction architecture, where x1(n) is the historical original noise signal, h1(n) is the output value of x1(n) after correction by S'(z), and e1(n) is the historical residual noise signal.

[0013] (z) is the estimate of the secondary channel in the adaptive feedforward active noise cancellation architecture;

[0014] S203. Using x1(n) and e1(n) as inputs, calculate the weight coefficients of the second neural network model using the BP algorithm;

[0015] S204. Using x1(n), the weight coefficients of the second neural network model, and h1(n) as training samples for the theoretical model of the second neural network model, with x1(n) and the weight coefficients of the second neural network model as inputs and h1(n) as outputs, the theoretical model is trained using machine learning.

[0016] S205. The constructed second neural network model is used as S'(z) in the adaptive feedforward active noise cancellation architecture. The second neural network model takes x1(n) and the current weight coefficients of the second neural network model as inputs and outputs h1(n) to the adaptive algorithm in the adaptive feedforward active noise cancellation architecture. After analysis by the adaptive algorithm, the weight coefficients are generated to adjust the inverse noise signal output by the feedforward filter and then output to the speaker in the adaptive feedforward active noise cancellation architecture for playback.

[0017] Furthermore, the trained second neural network model is fitted using a logistic regression algorithm. The logistic regression algorithm is configured as the LR algorithm.

[0018] Furthermore, the adaptive algorithm is the LMS algorithm.

[0019] An electronic device is also provided, wherein the electronic device includes:

[0020] Controller; and,

[0021] A memory is configured to store computer-executable instructions, which, when executed, cause the controller to perform the methods described above.

[0022] A computer-readable storage medium is also provided, wherein the computer-readable storage medium stores one or more programs that, when executed by a controller, implement the above-described method.

[0023] Beneficial effects:

[0024] This invention improves the estimation of the secondary channel by replacing S'(z) with a second neural network model, thereby reducing harmonic distortion and intermodulation distortion caused by nonlinear elements in the propagation path from the B-position speaker to the C-position error microphone.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0027] Figure 1 A schematic diagram of an adaptive feedforward noise cancellation architecture in traditional headphones is shown.

[0028] Figure 2 A schematic diagram illustrating the implementation of a digital noise reduction scheme based on an adaptive feedforward noise reduction architecture in traditional headphones is shown.

[0029] Figure 3 This diagram illustrates the implementation of the digital noise reduction scheme when the feedforward filter is replaced by the first neural network model in this embodiment.

[0030] Figure 4 A schematic diagram of the structure of the first neural network model is shown;

[0031] Figure 5 A schematic diagram of the neuron's structure is shown;

[0032] Figure 6 This diagram illustrates the structure of the first neural network model when there are several input signals.

[0033] Figure 7 This diagram illustrates the implementation of a digital noise reduction scheme when the secondary channel estimation S'(z) is replaced by a second neural network model in this embodiment.

[0034] Figure 8 This diagram illustrates the implementation of a digital noise reduction scheme where both the feedforward filter and the secondary channel estimation S'(z) are implemented using neural networks in this embodiment.

[0035] Figure 9 A schematic diagram of the electronic device of the present invention is shown;

[0036] Figure 10 A schematic diagram of the structure of the computer-readable storage medium of the present invention is shown. Detailed Implementation

[0037] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0038] When a nonlinear element exists in the propagation path from noise source A to speaker B, such as excessive original noise causing nonlinearity in the reference microphone, a nonlinear feedforward filter can better control the noise. Since a neural network is a nonlinear controller, it can be used to implement a feedforward filter, improving the harmonic and intermodulation distortion phenomena caused by the nonlinear element in the propagation path from noise source A to speaker B. Specifically, see... Figure 3 Based on the adaptive feedforward active noise reduction architecture of the background technology, the feedforward filter is replaced by a first neural network model, specifically including the following steps:

[0039] S101. Construct the theoretical model of the first neural network model, specifically:

[0040] Assume the first neural network model is composed of, for example Figure 4 The 3*N*1 feedforward neural network is implemented, that is, the input layer has 3 neurons, the hidden layer has N neurons, and the output layer has one neuron;

[0041] Among them, w ij,h w represents the weight value of the i-th input to the j-th neuron in the hidden layer. j,o This represents the weight values ​​from the neurons in the j hidden layers to the output;

[0042] The definition of the above neurons is as follows: Figure 5 As shown, based on this, assume there are several input signals x i If i = 1, 2, ..., N, then the first neural network model becomes as follows: Figure 6 As shown in the figure, w i Let represent the weight coefficient corresponding to the i-th input, θ represent the threshold of the neuron, which can also be understood as another constant input value; ∑ represents accumulation, Net represents the accumulated value after multiplying each input value by its weight value, y is the output value; f is a non-linear function, such as the sigmoid function, or the hyperbolic tangent function as follows:

[0043]

[0044] S102. Obtain S'(z) of the adaptive feedforward active noise cancellation architecture in the headphones, where S'(z) is an estimate of the secondary channel in the adaptive feedforward active noise cancellation architecture, specifically:

[0045] In this step, S'(z) is temporarily implemented using an FIR filter or an IIR filter. Taking an FIR filter as an example, the iterative calculation formula for the weight coefficients of S'(z) at each order is as follows:

[0046] w l (n+1)=w l(n)+λ e e(n)x(n)

[0047] S103. Collect x1(n), y1(n), and e1(n) from the adaptive feedforward active noise reduction architecture, where x1(n) is the historical original noise signal, y1(n) is the historical reverse noise signal output by the feedforward filter, and e1(n) is the historical residual noise signal.

[0048] S104. Using the value of x1(n) corrected by S'(z) and e1(n) as input, calculate the weight coefficients of the first neural network model using the BP (backpropagation) algorithm. Specifically:

[0049] According to the BP (Back propagation) algorithm, the iterative calculation formulas for each weight in the output layer of the first neural network model are as follows, where f' o (Net o ) is f o (Net o The derivative of )

[0050] w j，o (n+1)=w j，o (n)+2e(n)f′ o (Net o )x i，h (n)

[0051] The iterative calculation formulas for each weight in the hidden layer are as follows:

[0052] w ij，h (n+1)=w ij，h (n)+2e(n)f′ o (Net o )w j，o f h ′(Net j，h )x i (n)

[0053] The two formulas above can be written as follows:

[0054] w node，o (n+1)=w node，o (n)+λ1δ node (n)x node，h (n)

[0055] In the formula, λ1 is the convergence coefficient, and δ node (n) is calculated as follows:

[0056]

[0057] S105. Using x1(n), the weight coefficients of the first neural network model, and y1(n) as training samples for the theoretical model of the first neural network model, with x1(n) and the weight coefficients of the first neural network model as inputs and y1(n) as outputs, the theoretical model is trained using machine learning, and the first neural network model is fitted using the logistic regression (LR) algorithm until a model that meets the requirements is obtained, thus realizing the construction of the first neural network model;

[0058] S106. Use the constructed first neural network model as the feedforward filter W in the adaptive feedforward active noise reduction architecture. f (n) is used, for example, the first neural network model takes x(n) and the current weight coefficients of the first neural network model as input and outputs y(n) to the loudspeaker in the adaptive feedforward active noise reduction architecture to play sound, where x(n) is the current original noise signal and y(n) is the current inverse noise signal output by the feedforward filter.

[0059] Since there are nonlinear elements in the propagation path from the B-position speaker to the C-position error microphone, such as the speaker being saturated, if a nonlinear estimation is used for the secondary channel, the estimation of the secondary channel will be more accurate. Furthermore, considering that S(z) is nonlinear, S'(z) can be estimated using a second neural network model. The parameters of the second neural network model are iteratively solved using the BP (backpropagation) algorithm. Specifically, see... Figure 7 Based on the adaptive feedforward active noise reduction architecture of the background technology, S'(z) is modified to be implemented by a second neural network model, including the following steps:

[0060] S201. Construct the theoretical model of the second neural network model, wherein the theoretical model of the second neural network model is constructed with reference to the theoretical model of the first neural network model, which will not be elaborated here.

[0061] S202. Collect x1(n), h1(n), and e1(n) from the adaptive feedforward active noise reduction architecture, where x1(n) is the historical original noise signal, h1(n) is the output value of x1(n) after correction by S'(z), and e1(n) is the historical residual noise signal;

[0062] S203. Using x1(n) and e1(n) as inputs, calculate the weight coefficients of the second neural network model using the BP (backpropagation) algorithm. Specifically:

[0063] Based on the BP (Back propagation) algorithm, the iterative calculation formulas for each weight in the second neural network model are as follows:

[0064] w node，o(n+1)=w node，o (n)+λ1δ node (n)x node，h (n)

[0065]

[0066] S204. Using x1(n), the weight coefficients of the second neural network model, and h1(n) as training samples for the theoretical model of the second neural network model, with x1(n) and the weight coefficients of the second neural network model as inputs and h1(n) as outputs, the theoretical model is trained using machine learning, and the second neural network model is fitted using the logistic regression (LR) algorithm until a model that meets the requirements is obtained, thus realizing the construction of the second neural network model;

[0067] S205. The constructed second neural network model is used as S'(z) in the adaptive feedforward active noise reduction architecture, where S'(z) is the estimation of the secondary channel in the adaptive feedforward active noise reduction architecture. For example, with x(n) and e(n) as input, the current weight coefficients of the second neural network model are calculated using the BP algorithm. The second neural network model takes x(n) and the current weight coefficients of the second neural network model as input and outputs h(n) to the LMS algorithm in the adaptive feedforward active noise reduction architecture. After analysis by the LMS algorithm, weight coefficients are generated to adjust the inverse noise signal output by the feedforward filter. Here, x(n) is the current original noise signal, h(n) is the value of x(n) after correction by the second neural network model, and e(n) is the current residual noise signal.

[0068] When nonlinear elements exist simultaneously in the propagation paths from the noise source at position A to the speaker at position B, and from the speaker at position B to the error microphone at position C, both the feedforward filter and the secondary channel estimate S'(z) can be implemented using neural networks. Specifically, for example... Figure 8 As shown, the first neural network model trained above is used as a feedforward filter, and the second neural network model trained above is used as a secondary channel estimate S'(z).

[0069] It should be noted that:

[0070] The method used in this embodiment can be converted into program steps and apparatus that can be stored in a computer storage medium and implemented by being called and executed by a controller.

[0071] The algorithms and displays provided herein are not inherently related to any particular computer, virtual device, or other equipment. Various general-purpose devices can also be used in conjunction with the teachings herein. The required structure for constructing such devices is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0072] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0073] Similarly, it should be understood that, in order to simplify this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof. However, this method of disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into this detailed description, wherein each claim itself is a separate embodiment of the invention.

[0074] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0075] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0076] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the device for detecting the wearing status of an electronic device according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can take the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0077] For example, Figure 9 A schematic diagram of an electronic device according to an embodiment of the present invention is shown. The electronic device conventionally includes a processor 61 and a memory 62 arranged to store computer-executable instructions (program code). The memory 62 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. The memory 62 has storage space 63 for storing program code 64 for performing any method steps in the embodiments. For example, the storage space 63 for program code may include various program codes 64 respectively for implementing the various steps in the above methods. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. Such computer program products are typically, for example... Figure 10 The aforementioned computer-readable storage medium. This computer-readable storage medium may have the same characteristics as... Figure 6 The memory 62 in the electronic device is arranged in a similar manner as storage segments, storage spaces, etc. The program code can be compressed, for example, in a suitable form. Typically, the storage unit stores program code 71 for performing the steps of the method according to the invention, i.e., program code that can be read by a processor such as 61, which, when run by the electronic device, causes the electronic device to perform the various steps of the method described above.

[0078] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

Claims

1. A denoising method using a neural network to construct a secondary channel estimate, characterized in that, include: S201. Constructing the theoretical model of the second neural network model; S202. Collect x1(n), h1(n), and e1(n) from the adaptive feedforward active noise reduction architecture, where x1(n) is the historical original noise signal, h1(n) is the output value of x1(n) after correction by S'(z), e1(n) is the historical residual noise signal, and S'(z) is the estimation of the secondary channel in the adaptive feedforward active noise reduction architecture; S203. Using x1(n) and e1(n) as inputs, calculate the weight coefficients of the second neural network model using the BP algorithm; S204. Using x1(n), the weight coefficients of the second neural network model, and h1(n) as training samples for the theoretical model of the second neural network model, with x1(n) and the weight coefficients of the second neural network model as inputs and h1(n) as outputs, the theoretical model is trained using machine learning. S205. The constructed second neural network model is used as S'(z) in the adaptive feedforward active noise cancellation architecture. The second neural network model takes x1(n) and the current weight coefficients of the second neural network model as inputs and outputs h1(n) to the adaptive algorithm in the adaptive feedforward active noise cancellation architecture. After analysis by the adaptive algorithm, the weight coefficients are generated to adjust the inverse noise signal output by the feedforward filter and then output to the speaker in the adaptive feedforward active noise cancellation architecture for playback.

2. The method according to claim 1, characterized in that, The trained second neural network model was fitted using a logistic regression algorithm.

3. The method according to claim 2, characterized in that, The logistic regression algorithm is configured as the LR algorithm.

4. The method according to claim 1, characterized in that, The adaptive algorithm is the LMS algorithm.

5. A computer-readable storage medium, wherein, The computer-readable storage medium stores one or more programs that, when executed by a controller, implement the method of any one of claims 1-4.

6. An electronic device, wherein, The electronic device includes: Controller; and, A memory configured to store computer-executable instructions, which, when executed, cause the controller to perform the method of any one of claims 1-4.

Images