A method, apparatus, device, and storage medium for training a noise reduction neural network.

Abstract

This application belongs to the field of signal denoising technology and discloses a training method, apparatus, device, and storage medium for a denoising neural network. The method includes the following steps: acquiring a noisy signal and inputting it into an initial neural network to obtain a sampling sequence; performing Gaussian sampling on the sampling sequence to obtain a noise signal; obtaining the original signal based on the noisy signal and the noise signal; calculating the error of the initial neural network based on the original signal; determining whether the error is less than a preset error threshold; if not, returning to the training step; if yes, ending the training and using the initial neural network as the denoising neural network. This application can obtain a denoising neural network without acquiring a clean signal, realizing unsupervised learning and having extremely high applicability in engineering applications.

Description

Technical Field

[0001] This application relates to the field of signal denoising technology, and in particular to a training method, apparatus, device and storage medium for a denoising neural network. Background Technology

[0002] Currently, algorithms for denoising audio signals include frequency domain denoising algorithms, time domain denoising algorithms, wavelet domain denoising algorithms, and so on. These algorithms all require the use of denoising autoencoders. The working principle of an autoencoder is to input a set of clean signals and noisy signals, and then compare the encoded and decoded signals of the noisy signals with the original clean signals to obtain the noise function.

[0003] However, in practical applications, it is difficult to obtain truly pure audio signals without noise. They are usually mixed with some noise, which leads to low accuracy of the constructed noise function and poor noise reduction effect of the algorithm. Summary of the Invention

[0004] This application provides a training method, apparatus, device, and storage medium for a denoising neural network, which can obtain a denoising neural network without acquiring clean signals, realize unsupervised learning, and has extremely high applicability in engineering applications.

[0005] In a first aspect, embodiments of this application provide a method for training a denoising neural network, comprising:

[0006] Training steps: Acquire a noisy signal and input it into the initial neural network to obtain a sampling sequence;

[0007] Gaussian sampling is performed on the sampling sequence to obtain the noise signal;

[0008] The original signal is obtained from the noise-added signal and the noise signal;

[0009] Calculate the error of the initial neural network based on the original signal;

[0010] Determine if the error is less than the preset error threshold; if not, return to the training step; if yes, end the training and use the initial neural network as the denoising neural network.

[0011] Furthermore, the method also includes:

[0012] After obtaining the original signal, increment the training count by 1;

[0013] Determine if the number of training iterations has reached the preset training threshold; if not, return to the training step; if yes, end the training and use the initial neural network as the denoising neural network.

[0014] Furthermore, the process of obtaining the original signal from the noise signal and the added noise signal includes:

[0015] Subtract the noise signal from the noise signal to obtain the original signal.

[0016] Furthermore, the error in calculating the initial neural network based on the original signal includes:

[0017] Calculate the adjacent difference sequence of the original signal;

[0018] Calculate the variance of adjacent difference sequences; treat the variance as the error.

[0019] Furthermore, the sampling sequence includes the mean and variance of the noise signal in the noisy signal; the noisy signal is a noise-added frequency signal.

[0020] Secondly, embodiments of this application provide a training apparatus for a denoising neural network, comprising:

[0021] The training module is used to acquire the noisy signal and input it into the initial neural network to obtain the sampling sequence;

[0022] The sampling module is used to perform Gaussian sampling based on the sampling sequence to obtain the noise signal;

[0023] The noise reduction module is used to obtain the original signal from the noise-added signal and the noise signal;

[0024] The error calculation module is used to calculate the error of the initial neural network based on the original signal;

[0025] The judgment module is used to return to the training module when the error is greater than or equal to a preset error threshold; when the error is less than the preset error threshold, the initial neural network is used as the denoising neural network.

[0026] Furthermore, the noise reduction module is specifically used to subtract the noise signal from the noise signal to obtain the original signal.

[0027] Furthermore, the error calculation module is specifically used to calculate the variance of adjacent difference sequences and adjacent difference sequences of the original signal; the variance is used as the error.

[0028] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the steps of a training method for a noise reduction neural network as described in any of the above embodiments.

[0029] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of a training method for a noise reduction neural network as described in any of the above embodiments.

[0030] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following:

[0031] This application provides a training method for a denoising neural network. During training, only a noisy signal needs to be acquired. Then, a Gaussian signal is sampled using the output of the neural network to obtain the noise signal. The original signal is then obtained based on the noise signal and the noisy signal. Since the initial output of the neural network may not be accurate during training, the error needs to be calculated based on the original signal. If the error is less than a preset error threshold, it indicates that the output accuracy of the initial neural network has met the denoising requirements, and the initial neural network can be used to train the denoising neural network. The above method does not require the acquisition of clean signals during the training of the denoising neural network, realizing unsupervised learning and exhibiting high applicability in engineering applications. Attached Figure Description

[0032] Figure 1 A flowchart illustrating a training method for a noise reduction neural network, provided as an exemplary embodiment of this application.

[0033] Figure 2 This is a structural diagram of a training device for a noise reduction neural network, provided as an exemplary embodiment of this application. Detailed Implementation

[0034] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0035] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] Existing noise reduction algorithms include frequency domain noise reduction algorithms, time domain noise reduction algorithms, wavelet domain noise reduction algorithms, etc.

[0037] Frequency domain noise reduction algorithms, based on the frequency characteristics of a signal, analyze the differences between noise and the signal spectrum to separate the noise component from the signal, thus achieving noise elimination. In the field of audio processing, commonly used frequency domain noise reduction algorithms include FIR filters and Kalman filters based on Fast Fourier Transform (FFT).

[0038] Temporal denoising algorithms primarily separate and eliminate noise by analyzing the temporal waveform characteristics of noisy and non-noisy signals. These algorithms typically include adaptive denoising algorithms and deep learning-based denoising algorithms.

[0039] Wavelet transform is a signal decomposition and reconstruction technique that breaks down a signal into multiple frequency components and processes each component accordingly. In the field of noise reduction, wavelet domain noise reduction algorithms can perform analysis and processing based on either the frequency domain or the time domain.

[0040] These noise reduction algorithms all require the use of a noise reduction autoencoder. The working principle of an autoencoder is to input a set of clean signals and a noisy signal, and then compare the signal after encoding and decoding the noisy signal with the original clean signal to obtain the noise function.

[0041] However, in practical applications, it is difficult to obtain truly pure audio signals without noise. They are usually mixed with some noise, resulting in low accuracy of the constructed noise function and poor noise reduction effect.

[0042] Please see Figure 1 To address the difficulty in obtaining noise-free, clean signals in existing engineering projects, this application provides a training method for a denoising neural network, specifically including:

[0043] Step S1, Training Step: Acquire the noisy signal and input it into the initial neural network to obtain the sampling sequence.

[0044] Among them, the noise-added signal can be a noise-added frequency signal.

[0045] The initial neural network can be a convolutional neural network or a deep residual shrinking network.

[0046] The sampling sequence is a two-dimensional sequence, representing the mean and variance of the noise signal in the noisy signal, respectively.

[0047] Step S2: Perform Gaussian sampling based on the sampling sequence to obtain the noise signal.

[0048] The Gaussian signal used can be an existing one-dimensional Gaussian signal.

[0049] Step S3: Obtain the original signal based on the noise signal and the added noise signal.

[0050] Specifically, the noise signal is subtracted from the noise signal to obtain the original signal.

[0051] Step S4: Calculate the error of the initial neural network based on the original signal.

[0052] Specifically, the adjacent difference sequences of the original signal are calculated; the variance of the adjacent difference sequences is calculated; and the variance is used as the error.

[0053] Assuming the sampling sequence is [X1, X2, X3, X4], then the adjacent difference sequences of the original signal are [Y1 = X2 - X1, Y2 = X3 - X2, Y3 = X4 - X3]. The variance of these adjacent difference sequences is used as the error.

[0054] Step S5: Determine if the error is less than a preset error threshold; if not, return to the training step; if yes, end the training and use the initial neural network as the denoising neural network. Specifically, in addition to judging based on the error, ending the training can also be achieved by limiting the number of training iterations: after obtaining the original signal, increment the training iterations by 1; determine if the number of training iterations has reached the preset training threshold; if not, return to the training step; if yes, end the training and use the initial neural network as the denoising neural network.

[0055] In the specific implementation process, after obtaining the trained noise reduction neural network, the noise reduction signal to be noise-reduced can be input into the noise reduction neural network to obtain the sampling sequence. Then, Gaussian sampling is performed according to the sampling sequence, and the sampling result is used as the noise signal. The noise signal is then subtracted from the noise reduction signal to obtain the noise-reduced, clean original audio signal.

[0056] The above embodiment provides a training method for a denoising neural network. During training, only a noisy signal needs to be acquired. Then, a Gaussian signal is sampled using the output of the neural network to obtain the noise signal. The original signal is then obtained based on the noise signal and the noisy signal. Since the initial output of the neural network may not be accurate during training, the error needs to be calculated based on the original signal. If the error is less than a preset error threshold, it indicates that the output accuracy of the initial neural network has met the denoising requirements, and the initial neural network can be used to train the denoising neural network. This method does not require the acquisition of clean signals during the training of the denoising neural network, achieving unsupervised learning and possessing high applicability in engineering applications.

[0057] Please see Figure 2 Another embodiment of this application provides a training apparatus for a noise reduction neural network, comprising:

[0058] Training module 101 is used to acquire the noisy signal and input it into the initial neural network to obtain the sampling sequence.

[0059] The sampling module 102 is used to perform Gaussian sampling based on the sampling sequence to obtain a noise signal.

[0060] The noise reduction module 103 is used to obtain the original signal based on the noise signal and the noise signal.

[0061] Error calculation module 104 is used to calculate the error of the initial neural network based on the original signal.

[0062] The judgment module 105 is used to return to the training module when the error is greater than or equal to the preset error threshold; and to use the initial neural network as the denoising neural network when the error is less than the preset error threshold.

[0063] Furthermore, the noise reduction module 103 is specifically used to subtract the noise signal from the noise signal to obtain the original signal.

[0064] Furthermore, the error calculation module 104 is specifically used to calculate the variance of the adjacent difference sequences and the adjacent difference sequences of the original signal; and to use the variance as the error.

[0065] The specific limitations of the training device for a denoising neural network provided in this embodiment can be found in the embodiment of the training method for a denoising neural network described above, and will not be repeated here. Each module in the above-described training device for a denoising neural network can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0066] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of a noise reduction neural network training method as described in any of the above embodiments.

[0067] The working process, working details, and technical effects of the computer device provided in this embodiment can be found in the embodiment of a training method for a noise reduction neural network described above, and will not be repeated here.

[0068] This application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of a training method for a denoising neural network as described in any of the above embodiments. The computer-readable storage medium refers to a data storage carrier, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The working process, details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiments of a training method for a denoising neural network described above, and will not be repeated here.

[0069] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A training method for a denoising neural network, characterized in that, include: Training steps: Acquire a noisy signal and input it into an initial neural network to obtain a sampling sequence; wherein, the sampling sequence includes the mean and variance of the noise signal in the noisy signal, and the noisy signal is a noisy frequency signal; Gaussian sampling is performed on the sampling sequence to obtain a noise signal; The original signal is obtained by subtracting the noise signal from the noise signal; specifically, the original signal is obtained by subtracting the noise signal from the noise signal. The error of the initial neural network is calculated based on the original signal; specifically, the adjacent difference sequences of the original signal are calculated; the variance of the adjacent difference sequences is calculated; and the variance is used as the error. Determine whether the error is less than a preset error threshold; if not, return to the training step; if yes, end the training and use the initial neural network as the noise reduction neural network.

2. The training method for the denoising neural network according to claim 1, characterized in that, Also includes: After obtaining the original signal, increment the training count by 1; Determine whether the number of training iterations has reached a preset training threshold; If not, return to the training steps described above; If so, then end the training and use the initial neural network as the denoising neural network.

3. A training device for a noise reduction neural network, characterized in that, include: A training module is used to acquire a noisy signal and input it into an initial neural network to obtain a sampling sequence; wherein, the sampling sequence includes the mean and variance of the noise signal in the noisy signal, and the noisy signal is a noise frequency signal; The sampling module is used to perform Gaussian sampling based on the sampling sequence to obtain a noise signal; A noise reduction module is used to obtain an original signal based on the noise-added signal and the noise signal; specifically, the noise reduction module is used to subtract the noise signal from the noise-added signal to obtain the original signal. An error calculation module is used to calculate the error of the initial neural network based on the original signal; specifically, it calculates the adjacent difference sequence of the original signal; calculates the variance of the adjacent difference sequence; and uses the variance as the error. The judgment module is used to return to the training module when the error is greater than or equal to a preset error threshold; and to use the initial neural network as a noise reduction neural network when the error is less than the preset error threshold.

4. The training device for the noise reduction neural network according to claim 3, characterized in that, The error calculation module is specifically used to calculate the variance of the adjacent difference sequences and the adjacent difference sequences of the original signal; and to use the variance as the error.

5. 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, it implements the steps of the training method for the noise reduction neural network as described in any one of claims 1 to 2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the training method for the denoising neural network as described in any one of claims 1 to 2.

Images