Signal denoising method and device based on neural network and comb filter

Abstract

The present invention discloses a signal denoising method and device based on a neural network and a comb filter, wherein the method comprises: inputting a first signal to be denoised; preprocessing the first signal to obtain a characteristic graph of the first signal; inputting the characteristic graph of the first signal into a pre-trained neural network model, and outputting the fundamental frequency of the first signal; determining the control parameters of the comb filter according to the fundamental frequency of the first signal; and controlling the comb filter to filter the first signal according to the determined control parameters to obtain a second signal after denoising. The present invention can adaptively adjust the filter parameters according to the signal to be denoised, and then filter the signal to be denoised, without requiring the filter function to be differentiable.

Description

Technical Field

[0001] The present invention relates to the field of signal processing, and in particular to a signal noise reduction method and device based on a neural network and a comb filter. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] A filter is a frequency selection device. After setting the frequency, it can filter the input signal and allow the signal of the specified frequency to pass through, thus realizing the frequency screening function of the input signal. At present, filters are widely used in many aspects. For example, in image processing, median filtering is used to remove salt and pepper noise from images; bilateral filtering can smooth and reduce noise on images while retaining the edges of images; wavelet transform is used to decompose and reconstruct images. In speech processing, Wiener filtering can reduce noise on speech; Mel filters and gamma filters are also often used to extract more advanced acoustic features. However, the design of traditional filters and the setting of parameters generally need to be determined manually, usually relying on a lot of professional knowledge and experience.

[0004] To solve the above problems, in the prior art, a neural network is used to control the filter, that is, the filter is integrated into the neural network, the filter is used to process linear work, the neural network is used to process nonlinear work, and the GPU is used for calculation, which can greatly improve the calculation speed and efficiency. Experiments have shown that adding a filter to a neural network can reduce the training parameters of the network and increase the convergence speed of the network. It can be seen that the research on the combination of neural networks and filters is very meaningful.

[0005] At present, the existing methods of combining neural networks with filters can only use differentiable filter functions. If a non-differentiable function is to be used as the shape constraint of the filter, it is necessary to fit it with a differentiable function. In addition, filter parameters whose orders change by integers cannot be learned through back propagation.

[0006] To address the above problems, no effective solution has been proposed yet. Summary of the invention

[0007] In an embodiment of the present invention, a signal denoising method based on a neural network and a comb filter is provided to realize the control of a non-differentiable filter function and a discontinuous variable using a neural network. The method comprises: inputting a first signal to be denoised; preprocessing the first signal to obtain a characteristic graph of the first signal; inputting the characteristic graph of the first signal into a pre-trained neural network model, and outputting the fundamental frequency of the first signal; determining control parameters of the comb filter according to the fundamental frequency of the first signal; and controlling the comb filter to filter the first signal according to the determined control parameters to obtain a second signal after denoising.

[0008] The embodiment of the present invention also provides a signal denoising device based on a neural network and a comb filter, which is used to realize the control of non-differentiable filter functions and discontinuous variables using a neural network. The device includes: a signal input module, which is used to input a first signal to be denoised; a signal preprocessing module, which is used to preprocess the first signal to obtain a feature map of the first signal; a neural network model prediction module, which is used to input the feature map of the first signal into a pre-trained neural network model and output the fundamental frequency of the first signal; a comb filter parameter determination module, which is used to determine the control parameters of the comb filter according to the fundamental frequency of the first signal; and a signal filtering module, which is used to control the comb filter to filter the first signal according to the determined control parameters to obtain a second signal after denoising.

[0009] A computer device is also provided in an embodiment of the present invention for realizing the control of non-differentiable filter functions and discontinuous variables using a neural network. The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the signal denoising method based on the neural network and comb filter is realized.

[0010] A computer-readable storage medium is also provided in an embodiment of the present invention for realizing the control of non-differentiable filter functions and discontinuous variables using a neural network. The computer-readable storage medium stores a computer program for executing the above-mentioned signal denoising method based on a neural network and a comb filter.

[0011] In the embodiment of the present invention, the feature map of the signal to be denoised is input into a pre-trained neural network, and the neural network model is used to predict the control parameters of the comb filter, thereby controlling the comb filter to filter the signal to be denoised. Through the embodiment of the present invention, the filter parameters are adaptively adjusted according to the signal to be denoised, thereby filtering the signal to be denoised, without requiring the filter function to be differentiable. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0013] Figure 1 A schematic diagram of a spectrum of a comb filter provided in an embodiment of the present invention;

[0014] Figure 2 A flow chart of a signal noise reduction method based on a neural network and a comb filter provided in an embodiment of the present invention;

[0015] Figure 3 A specific implementation flow chart of a signal noise reduction method based on a neural network and a comb filter provided in an embodiment of the present invention;

[0016] Figure 4 A time domain waveform diagram of a speech signal to be denoised provided in an embodiment of the present invention;

[0017] Figure 5 A time domain waveform diagram of a normalized speech signal to be denoised provided in an embodiment of the present invention;

[0018] Figure 6 A schematic diagram of a signal after framing and windowing provided in an embodiment of the present invention;

[0019] Figure 7 A characteristic graph of a speech signal to be denoised provided in an embodiment of the present invention;

[0020] Figure 8 A schematic diagram of how the filter spectrum characteristics and shape vary with input signals provided in an embodiment of the present invention;

[0021] Fig. 9 A schematic diagram of a signal noise reduction device based on a neural network and a comb filter provided in an embodiment of the present invention;

[0022] Fig.10 A schematic diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0024] The comb filter is a type of infinite impulse response (IIR) filter, because its spectral characteristics are similar to the shape of a comb, such as Figure 1 As shown, the comb filter has a good filtering effect on the signal whose spectrum has a frequency doubling characteristic.

[0025] However, since the comb filter function is a non-differentiable function and the order of the filter is a discontinuous variable, the existing method of combining a neural network with a filter is not applicable to the comb filter. To this end, an embodiment of the present invention provides a signal denoising method based on a neural network and a comb filter, which can be applied to but not limited to the denoising processing of speech signals. The method uses a neural network model to predict the order of the comb filter according to the feature map of the signal to be denoised, controls the comb filter to filter the signal to be denoised, and can select different filter parameters for different signals.

[0026] Figure 2 FIG. 1 is a flow chart of a signal noise reduction method based on a neural network and a comb filter provided in an embodiment of the present invention. Figure 2 As shown, the method comprises the following steps:

[0027] S201, input a first signal to be denoised.

[0028] It should be noted that, in the embodiment of the present invention, the first signal input through S201 may be any signal to be denoised, and may be an original noisy signal, or a signal containing a small amount of noise after being filtered once or multiple times.

[0029] S202, preprocessing the first signal to obtain a feature map of the first signal.

[0030] In specific implementation, the above S202 can be achieved through the following steps: performing amplitude normalization processing on the first signal; framing the normalized signal, and performing windowing operation on the framed signal; removing frame signals whose signal strength is less than a preset threshold; performing Fourier transform on the remaining frame signals to obtain the corresponding frequency spectrum as the feature map of the first signal.

[0031] In one embodiment, the frame signal may be Fourier transformed using the following formula:

[0032]

[0033] Wherein, Y(ω) represents the characteristic graph of the signal; STFT represents Fourier transform; represents the nth frame signal; e represents the natural base; ω represents the frequency; j represents the imaginary unit; t represents the time.

[0034] S203, inputting the feature map of the first signal into a pre-trained neural network model, and outputting the fundamental frequency of the first signal.

[0035] It should be noted that the neural network model in the embodiment of the present invention can be a model for predicting the fundamental frequency of a signal obtained by machine learning of any neural network. In specific implementation, a neural network based on LSTM can be constructed to predict the fundamental frequency of the signal. Therefore, in one embodiment, the signal denoising method based on a neural network and a comb filter provided in the embodiment of the present invention can also include the following steps: training a neural network based on LSTM to obtain a neural network model for predicting the fundamental frequency of the signal.

[0036] S204: Determine control parameters of the comb filter according to the fundamental frequency of the first signal.

[0037] In one embodiment, in the above S204, the fundamental frequency of the first signal may be determined as the order of the comb filter, and the system function of the comb filter may be determined by the following formula:

[0038] H(Z)=b0+b M Z -M (2)

[0039] Wherein, H(Z) represents the system function of the comb filter; M represents the order of the comb filter; Z represents the independent variable of the signal after Z transformation; b0 represents the coefficient of the 0th order filter; b M Represents the coefficients of the Mth order filter.

[0040] S205, controlling the comb filter to filter the first signal according to the determined control parameters to obtain a second signal after noise reduction.

[0041] In one embodiment, in the above S205, the system function of the comb filter can be inversely Fourier transformed by the following formula (3) to obtain the time domain quantity corresponding to the system function of the comb filter, and then the time domain quantity corresponding to the system function of the comb filter is convolved with the first signal by the following formula (4) to obtain the second signal after noise reduction:

[0042]

[0043] OUT(t)=y(t)*h(t) (4)

[0044] Among them, h(t) represents the time domain quantity corresponding to the comb filter system function; ISTFT represents the inverse Fourier transform; H(Z) represents the system function of the comb filter; Z represents the independent variable of the signal after Z transformation; e represents the natural base; ω represents the frequency; j represents the imaginary unit; t represents the time; OUT(t) represents the signal output by the comb filter; y(t) represents the signal input to the comb filter; * represents the convolution operation.

[0045] Figure 3 FIG. 1 is a specific implementation flow chart of a signal noise reduction method based on a neural network and a comb filter provided in an embodiment of the present invention. Figure 3 As shown, the input signal is preprocessed (for example, amplitude normalization, framing, windowing, etc.), the preprocessed signal is mapped to the spectrum, and the feature map of the input signal is obtained, which is input into a pre-trained neural network model. The fundamental frequency of the input signal is predicted by the neural network model, and the order of the comb filter is determined by the fundamental frequency, and then the system function of the comb filter is determined. Finally, the comb filter of the system function is used to filter the input signal to achieve adaptive filtering.

[0046] The following describes the embodiment of the present invention in detail by taking a speech signal as an example.

[0047] Assuming x(t) is a clean speech signal and n(t) is noise (for example, Gaussian white noise), the input signal y(t) can be expressed as follows:

[0048] y(t)=x(t)+n(t) (5)

[0049] Figure 4 A time domain waveform diagram of a speech signal to be denoised provided in an embodiment of the present invention, such as Figure 4 As shown, the input signal is a speech signal with additive noise.

[0050] 1. Signal preprocessing stage:

[0051] ① The input signal amplitude is normalized by the following formulas (6) and (7):

[0052]

[0053]

[0054] Among them, mean represents the mean function; max represents the maximum value function; abs represents the absolute value function.

[0055] right Figure 4 The result after normalization of the speech signal shown in Figure 5 shown.

[0056] ② Framing and windowing: Frame processing is performed, wherein the frame length is 25 ms and the frame shift is 10 ms. To reduce the leakage problem caused by frame division, a windowing operation is performed after frame division, and the window function selected in the embodiment of the present invention is a Hamming window.

[0057] ③ Remove the leading silent frame of the audio: If the signal strength of a frame after the signal is divided into frames is too small, then remove the frame; for example, Figure 5 After the signal shown in the figure is framed and windowed, the following is obtained: Figure 6 The signal shown.

[0058] 2. Neural network prediction stage:

[0059] A LSTM-based neural network is used to predict the fundamental frequency of the signal to be denoised. The input of the neural network is the feature map of the signal to be denoised, and the output of the neural network is a real number M, which is the fundamental frequency of the signal. Figure 7 A characteristic graph of a speech signal to be denoised provided in an embodiment of the present invention.

[0060] 3. Comb filter filtering stage:

[0061] The output neural network of the neural network is used as the order of the comb filter, and then the system function of the comb filter is obtained through the above formula (2). The system function of the comb filter is inverse Fourier transformed through the above formula (3) to obtain its time domain expression.

[0062] 4. Signal output stage:

[0063] The speech signal to be denoised is convolved with the time domain expression of the comb filter using the above formula (4) to obtain the final denoised speech signal.

[0064] In the embodiment of the present invention, the Straight-Through Estimator method is used to solve the problem of gradient back propagation of non-differentiable functions in neural networks. It relies on non-differentiable functions to calculate loss during forward propagation, and realizes gradient propagation through the gradient designed by us during gradient back propagation.

[0065] The objective function of the comb filter parameter estimation module is as follows:

[0066]

[0067] Among them, x represents the input data, p(x) represents the comb filter order calculation function, which is a non-differentiable function. sg[] represents the stop calculation gradient identifier, comb() represents the comb filter system function calculation function, and y represents the label data. When the data is forward propagated, the function e is equivalent to:

[0068]

[0069] When the gradient is back propagated, the function e is equivalent to:

[0070]

[0071] This function is a differentiable function and can perform gradient backpropagation.

[0072] In the embodiment of the present invention, after the input signal to be denoised is normalized, framed, and silent frames are removed, it is input into the neural network. Experimental results show that by filtering the speech signal with 3db Gaussian white noise through the signal denoising method based on neural network and comb filter provided in the embodiment of the present invention, the signal-to-noise ratio of the speech signal is improved to about 8db, which is a significant improvement compared with the original signal. More importantly, the order of the comb filter can be modified in real time according to the output of the neural network, that is, its spectral characteristics and shape can be adaptively adjusted according to the input signal. Figure 8 It shows that the filter spectrum characteristics and shape vary with the input signal, such as Figure 8 As shown, the effectiveness of the embodiment of the present invention is proved.

[0073] Different from the existing algorithms combining neural networks with filters, the signal denoising method based on neural networks and comb filters provided in the embodiments of the present invention does not directly use the filter function as the weight of the first layer of the neural network to participate in the parameter update of the neural network, but uses filtering as the last step of the system, uses the output of the trained neural network to determine the control parameters of the filter, and uses the result of the signal after the comb filter is filtered as a reference for updating the neural network parameters, so that the non-differentiable filter function is used as a link in the back propagation algorithm to participate in the gradient calculation and parameter update. At the same time, the method also realizes the use of neural networks to guide integer variables such as filter orders. The advantage of this is that it breaks the limitation of neural networks on the differentiability of functions and explores the possibility of using back propagation to update non-continuous variables.

[0074] Based on the same inventive concept, the present invention also provides a signal noise reduction device based on a neural network and a comb filter in an embodiment, as described in the following embodiment. Since the principle of solving the problem by the device is similar to that of the signal noise reduction method based on a neural network and a comb filter, the implementation of the device can refer to the implementation of the signal noise reduction method based on a neural network and a comb filter, and the repeated parts will not be repeated.

[0075] Fig. 9 FIG. 1 is a schematic diagram of a signal noise reduction device based on a neural network and a comb filter provided in an embodiment of the present invention. Fig. 9 As shown, the device includes: a signal input module 91, a signal preprocessing module 92, a neural network model prediction module 93, a comb filter parameter determination module 94 and a signal filtering module 95.

[0076] Among them, the signal input module 91 is used to input the first signal to be denoised; the signal preprocessing module 92 is used to preprocess the first signal to obtain a characteristic map of the first signal; the neural network model prediction module 93 is used to input the characteristic map of the first signal into a pre-trained neural network model, and output the fundamental frequency of the first signal; the comb filter parameter determination module 94 is used to determine the control parameters of the comb filter according to the fundamental frequency of the first signal; the signal filtering module 95 is used to control the comb filter to filter the first signal according to the determined control parameters to obtain the second signal after denoising.

[0077] In one embodiment, the signal denoising device based on a neural network and a comb filter provided in the embodiment of the present invention may further include: a model training module 96, which is used to train a neural network based on LSTM to obtain a neural network model for predicting the fundamental frequency of the signal.

[0078] In one embodiment, the signal preprocessing module 92 specifically includes: a normalization unit 921, used to perform amplitude normalization processing on the first signal; a frame dividing and windowing unit 922, used to frame the normalized signal and perform windowing operation on the framed signal; a silent frame removal unit 923, used to remove frame signals whose signal strength is less than a preset threshold; a Fourier transform unit 924, used to perform Fourier transform on the remaining frame signals to obtain a corresponding frequency spectrum as a feature map of the first signal.

[0079] In one embodiment, the Fourier transform unit 924 is further configured to perform Fourier transform on the frame signal according to the above formula (1).

[0080] In one embodiment, the comb filter parameter determination module 94 is further used to determine the fundamental frequency of the first signal as the order of the comb filter, and determine the system function of the comb filter by the following formula.

[0081] In one embodiment, the signal filtering module 95 specifically includes: an inverse Fourier transform unit 951, which is used to perform an inverse Fourier transform on the system function of the comb filter through the above formula (3) to obtain a time domain quantity corresponding to the system function of the comb filter; a convolution operation unit 952, which is used to convolve the time domain quantity corresponding to the system function of the comb filter with the first signal through the above formula (4) to obtain a second signal after noise reduction.

[0082] Based on the same inventive concept, an embodiment of the present invention further provides a computer device for realizing the control of non-differentiable filter functions and discontinuous variables using a neural network. Fig.10 A schematic diagram of a computer device provided in an embodiment of the present invention is shown in FIG. Fig.10 As shown, the computer device 10 includes a memory 101, a processor 102, and a computer program stored in the memory 101 and executable on the processor 102. When the processor 102 executes the computer program, the above-mentioned signal denoising method based on a neural network and a comb filter is implemented.

[0083] Based on the same inventive concept, a computer-readable storage medium is also provided in an embodiment of the present invention for realizing the control of non-differentiable filter functions and discontinuous variables using a neural network. The computer-readable storage medium stores a computer program for executing the above-mentioned signal denoising method based on a neural network and a comb filter.

[0084] In summary, an embodiment of the present invention provides a signal denoising method, apparatus, computer device and computer-readable storage medium based on a neural network and a comb filter, wherein a feature map of a signal to be denoised is input into a pre-trained neural network, and the neural network model is used to predict control parameters of the comb filter, thereby controlling the comb filter to filter the signal to be denoised, and the filter parameters can be adaptively adjusted according to the signal to be denoised, thereby filtering the signal to be denoised, without requiring the filter function to be differentiable.

[0085] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A signal denoising method based on a neural network and a comb filter, characterized in that: include: Input a first signal to be denoised; Preprocessing the first signal to obtain a characteristic graph of the first signal; Inputting the characteristic graph of the first signal into a pre-trained neural network model, and outputting the fundamental frequency of the first signal; Determining control parameters of the comb filter according to the fundamental frequency of the first signal; According to the control parameter, controlling the comb filter to filter the first signal to obtain a second signal after noise reduction; Determining the control parameters of the comb filter according to the fundamental frequency of the first signal includes: determining the fundamental frequency of the first signal as the order of the comb filter, and determining the system function of the comb filter by the following formula: H(Z)=b0+b M From -M ; Wherein, H(Z) represents the system function of the comb filter; M represents the order of the comb filter; Z represents the independent variable of the signal after Z transformation; b0 represents the coefficient of the 0th order filter; b M Represents the coefficients of the Mth order filter; According to the control parameter, controlling the comb filter to filter the first signal to obtain a second signal after noise reduction, including: performing an inverse Fourier transform on the system function of the comb filter by the following formula to obtain a time domain quantity corresponding to the system function of the comb filter: The time domain quantity corresponding to the comb filter system function is convolved with the first signal through the following formula to obtain the second signal after noise reduction: OUT(t)=y(t)*h(t); Among them, h(t) represents the time domain quantity corresponding to the comb filter system function; ISTFT represents the inverse Fourier transform; H(Z) represents the system function of the comb filter; Z represents the independent variable of the signal after Z transformation; e represents the natural base; ω represents the frequency; j represents the imaginary unit; t represents the time; OUT(t) represents the signal output by the comb filter; y(t) represents the signal input to the comb filter; * represents the convolution operation.

2. The method according to claim 1, characterized in that The method further comprises: The LSTM-based neural network is trained to obtain a neural network model for predicting the fundamental frequency of the signal.

3. The method according to claim 1, characterized in that Preprocessing the first signal to obtain a characteristic graph of the first signal includes: Performing amplitude normalization processing on the first signal; Frame the normalized signal, and perform windowing on the framed signal; Remove frame signals whose signal strength is less than a preset threshold; Perform Fourier transform on the remaining frame signals to obtain the corresponding frequency spectrum as the feature map of the first signal.

4. The method according to claim 3, characterized in that The frame signal is Fourier transformed using the following formula: Wherein, Y(ω) represents the characteristic graph of the signal; STFT represents Fourier transform; represents the nth frame signal; e represents the natural base; ω represents the frequency; j represents the imaginary unit; t represents the time.

5. A signal noise reduction device based on a neural network and a comb filter, characterized in that: include: A signal input module, used for inputting a first signal to be denoised; A signal preprocessing module, used for preprocessing the first signal to obtain a characteristic graph of the first signal; A neural network model prediction module, used for inputting the feature map of the first signal into a pre-trained neural network model, and outputting the fundamental frequency of the first signal; A comb filter parameter determination module, used to determine the control parameters of the comb filter according to the fundamental frequency of the first signal; A signal filtering module, used for controlling the comb filter to filter the first signal according to the control parameter to obtain a second signal after noise reduction; The comb filter parameter determination module is specifically used to determine the fundamental frequency of the first signal as the order of the comb filter, and determine the system function of the comb filter through the following formula: H(Z)=b0+b M From -M ; Wherein, H(Z) represents the system function of the comb filter; M represents the order of the comb filter; Z represents the independent variable of the signal after Z transformation; b0 represents the coefficient of the 0th order filter; b M Represents the coefficients of the Mth order filter The signal filtering module is specifically used to perform an inverse Fourier transform on the system function of the comb filter through the following formula to obtain the time domain quantity corresponding to the system function of the comb filter: The time domain quantity corresponding to the comb filter system function is convolved with the first signal through the following formula to obtain the second signal after noise reduction: OUT(t)=y(t)*h(t); Among them, h(t) represents the time domain quantity corresponding to the comb filter system function; ISTFT represents the inverse Fourier transform; H(Z) represents the system function of the comb filter; Z represents the independent variable of the signal after Z transformation; e represents the natural base; ω represents the frequency; j represents the imaginary unit; t represents the time; OUT(t) represents the signal output by the comb filter; y(t) represents the signal input to the comb filter; * represents the convolution operation.

6. The device according to claim 5, characterized in that The device also includes: The model training module is used to train the LSTM-based neural network to obtain a neural network model for predicting the fundamental frequency of the signal.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the signal noise reduction method based on a neural network and a comb filter according to any one of claims 1 to 4 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the signal denoising method based on a neural network and a comb filter according to any one of claims 1 to 4 is implemented.

Images