Audio noise reduction method, device and medium for industrial audio processing

Abstract

The present invention proposes an audio noise reduction method, device, and medium for industrial audio processing. The method performs super-subtraction on a noisy audio signal, extracts a pre-processed signal, and performs Wiener filtering on the pre-processed signal using a parametric Wiener filter to obtain an audio signal from which ambient noise has been removed. The audio signal is then subjected to super-subtraction again to obtain a denoised audio signal. The advantages of the present invention are that super-subtraction solves the spectral line discontinuity problem caused by Berouti spectral subtraction, Wiener filtering suppresses ambient noise in the audio signal, and super-subtraction is used again to process the audio signal after Wiener filtering, solving the problem of adhesion caused by unfiltered ambient noise in the audio signal. The present invention utilizes a combination of super-subtraction and Wiener filtering to process complex audio signals in the industrial field, thereby improving the effectiveness of eliminating ambient noise.

Description

Technical Field

[0001] The present invention relates to the field of industrial audio processing, and in particular to an audio noise reduction method, device and medium applied to industrial audio processing. Background Art

[0002] Wind turbines generate various noises during operation, two of the most common being mechanical noise and aerodynamic noise. Aerodynamic noise is the sound of wind sweeping through the turbine blades. If a blade is damaged, the aerodynamic audio signal may exhibit some anomalies. For example, a debonded wind turbine blade may produce a sound when rotating, while a blade with cracks or pinholes may produce a whistling sound when rotating. Experienced wind farm workers can determine blade damage and the specific type of damage by listening to aerodynamic signals. Modern industry utilizes this characteristic to collect the sound of wind sweeping through the turbine blades, effectively monitoring the health of wind turbine blades and ensuring safe operation. However, the collected aerodynamic signal data from wind turbine blades often contains a significant amount of noise mixed with useful signal sources, including noise from the heat dissipation equipment used to cool the wind turbine, noise generated by the generator, and wind noise.

[0003] Existing Methods: Most existing audio noise reduction solutions are primarily used for speech enhancement research. For example, the most classic methods include spectral subtraction and Wiener filtering. Meanwhile, many methods have been proposed to improve upon these methods. However, existing methods are unable to effectively capture the clean sweeping sound of wind turbine blades. This ambient noise is always generated simultaneously with the blade's aerodynamic audio signal.

[0004] In summary, there is an urgent need for an audio noise reduction method, device and medium applied to industrial audio processing to solve the problems existing in the prior art. Summary of the Invention

[0005] The present invention aims to provide an audio noise reduction method, device, and medium for industrial audio processing, which are used to effectively collect the clean sweeping sound of wind turbine blades, that is, to eliminate the ambient noise in the blade aerodynamic audio signal. The specific technical solution is as follows:

[0006] An audio noise reduction method applied to industrial audio processing, the steps are as follows:

[0007] Step S1: performing super-subtraction processing on the noisy frequency signal to extract the pre-processed signal;

[0008] Step S2: performing Wiener filtering on the preprocessed signal obtained in step S1 to obtain an audio signal with ambient noise removed;

[0009] Step S3: performing the super-subtraction process in step S1 on the audio signal obtained in step S2 to extract the noise-removed audio signal;

[0010] In step S1, the super-subtraction method is specifically as follows:

[0011] The Berouti spectral subtraction method is used to process the noisy frequency signal to obtain frequency domain audio data, and the maximum noise residual value is calculated. The maximum noise residual value is then used to limit the frequency domain audio data to obtain preprocessed audio data, and finally the preprocessed audio data is subjected to inverse Fourier transform to obtain the preprocessed signal.

[0012] Preferably, in step S1, the maximum noise residual value is calculated by using a parameter function to calculate the noise data E of each frame after framing. t (ω) is subtracted from the estimated noise data V(ω), and then the E t The difference between (ω) and V(ω) is parameter-accumulated to obtain the maximum noise residual value, which is expressed as follows:

[0013]

[0014] Where max(ω) represents the maximum noise residual value; T noise Represents the frame number of the noise data.

[0015] Preferably, in step S1, the frequency domain audio data is limited to using the maximum noise residual value to replace the too small part of the frequency domain audio data with the minimum value of the adjacent frames, and the expression is as follows:

[0016]

[0017] Where |X(ω)| represents the preprocessed audio data, Indicates selecting the minimum value among adjacent frames.

[0018] Preferably, in step S2, the Wiener filtering process is to design a parametric Wiener filter, and use the parametric Wiener filter to filter the noisy frequency signal to obtain a filtered signal, so that the mean square error value between the filtered signal and the preprocessed signal is minimized.

[0019] Preferably, in step S2, the expression of the parametric Wiener filter is as follows:

[0020]

[0021] Among them, H(ω k ) represents the filter data at frequency point k, P xx (ω k ) represents the mean square error of the preprocessed signal at frequency k, P vv (ωk ) represents the mean square error of the noise signal at frequency k.

[0022] In addition, the present invention also provides a computer device, comprising:

[0023] Memory: Memory stores computer programs;

[0024] Processor: When the processor executes the computer program, the audio noise reduction method as described above is implemented.

[0025] In addition, the present invention also proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the audio noise reduction method as described above is implemented.

[0026] The application of the technical solution of the present invention has the following beneficial effects:

[0027] (1) The present invention improves the Berouti spectral subtraction method, calculates the maximum noise residual value, and uses the maximum noise residual value to limit the frequency domain audio data in the Berouti spectral subtraction method, thereby solving the problem of spectral line discontinuity caused by the Berouti spectral subtraction process; the method of limiting the frequency domain audio data in the Berouti spectral subtraction method in the present invention is to replace the excessively small part of the frequency domain audio data with the minimum value of the adjacent frames, thereby better solving the problem of spectral line discontinuity caused by the Berouti spectral subtraction process, and can effectively denoise industrial audio.

[0028] (2) The present invention uses a parametric Wiener filter to control the output of the Wiener filter by adjusting the first control parameter and the second control parameter to cope with frequency bands with different noise levels and suppress environmental noise in the audio signal.

[0029] (3) The present invention uses super-subtraction to process the audio signal again after Wiener filtering, which solves the problem of adhesion caused by the part of the environmental noise that is not filtered out. The present invention uses a combination of super-subtraction and Wiener filtering to process complex audio signals in the industrial field, thereby improving the effect of eliminating environmental noise.

[0030] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0032] Figure 1 is a flowchart of the steps of the audio noise reduction method in Example 1 of the present invention;

[0033] Figure 2 is a spectrum diagram of a noise-containing frequency signal of a wind turbine blade in Example 1 of the present invention;

[0034] Figure 3 Schematic diagram of the inhibitory effect of the first control parameter on the output signal in Example 1 of the present invention;

[0035] Figure 4 Schematic diagram of the inhibitory effect of the second control parameter on the output signal in Example 1 of the present invention;

[0036] Figure 5 This is a spectrum diagram of the noise-removed frequency signal of the wind turbine blade in Example 1 of the present invention.

[0037] Figure 6 is a spectrum diagram of the preprocessed signal of the wind turbine blade in Comparative Example 1 of the present invention;

[0038] Figure 7 1 is a spectrum diagram of the audio signal of the wind turbine blade in Comparative Example 2 of the present invention. DETAILED DESCRIPTION

[0039] In order to solve the problem that existing methods cannot effectively collect the clean sweeping sound of wind turbine blades and eliminate the noise signals simultaneously generated in the aerodynamic audio signals of the blades, the present invention proposes an audio denoising method for industrial audio processing.

[0040] Berouti spectral subtraction utilizes over-subtraction technology and sets a lower limit for negative values ​​after spectral subtraction to replace the 0 value in traditional spectral subtraction, thereby processing the problem of musical noise in audio that cannot be processed by traditional general subtraction.

[0041] The basic principle of traditional spectral subtraction is that, assuming the original clean audio signal and additive noise are independent of each other and the additive noise signal is stationary, the power spectrum of the noise signal can be subtracted from the power spectrum of the noisy speech signal to obtain the power spectrum of the noise-free speech signal. The spectral subtraction process can be expressed mathematically as follows:

[0042] s(n)=x(n)+v(n)

[0043] Among them, s(n) represents the noisy audio signal, x(n) represents the original clean signal, and v(n) represents the noise signal.

[0044] Furthermore, by performing Fourier transform on the audio signal, the original time domain data is converted into frequency domain data.

[0045] S(ω)=X(ω)+V(ω);

[0046] Among them, S(ω) represents the frequency domain audio signal data, X(ω) represents the frequency domain clean signal data, and V(ω) represents the frequency domain noise signal data.

[0047] Furthermore, the amplitude spectrum after noise addition is as follows:

[0048] |S(ω)| 2 =|X(ω)| 2 +|V(ω)| 2 +X * (ω)V(ω)+V * (ω)X(ω);

[0049] Here, * represents conjugation.

[0050] Because the noise signal is not correlated with the clean audio signal, and the clean signal data is predicted (i.e., by enhancing the speech signal) It should be infinitely close to the original clean signal data |X(ω)|, so we can get:

[0051]

[0052] Right now:

[0053]

[0054] Furthermore, the phase spectrum of the noisy signal S(ω) is used and prediction signals Multiply the amplitude spectrum by , and get the frequency domain audio data

[0055]

[0056] Furthermore, the inverse Fourier transform is used to transform the data in the frequency domain into the time domain, and the speech signal enhanced by spectral subtraction is obtained:

[0057]

[0058] Among them, the spectral subtraction method stipulates that The value range should be:

[0059]

[0060] That is Time The rest of the values ​​should be 0.

[0061] However, when spectral subtraction is actually applied to speech enhancement processing, there is a high probability that spectral subtraction will introduce noise. Therefore, the Berouti spectral subtraction method substitutes the first control parameter α and the second control parameter β into the traditional spectral subtraction method for over-subtraction, and sets a lower limit for the negative value after spectral subtraction to replace the 0 value in the traditional spectral subtraction to eliminate noise.

[0062] Then perform inverse Fourier transform to convert the data in the frequency domain to the time domain to obtain the speech signal enhanced by Berouti spectrum subtraction.

[0063]

[0064] in, Express Perform an inverse Fourier transform.

[0065] Using the improved spectral subtraction method developed by M. Berouti et al., by adjusting the first and second control parameters, it is possible to effectively remove the noise that can occur with traditional spectral subtraction for speech enhancement. However, experiments have shown that while this technique works well for human voice tasks, when processing industrial audio data, it can significantly cause spectral discontinuities while removing noise. Analysis has revealed that this problem is often caused by the presence of residual noise.

[0066] The present invention proposes an audio noise reduction method for industrial audio processing. By improving the Berouti spectral subtraction method, the maximum noise residual value is used to limit the frequency domain audio data in the Berouti spectral subtraction method before performing the inverse Fourier transform. This method solves the spectral line discontinuity problem caused by the Berouti spectral subtraction process. The following describes an embodiment of the present invention in detail with reference to the accompanying drawings. However, the present invention may be implemented in a variety of different ways as defined and covered by the claims.

[0067] Example 1:

[0068] See also Figure 1 This embodiment discloses an audio denoising method for industrial audio processing, specifically for the original collected wind turbine blade noise-containing frequency signal (such as Figure 2 As shown in the figure, the collected wind turbine blade noise frequency signal does not show the frequency band of the engine blade audio signal, which shows that the noise pollution is relatively serious) and the steps are as follows:

[0069] Step S1: performing super-subtraction processing on the noisy frequency signal to extract the pre-processed signal;

[0070] Step S2: performing Wiener filtering on the pre-processed signal in step S1 to obtain an audio signal with environmental noise removed;

[0071] Step S3: performing the super-subtraction process in step S1 on the audio signal in step S2 to extract the noise-removed audio signal;

[0072] In step S1, the super-subtraction method is specifically as follows:

[0073] The Berouti spectral subtraction method is used to process the noisy frequency signal to obtain frequency domain audio data, and the maximum noise residual value is calculated. The maximum noise residual value is then used to limit the frequency domain audio data to obtain preprocessed audio data, and finally the preprocessed audio data is subjected to inverse Fourier transform to obtain the preprocessed signal.

[0074] Preferably, in step S1, the maximum noise residual value is calculated by using a parameter function to calculate the noise data E of each frame after framing. t (ω) is subtracted from the estimated noise data V(ω), and then the E t The difference between (ω) and V(ω) is parameter-accumulated to obtain the maximum noise residual value, which is expressed as follows:

[0075]

[0076] Where max(ω) represents the maximum noise residual value; T noise Represents the frame number of the noise data.

[0077] Preferably, in step S1, the frequency domain audio data is limited to using the maximum noise residual value to replace the too small part of the frequency domain audio data with the minimum value of the adjacent frames, and the expression is as follows:

[0078]

[0079] Where |X(ω)| represents the preprocessed audio data, Indicates selecting the minimum value among adjacent frames.

[0080] Preferably, in step S2, the Wiener filtering process is to design a parametric Wiener filter, and use the parametric Wiener filter to filter the noisy frequency signal to obtain a filtered signal, so that the mean square error value between the filtered signal and the preprocessed signal is minimized.

[0081] Preferably, in step S2, the expression of the parametric Wiener filter is as follows:

[0082]

[0083] Among them, H(ω k) represents the filter data at frequency point k, P xx (ω k ) represents the mean square error of the preprocessed signal at frequency k, P vv (ω k ) represents the mean square error of the noise signal at frequency k.

[0084] Furthermore, Wiener filtering can be considered as a linear minimum mean square error (LMMSE) estimator. Linearity refers to the linearity of the estimation form, and the minimum mean square error is the optimization criterion for constructing the filter later. When the random signal s(n), that is, the noisy raw data, meets the following conditions:

[0085] s(n)=x(n)+v(n);

[0086] Among them, x(n) represents the original clean signal and v(n) represents the noise signal.

[0087] Input the signal s(n) into the linear filter h(n), and output for:

[0088]

[0089] It represents the filtered signal after passing through the linear filter h(n). ∑ means the addition sign. ∑h(n)*s(n) means h(1)*s(2)+h(2)*s(3)+...+h(n)*s(n).

[0090] That is, by convolving the input noisy signal s(n) with the filter h(n), the filtered signal is obtained.

[0091] The signal data in the time domain is further converted to the frequency domain through Fourier transform. That is, the frequency domain audio data is obtained.

[0092]

[0093] Here, H(ω) represents filter data, and S(ω) represents random signal data.

[0094] And because each frequency band is independent of each other, we can further analyze each frequency point k to obtain the frequency domain audio data at the kth frequency point.

[0095]

[0096] Among them, S(ω k ) represents the random signal data at the kth frequency point.

[0097] Furthermore, the preprocessed audio data can be obtained by calculating the minimum mean square error value.

[0098] Furthermore, since the frequency domain noisy signal data S(ω k ) is the original clean signal data X(ω k ) and the noise signal data V(ω k ), that is:

[0099] S(ω k )=X(ω k )+V(ω k );

[0100] And the original clean signal X(ω k ) and the noise signal V(ω k ) are uncorrelated, the filter expression can be derived:

[0101]

[0102] Furthermore, in order to meet the demand for output control of the Wiener filter, the first control parameter α and the second control parameter β are added to the filter to obtain an improved parametric Wiener filter.

[0103] On this basis, see Figure 3 as well as Figure 4 , it can be seen that the first and second control parameters in this embodiment have a suppressive effect on different output signals. Due to the suppressive noise reduction effect of the Wiener filter, while suppressing the ambient noise, it also suppresses the output of some useful signals, resulting in the suppression of the clean signal. At this time, the audio signal and the unfiltered portion of the ambient noise are adhered to each other. To address this problem, this embodiment uses the above-mentioned super-subtraction method to reprocess the filtered signal, eliminating the noise signal that causes adhesion, and thus obtaining a clearer denoised audio signal.

[0104] like Figure 5 As shown, comparing the spectrum of the noisy data before filtering with the spectrum after overall filtering, it can be seen that the periodic audio signal generated by the aerodynamics of the wind turbine blades is clearly visible and has stronger and more obvious characteristics compared to before processing using the audio denoising method in Example 1. There is no obvious noise in the denoised audio signal.

[0105] In addition, this embodiment also provides a computer device, including:

[0106] Memory: Memory stores computer programs;

[0107] Processor: When the processor executes the computer program, the audio noise reduction method as described above is implemented.

[0108] In addition, this embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the audio noise reduction method as described above is implemented.

[0109] Comparative Example 1:

[0110] This comparative example is processed by only using step S1 of Example 1. Figure 6 As shown in FIG, the pre-processed signal of the wind turbine blade obtained by processing the noise-containing frequency signal of the wind turbine blade using the super-subtraction method has shown a certain frequency band, which shows that the super-subtraction method has the ability to eliminate noise signals.

[0111] Comparative Example 2:

[0112] This comparative example only uses the steps S1 and S2 used in Example 1 for processing. The effect after processing is as follows: Figure 7 As shown in the figure, the spectrum data after filtering the wind turbine blade preprocessing signal using a parametric Wiener filter with α being 1 and β being 8, the noise of the signal is suppressed again, but the output of some useful signals is also suppressed at the same time. At this time, the clean signal in the audio signal is suppressed, so it is necessary to use super subtraction to extract the audio signal again.

[0113] It can be seen from Example 1 and Comparative Examples 1-2 that the present invention uses steps S1, S2, and S3 to perform processing according to specific steps, and there is no obvious noise in the processed audio signal, and the effect is significant.

[0114] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An audio noise reduction method applied to industrial audio processing, characterized in that: Here are the steps: Step S1: performing super-subtraction processing on the noisy frequency signal to extract the pre-processed signal; Step S2: performing Wiener filtering on the preprocessed signal obtained in step S1 to obtain an audio signal with ambient noise removed; Step S3: performing the super-subtraction process in step S1 on the audio signal obtained in step S2 to extract the noise-removed audio signal; In step S1, the super-subtraction method is specifically as follows: The Berouti spectral subtraction method is used to process the noisy frequency signal to obtain frequency domain audio data, the maximum noise residual value is calculated, and then the maximum noise residual value is used to limit the frequency domain audio data to obtain preprocessed audio data, and finally the preprocessed audio data is subjected to inverse Fourier transform to obtain the preprocessed signal.

2. The audio noise reduction method according to claim 1, wherein: In step S1, the maximum noise residual value is calculated by using a parameter function to convert the noise data E of each frame after framing into t (ω) is subtracted from the estimated noise data V(ω), and then the E t The difference between (ω) and V(ω) is parameter-accumulated to obtain the maximum noise residual value, which is expressed as follows: Where max(ω) represents the maximum noise residual value; T noise Represents the frame number of the noise data.

3. The audio noise reduction method according to claim 2, wherein: In step S1, the frequency domain audio data is limited to using the maximum noise residual value to replace the smaller part of the frequency domain audio data with the minimum value of the adjacent frames. The expression is as follows: Where |X(ω)| represents the preprocessed audio data, Indicates selecting the minimum value among adjacent frames.

4. The audio noise reduction method according to claim 3, wherein: In step S2, the Wiener filtering process is to design a parametric Wiener filter, and use the parametric Wiener filter to filter the noisy frequency signal to obtain a filtered signal, so that the mean square error value between the filtered signal and the preprocessed signal is minimized.

5. The audio noise reduction method according to claim 4, characterized in that: In step S2, the expression of the parametric Wiener filter is as follows: Among them, H(ω k ) represents the filter data at frequency point k, P xx (ω k ) represents the mean square error of the preprocessed signal at frequency k, P vv (ω k ) represents the mean square error of the noise signal at frequency k.

6. A computer device, characterized in that: include: Memory: Memory stores computer programs; Processor: When executing the computer program, the processor implements the audio noise reduction method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the audio noise reduction method according to any one of claims 1 to 5 is implemented.

Images